| 1 | """Director tools for structured video-workspace orchestration. |
| 2 | |
| 3 | These tools keep a work-specific state machine on disk so the model can rely |
| 4 | on tool-managed state instead of reconstructing progress from conversation |
| 5 | history alone. |
| 6 | """ |
| 7 | |
| 8 | from __future__ import annotations |
| 9 | |
| 10 | import asyncio |
| 11 | import hashlib |
| 12 | import json |
| 13 | import re |
| 14 | import time |
| 15 | from contextvars import ContextVar |
| 16 | from datetime import datetime, timezone |
| 17 | from pathlib import Path |
| 18 | from typing import Any |
| 19 | from urllib import error as urllib_error |
| 20 | from urllib import request as urllib_request |
| 21 | from urllib.parse import unquote, urlparse |
| 22 | |
| 23 | from loguru import logger |
| 24 | |
| 25 | from nanobot.agent.tools import ( |
| 26 | ArraySchema, |
| 27 | BooleanSchema, |
| 28 | IntegerSchema, |
| 29 | ObjectSchema, |
| 30 | StringSchema, |
| 31 | Tool, |
| 32 | tool_parameters, |
| 33 | tool_parameters_schema, |
| 34 | ) |
| 35 | from nanobot.integrations.echo_admission import ( |
| 36 | UNAVAILABLE_MESSAGE, |
| 37 | EchoAdmissionController, |
| 38 | EchoGeneratorBusyError, |
| 39 | EchoGeneratorUnavailableError, |
| 40 | is_connection_refused, |
| 41 | ) |
| 42 | from nanobot.prompts import prompts |
| 43 | from nanobot.prompts.manager import PEManager |
| 44 | from nanobot.session.auto_generate import ( |
| 45 | effective_auto_generate_shot_count, |
| 46 | get_auto_generate, |
| 47 | locked_shot_count_from_goal, |
| 48 | ) |
| 49 | from nanobot.session.reference_image import ( |
| 50 | clear_reference_image_needs_story_rewrite, |
| 51 | is_reference_image_locked, |
| 52 | lock_reference_image, |
| 53 | normalize_reference_image, |
| 54 | reference_image_needs_story_rewrite, |
| 55 | reference_image_present, |
| 56 | ) |
| 57 | from nanobot.utils.helpers import write_json_atomic |
| 58 | |
| 59 | DIRECTOR_CONTEXT_TOOL_NAMES = frozenset( |
| 60 | { |
| 61 | "start_director", |
| 62 | "set_director_goal", |
| 63 | "get_workplace_status", |
| 64 | "get_story", |
| 65 | "write_story", |
| 66 | "get_shot", |
| 67 | "create_shot_prompt", |
| 68 | "review_shot", |
| 69 | "set_shot_references", |
| 70 | "set_shot_memory_recommendations", |
| 71 | "generate_echo_shot", |
| 72 | "merge_shot", |
| 73 | } |
| 74 | ) |
| 75 | |
| 76 | DIRECTOR_MUTATING_TOOL_NAMES = frozenset( |
| 77 | { |
| 78 | "start_director", |
| 79 | "set_director_goal", |
| 80 | "write_story", |
| 81 | "create_shot_prompt", |
| 82 | "review_shot", |
| 83 | "set_shot_references", |
| 84 | "set_shot_memory_recommendations", |
| 85 | "generate_echo_shot", |
| 86 | "merge_shot", |
| 87 | } |
| 88 | ) |
| 89 | |
| 90 | # Shown in stepwise chat after the user locks shot_count and before they click |
| 91 | # Workplace 01 「下一步」. Never used for input-box one-click (auto_generate). |
| 92 | SHOT_COUNT_NEXT_STEP_HINT = ( |
| 93 | "点击「下一步」即可预览分镜脚本。满意脚本后接下来可以生成分镜镜头，" |
| 94 | "确认无误并接受所有分镜后，就能合成最终成片了。有任何问题可以随时找我~" |
| 95 | ) |
| 96 | SHOT_COUNT_NEXT_STEP_HINT_PENDING_KEY = "shot_count_next_step_hint_pending" |
| 97 | _STAGES_PAST_SHOT_COUNT_HINT = frozenset( |
| 98 | { |
| 99 | "shot_planning", |
| 100 | "shot_generating", |
| 101 | "shot_reviewing", |
| 102 | "shot_revising", |
| 103 | "merging", |
| 104 | "done", |
| 105 | "cancelled", |
| 106 | "awaiting_memory_review", |
| 107 | "failed", |
| 108 | } |
| 109 | ) |
| 110 | |
| 111 | |
| 112 | def consume_shot_count_next_step_hint( |
| 113 | workspace: Path, |
| 114 | session_key: str | None, |
| 115 | *, |
| 116 | auto_generate: bool = False, |
| 117 | emit: bool = True, |
| 118 | ) -> str | None: |
| 119 | """Return the stepwise 「下一步」 hint once after shot_count is first locked.""" |
| 120 | if not session_key: |
| 121 | return None |
| 122 | tool = GetWorkplaceStatusTool(workspace=workspace) |
| 123 | tool.set_context("websocket", "direct", effective_key=session_key) |
| 124 | return tool._consume_shot_count_next_step_hint( |
| 125 | auto_generate=auto_generate, |
| 126 | emit=emit, |
| 127 | ) |
| 128 | |
| 129 | |
| 130 | def stepwise_shot_count_next_step_hint_eligible( |
| 131 | workspace: Path, |
| 132 | session_key: str | None, |
| 133 | *, |
| 134 | auto_generate: bool = False, |
| 135 | ) -> bool: |
| 136 | """True when stepwise work is still on 01 with a locked shot_count.""" |
| 137 | if not session_key or auto_generate: |
| 138 | return False |
| 139 | tool = GetWorkplaceStatusTool(workspace=workspace) |
| 140 | tool.set_context("websocket", "direct", effective_key=session_key) |
| 141 | work_id = tool._active_work_id() |
| 142 | if not work_id: |
| 143 | return False |
| 144 | state = tool._load_state(work_id) |
| 145 | if bool(state.get("auto_generate")): |
| 146 | return False |
| 147 | if str(state.get("stage") or "") in _STAGES_PAST_SHOT_COUNT_HINT: |
| 148 | return False |
| 149 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 150 | try: |
| 151 | return int(goal.get("shot_count") or 0) > 0 |
| 152 | except (TypeError, ValueError): |
| 153 | return False |
| 154 | |
| 155 | _REMOTE_PROTOCOL_VERSION = "director-http-v1" |
| 156 | _R2V_SUBMIT_ATTEMPTS = 3 |
| 157 | _R2V_TRANSIENT_HTTP_CODES = frozenset({429, 502, 503, 504}) |
| 158 | |
| 159 | _REMOTE_ENDPOINT_PATHS = { |
| 160 | # Stable Echo Server routes used by the release workflow. |
| 161 | "merge_shot": "/merge", |
| 162 | # R2V unified generation (T2V / I2V / R2V) |
| 163 | "r2v_generate": "/r2v", |
| 164 | } |
| 165 | |
| 166 | _REMOTE_CALLBACK_PATHS = { |
| 167 | "generate_echo_shot": "/api/director/echo-generate-shot/callback", |
| 168 | "merge_shot": "/api/director/merge-shot/callback", |
| 169 | } |
| 170 | |
| 171 | # Workplace workflow transitions are button-driven. Chat turns may only advance |
| 172 | # stages when the agent loop sets one of these injected workplace events. |
| 173 | WORKFLOW_GATE_BYPASS = "workplace_test_bypass" |
| 174 | _WORKFLOW_INJECTED_EVENT: ContextVar[str | None] = ContextVar( |
| 175 | "director_workflow_injected_event", |
| 176 | default=None, |
| 177 | ) |
| 178 | _WORKFLOW_CONTEXT_UNSET = object() |
| 179 | _WORKFLOW_GATE_OPERATIONS: dict[str, frozenset[str]] = { |
| 180 | "write_story_confirmed": frozenset( |
| 181 | { |
| 182 | "workplace_workflow_confirm_story", |
| 183 | "workplace_workflow_start_generation", |
| 184 | "workplace_beats_edit", |
| 185 | WORKFLOW_GATE_BYPASS, |
| 186 | } |
| 187 | ), |
| 188 | "create_shot_prompt": frozenset( |
| 189 | { |
| 190 | "workplace_workflow_start_generation", |
| 191 | "workplace_beats_edit", |
| 192 | "workplace_shot_revision", |
| 193 | WORKFLOW_GATE_BYPASS, |
| 194 | } |
| 195 | ), |
| 196 | "set_shot_references": frozenset( |
| 197 | { |
| 198 | "workplace_workflow_start_generation", |
| 199 | WORKFLOW_GATE_BYPASS, |
| 200 | } |
| 201 | ), |
| 202 | "set_shot_memory_recommendations": frozenset( |
| 203 | { |
| 204 | "workplace_memory_recommendation", |
| 205 | WORKFLOW_GATE_BYPASS, |
| 206 | } |
| 207 | ), |
| 208 | "generate_echo_shot": frozenset( |
| 209 | { |
| 210 | "workplace_workflow_start_generation", |
| 211 | "workplace_shot_revision", |
| 212 | WORKFLOW_GATE_BYPASS, |
| 213 | } |
| 214 | ), |
| 215 | "merge_shot": frozenset( |
| 216 | { |
| 217 | "workplace_workflow_start_merge", |
| 218 | WORKFLOW_GATE_BYPASS, |
| 219 | } |
| 220 | ), |
| 221 | "review_shot": frozenset({WORKFLOW_GATE_BYPASS}), |
| 222 | } |
| 223 | |
| 224 | |
| 225 | def _workflow_gate_error(operation: str) -> str: |
| 226 | return prompts.text("director.workflow_gate_error", operation=operation) |
| 227 | |
| 228 | |
| 229 | def _allow_workflow_operation(operation: str) -> bool: |
| 230 | event = _WORKFLOW_INJECTED_EVENT.get() |
| 231 | if event == WORKFLOW_GATE_BYPASS: |
| 232 | return True |
| 233 | if not event: |
| 234 | return False |
| 235 | return event in _WORKFLOW_GATE_OPERATIONS.get(operation, frozenset()) |
| 236 | |
| 237 | |
| 238 | def _now_iso() -> str: |
| 239 | return datetime.now(timezone.utc).isoformat(timespec="seconds").replace("+00:00", "Z") |
| 240 | |
| 241 | |
| 242 | def _slugify(value: str, *, fallback: str) -> str: |
| 243 | slug = re.sub(r"[^a-z0-9]+", "-", value.strip().lower()) |
| 244 | slug = re.sub(r"-{2,}", "-", slug).strip("-") |
| 245 | return slug or fallback |
| 246 | |
| 247 | |
| 248 | def _json_dump(data: Any) -> str: |
| 249 | return json.dumps(data, ensure_ascii=False, indent=2, sort_keys=True) |
| 250 | |
| 251 | |
| 252 | def _story_profile_validation_error(story_profile: Any) -> str | None: |
| 253 | if not isinstance(story_profile, dict): |
| 254 | return "Error: story_profile must be a JSON object with summary and beats." |
| 255 | summary = story_profile.get("summary") |
| 256 | if not isinstance(summary, str) or not summary.strip(): |
| 257 | return "Error: story_profile.summary must be a non-empty string." |
| 258 | beats = story_profile.get("beats") |
| 259 | if not isinstance(beats, list) or len(beats) < 1: |
| 260 | return "Error: story_profile.beats must contain at least one beat." |
| 261 | for index, beat in enumerate(beats): |
| 262 | if not isinstance(beat, dict): |
| 263 | return ( |
| 264 | f"Error: story_profile.beats[{index}] must be an object with shot_id and summary." |
| 265 | ) |
| 266 | beat_summary = beat.get("summary") |
| 267 | if not isinstance(beat_summary, str) or not beat_summary.strip(): |
| 268 | return f"Error: story_profile.beats[{index}].summary must be a non-empty string." |
| 269 | return None |
| 270 | |
| 271 | |
| 272 | def _story_profile_language_validation_error( |
| 273 | story_profile: dict[str, Any], |
| 274 | ) -> str | None: |
| 275 | """Keep all natural-language story-profile prose in the selected language.""" |
| 276 | from nanobot.session.generation_settings import normalize_language |
| 277 | |
| 278 | normalized = normalize_language(story_profile.get("language")) |
| 279 | if normalized is None: |
| 280 | normalized = normalize_language(story_profile.get("caption_language")) |
| 281 | if normalized is None: |
| 282 | normalized = normalize_language(story_profile.get("dialogue_language")) |
| 283 | if normalized is None: |
| 284 | return None |
| 285 | |
| 286 | summaries: list[tuple[str, str]] = [] |
| 287 | prose: list[tuple[str, str]] = [] |
| 288 | |
| 289 | def collect_text(value: Any, path: str) -> None: |
| 290 | if isinstance(value, str): |
| 291 | if value.strip(): |
| 292 | prose.append((path, value)) |
| 293 | return |
| 294 | if isinstance(value, list): |
| 295 | for index, item in enumerate(value): |
| 296 | collect_text(item, f"{path}[{index}]") |
| 297 | return |
| 298 | if isinstance(value, dict): |
| 299 | for key, item in value.items(): |
| 300 | collect_text(item, f"{path}.{key}") |
| 301 | |
| 302 | summary = story_profile.get("summary") |
| 303 | if isinstance(summary, str): |
| 304 | summaries.append(("story_profile.summary", summary)) |
| 305 | beats = story_profile.get("beats") |
| 306 | if isinstance(beats, list): |
| 307 | for index, beat in enumerate(beats): |
| 308 | if not isinstance(beat, dict): |
| 309 | continue |
| 310 | if isinstance(beat.get("summary"), str): |
| 311 | summaries.append((f"story_profile.beats[{index}].summary", beat["summary"])) |
| 312 | collect_text( |
| 313 | beat.get("dialogue_intent"), |
| 314 | f"story_profile.beats[{index}].dialogue_intent", |
| 315 | ) |
| 316 | |
| 317 | for field in ("anchors", "scene_anchors", "shot_to_content"): |
| 318 | collect_text(story_profile.get(field), f"story_profile.{field}") |
| 319 | |
| 320 | def _has_chinese(value: str) -> bool: |
| 321 | return bool(re.search(r"[\u3400-\u4dbf\u4e00-\u9fff]", value)) |
| 322 | |
| 323 | if normalized == "zh": |
| 324 | invalid_summaries = [name for name, value in summaries if not _has_chinese(value)] |
| 325 | invalid_prose = [name for name, value in prose if not _has_chinese(value)] |
| 326 | if invalid_prose: |
| 327 | invalid = invalid_summaries + invalid_prose |
| 328 | return ( |
| 329 | "Error: Chinese story-profile prose is required for this work. " |
| 330 | "Rewrite all natural-language story_profile fields in Simplified Chinese. " |
| 331 | f"Invalid fields: {', '.join(invalid)}." |
| 332 | ) |
| 333 | if invalid_summaries: |
| 334 | return ( |
| 335 | "Error: Chinese storyboard summaries are required for this work. " |
| 336 | "Rewrite story_profile.summary and every beats[].summary in Simplified Chinese. " |
| 337 | f"Invalid fields: {', '.join(invalid_summaries)}." |
| 338 | ) |
| 339 | elif normalized == "en": |
| 340 | invalid_summaries = [name for name, value in summaries if _has_chinese(value)] |
| 341 | invalid_prose = [name for name, value in prose if _has_chinese(value)] |
| 342 | if invalid_prose: |
| 343 | invalid = invalid_summaries + invalid_prose |
| 344 | return ( |
| 345 | "Error: English story-profile prose is required for this work. " |
| 346 | "Rewrite all natural-language story_profile fields in English. " |
| 347 | f"Invalid fields: {', '.join(invalid)}." |
| 348 | ) |
| 349 | if invalid_summaries: |
| 350 | return ( |
| 351 | "Error: English storyboard summaries are required for this work. " |
| 352 | "Rewrite story_profile.summary and every beats[].summary in English. " |
| 353 | f"Invalid fields: {', '.join(invalid_summaries)}." |
| 354 | ) |
| 355 | return None |
| 356 | |
| 357 | |
| 358 | def _story_md_language_validation_error( |
| 359 | story_md: str, |
| 360 | story_profile: dict[str, Any], |
| 361 | ) -> str | None: |
| 362 | """Keep the displayed screenplay aligned with the selected story language.""" |
| 363 | from nanobot.session.generation_settings import normalize_language |
| 364 | |
| 365 | normalized = normalize_language(story_profile.get("language")) |
| 366 | if normalized is None: |
| 367 | normalized = normalize_language(story_profile.get("caption_language")) |
| 368 | if normalized is None: |
| 369 | normalized = normalize_language(story_profile.get("dialogue_language")) |
| 370 | if normalized is None or not story_md.strip(): |
| 371 | return None |
| 372 | |
| 373 | has_chinese = bool(re.search(r"[\u3400-\u4dbf\u4e00-\u9fff]", story_md)) |
| 374 | if normalized == "zh" and not has_chinese: |
| 375 | return ( |
| 376 | "Error: Chinese screenplay prose is required for this work. " |
| 377 | "Rewrite story_md in Simplified Chinese." |
| 378 | ) |
| 379 | if normalized == "en" and has_chinese: |
| 380 | return ( |
| 381 | "Error: English screenplay prose is required for this work. " |
| 382 | "Rewrite story_md in English." |
| 383 | ) |
| 384 | return None |
| 385 | |
| 386 | |
| 387 | def _normalize_story_profile(profile: dict[str, Any]) -> None: |
| 388 | beats = profile.get("beats") |
| 389 | if not isinstance(beats, list): |
| 390 | return |
| 391 | normalized: list[dict[str, Any]] = [] |
| 392 | for index, beat in enumerate(beats): |
| 393 | if not isinstance(beat, dict): |
| 394 | continue |
| 395 | summary = str(beat.get("summary") or "").strip() |
| 396 | if not summary: |
| 397 | continue |
| 398 | normalized.append({"shot_id": index + 1, "summary": summary}) |
| 399 | profile["beats"] = normalized |
| 400 | shot_to_content: dict[str, str] = {} |
| 401 | content_to_shots: dict[str, list[str]] = {} |
| 402 | for beat in normalized: |
| 403 | shot_id = int(beat["shot_id"]) |
| 404 | shot_key = f"shot_{shot_id:03d}" |
| 405 | shot_to_content[shot_key] = str(beat["summary"]) |
| 406 | content_to_shots[f"beat_{shot_id:03d}"] = [shot_key] |
| 407 | profile["shot_to_content"] = shot_to_content |
| 408 | profile["content_to_shots"] = content_to_shots |
| 409 | |
| 410 | |
| 411 | def _apply_story_profile_language(profile: dict[str, Any], language: str | None) -> None: |
| 412 | """Stamp UI language onto story_profile and keep caption/dialogue locks in sync.""" |
| 413 | from nanobot.session.generation_settings import ( |
| 414 | language_to_caption_language, |
| 415 | language_to_dialogue_language, |
| 416 | normalize_language, |
| 417 | ) |
| 418 | |
| 419 | normalized = normalize_language(language) |
| 420 | if normalized is None: |
| 421 | return |
| 422 | profile["language"] = normalized |
| 423 | dialogue = language_to_dialogue_language(normalized) |
| 424 | if dialogue: |
| 425 | profile["dialogue_language"] = dialogue |
| 426 | caption = language_to_caption_language(normalized) |
| 427 | if caption and not str(profile.get("caption_language") or "").strip(): |
| 428 | profile["caption_language"] = caption |
| 429 | |
| 430 | |
| 431 | def _ensure_story_profile_caption_language(profile: dict[str, Any]) -> None: |
| 432 | """Derive the full-caption language for legacy profiles that only locked dialogue.""" |
| 433 | from nanobot.session.generation_settings import ( |
| 434 | language_to_caption_language, |
| 435 | normalize_language, |
| 436 | ) |
| 437 | |
| 438 | if str(profile.get("caption_language") or "").strip(): |
| 439 | return |
| 440 | normalized = normalize_language(profile.get("language")) |
| 441 | if normalized is None: |
| 442 | normalized = normalize_language(profile.get("dialogue_language")) |
| 443 | caption = language_to_caption_language(normalized) |
| 444 | if caption: |
| 445 | profile["caption_language"] = caption |
| 446 | |
| 447 | |
| 448 | def _preserve_story_profile_language( |
| 449 | profile: dict[str, Any], |
| 450 | previous: dict[str, Any] | None = None, |
| 451 | ) -> None: |
| 452 | """Keep language / dialogue_language when an overwrite omits them.""" |
| 453 | from nanobot.session.generation_settings import normalize_language |
| 454 | |
| 455 | if normalize_language(profile.get("language")) is not None: |
| 456 | _apply_story_profile_language(profile, profile.get("language")) |
| 457 | return |
| 458 | if isinstance(previous, dict): |
| 459 | prev_language = normalize_language(previous.get("language")) |
| 460 | if prev_language is not None: |
| 461 | _apply_story_profile_language(profile, prev_language) |
| 462 | return |
| 463 | prev_dialogue = previous.get("dialogue_language") |
| 464 | if isinstance(prev_dialogue, str) and prev_dialogue.strip() and "dialogue_language" not in profile: |
| 465 | profile["dialogue_language"] = prev_dialogue.strip() |
| 466 | _ensure_story_profile_caption_language(profile) |
| 467 | |
| 468 | |
| 469 | def _caption_language_validation_error( |
| 470 | caption: str, |
| 471 | story_profile: dict[str, Any], |
| 472 | ) -> str | None: |
| 473 | """Reject natural-language prose that violates the work's caption-language lock. |
| 474 | |
| 475 | The target language must account for at least 90% of the meaningful character |
| 476 | count (excluding technical tokens). A handful of proper-noun transliterations |
| 477 | in the source language are tolerated below the 10% threshold. |
| 478 | """ |
| 479 | from nanobot.session.generation_settings import normalize_language |
| 480 | |
| 481 | caption_language = story_profile.get("caption_language") |
| 482 | normalized = normalize_language(caption_language) |
| 483 | if normalized is None: |
| 484 | normalized = normalize_language(story_profile.get("language")) |
| 485 | if normalized is None: |
| 486 | normalized = normalize_language(story_profile.get("dialogue_language")) |
| 487 | if normalized is None: |
| 488 | return None |
| 489 | |
| 490 | # Strip technical tokens that are allowed in either language. |
| 491 | prose = re.sub( |
| 492 | r"(?<![A-Za-z0-9_])ID_[A-Z0-9]+(?![A-Za-z0-9_])", |
| 493 | "", |
| 494 | caption, |
| 495 | flags=re.IGNORECASE, |
| 496 | ) |
| 497 | prose = re.sub( |
| 498 | r"(?<![A-Za-z0-9_])shot\d+(?![A-Za-z0-9_])", |
| 499 | "", |
| 500 | prose, |
| 501 | flags=re.IGNORECASE, |
| 502 | ) |
| 503 | prose = re.sub( |
| 504 | r"(?<![A-Za-z])OCR(?![A-Za-z])", |
| 505 | "", |
| 506 | prose, |
| 507 | flags=re.IGNORECASE, |
| 508 | ) |
| 509 | |
| 510 | chinese_chars = re.findall(r"[\u3400-\u4dbf\u4e00-\u9fff]", prose) |
| 511 | english_words = re.findall(r"[A-Za-z]+(?:'[A-Za-z]+)?", prose) |
| 512 | |
| 513 | chinese_char_count = len(chinese_chars) |
| 514 | english_char_count = sum(len(w) for w in english_words) |
| 515 | |
| 516 | total_meaningful = chinese_char_count + english_char_count |
| 517 | if total_meaningful == 0: |
| 518 | return None |
| 519 | |
| 520 | if normalized == "en": |
| 521 | en_ratio = english_char_count / total_meaningful |
| 522 | if en_ratio >= 0.9: |
| 523 | return None |
| 524 | preview = "".join(chinese_chars[:16]) |
| 525 | return ( |
| 526 | "Error: English caption required for this work (currently " |
| 527 | f"{en_ratio:.0%} English). Rewrite the entire caption in English " |
| 528 | "before calling create_shot_prompt again; translate all Chinese " |
| 529 | f"descriptions, actions, dialogue, and declarations. " |
| 530 | f"Chinese found: {preview}." |
| 531 | ) |
| 532 | |
| 533 | if normalized == "zh": |
| 534 | zh_ratio = chinese_char_count / total_meaningful |
| 535 | if zh_ratio >= 0.9: |
| 536 | return None |
| 537 | preview = ", ".join(english_words[:8]) |
| 538 | return ( |
| 539 | "Error: Chinese caption required for this work (currently " |
| 540 | f"{zh_ratio:.0%} Chinese). Rewrite the entire caption in Chinese " |
| 541 | "before calling create_shot_prompt again. Keep only required " |
| 542 | "technical tokens such as ID_A, shot1:, and OCR; use ID_A说 for " |
| 543 | "speech; translate all descriptions, actions, camera, sound, music, " |
| 544 | f"and declarations. English found: {preview}." |
| 545 | ) |
| 546 | |
| 547 | return None |
| 548 | |
| 549 | |
| 550 | def _shot_key(shot_id: int) -> str: |
| 551 | return f"shot_{shot_id:03d}" |
| 552 | |
| 553 | |
| 554 | def _job_id(kind: str, work_id: str, suffix: str) -> str: |
| 555 | stamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ") |
| 556 | return f"{kind}-{work_id}-{suffix}-{stamp}" |
| 557 | |
| 558 | |
| 559 | def _shot_id_from_key(shot_key: str) -> int: |
| 560 | if not shot_key.startswith("shot_"): |
| 561 | raise ValueError(f"Invalid shot key: {shot_key}") |
| 562 | return int(shot_key.split("_", 1)[1]) |
| 563 | |
| 564 | |
| 565 | # Echo generate-shot timing: see echo_generate_shot.md (25fps, num_frames = 1 + 8k, clamp [25, 241]). |
| 566 | ECHO_SHOT_FPS = 25 |
| 567 | ECHO_MIN_NUM_FRAMES = 25 |
| 568 | ECHO_MAX_NUM_FRAMES = 241 |
| 569 | ECHO_DEFAULT_NUM_FRAMES = 241 |
| 570 | ECHO_DEFAULT_DURATION_SEC = 4.0 |
| 571 | |
| 572 | |
| 573 | def snap_echo_num_frames(raw_frames: int) -> int: |
| 574 | """Snap upward to the nearest valid 1+8k frame count and clamp to Echo bounds. |
| 575 | |
| 576 | Echo requires ``num_frames = 1 + 8k``. Snapping up keeps generated duration |
| 577 | from falling below the caller's requested length (except at the hard max). |
| 578 | """ |
| 579 | frames = int(raw_frames) |
| 580 | frames = max(ECHO_MIN_NUM_FRAMES, min(ECHO_MAX_NUM_FRAMES, frames)) |
| 581 | remainder = (frames - 1) % 8 |
| 582 | if remainder: |
| 583 | frames += 8 - remainder |
| 584 | if frames > ECHO_MAX_NUM_FRAMES: |
| 585 | # Largest valid 1+8k at or below the hard max (241 == 1+8*30). |
| 586 | frames = ECHO_MAX_NUM_FRAMES |
| 587 | remainder = (frames - 1) % 8 |
| 588 | if remainder: |
| 589 | frames -= remainder |
| 590 | return max(ECHO_MIN_NUM_FRAMES, min(ECHO_MAX_NUM_FRAMES, frames)) |
| 591 | |
| 592 | |
| 593 | def duration_sec_to_num_frames(duration_sec: float) -> int: |
| 594 | """Convert desired seconds to the frame count Echo will actually generate. |
| 595 | |
| 596 | Ceil to frames then snap upward so playback length is >= the request |
| 597 | (unless capped by ``ECHO_MAX_NUM_FRAMES``). |
| 598 | """ |
| 599 | import math |
| 600 | |
| 601 | return snap_echo_num_frames(math.ceil(float(duration_sec) * ECHO_SHOT_FPS)) |
| 602 | |
| 603 | |
| 604 | def num_frames_to_duration_sec(num_frames: int) -> float: |
| 605 | """Return duration in whole seconds (nearest second of frames/fps) for UI/state.""" |
| 606 | return float(round(int(num_frames) / ECHO_SHOT_FPS)) |
| 607 | |
| 608 | |
| 609 | def num_frames_to_exact_duration_sec(num_frames: int) -> float: |
| 610 | """Exact playback seconds for the given frame count (no rounding). |
| 611 | |
| 612 | Sent to the Echo backend so it does not re-derive frames from a rounded |
| 613 | ``duration_sec`` (e.g. 7.0) and snap back downward. |
| 614 | """ |
| 615 | return float(num_frames) / float(ECHO_SHOT_FPS) |
| 616 | |
| 617 | |
| 618 | def sync_shot_echo_duration(shot: dict[str, Any], duration_sec: float) -> int: |
| 619 | """Persist snapped Echo timing on the shot record.""" |
| 620 | num_frames = duration_sec_to_num_frames(duration_sec) |
| 621 | actual_duration_sec = num_frames_to_duration_sec(num_frames) |
| 622 | shot["duration_sec"] = actual_duration_sec |
| 623 | shot["num_frames"] = num_frames |
| 624 | return num_frames |
| 625 | |
| 626 | |
| 627 | def resolve_echo_duration_seconds( |
| 628 | shot: dict[str, Any], |
| 629 | state: dict[str, Any] | None = None, |
| 630 | ) -> float: |
| 631 | """Resolve per-shot seconds for Echo generation (aligned with workplace UI defaults).""" |
| 632 | for key in ("duration_sec", "duration_seconds"): |
| 633 | try: |
| 634 | value = float(shot.get(key)) |
| 635 | if value > 0: |
| 636 | return value |
| 637 | except (TypeError, ValueError): |
| 638 | pass |
| 639 | try: |
| 640 | shot_frames = shot.get("num_frames") |
| 641 | if shot_frames is not None: |
| 642 | value = float(num_frames_to_duration_sec(int(shot_frames))) |
| 643 | if value > 0: |
| 644 | return value |
| 645 | except (TypeError, ValueError): |
| 646 | pass |
| 647 | goal = ( |
| 648 | state.get("goal") if isinstance(state, dict) and isinstance(state.get("goal"), dict) else {} |
| 649 | ) |
| 650 | try: |
| 651 | goal_duration = float(goal.get("shot_duration_sec") or 0) |
| 652 | if goal_duration > 0: |
| 653 | return goal_duration |
| 654 | except (TypeError, ValueError): |
| 655 | pass |
| 656 | return ECHO_DEFAULT_DURATION_SEC |
| 657 | |
| 658 | |
| 659 | def rewrite_prompt_for_i2v(original_prompt: str, caption_language: str) -> str: |
| 660 | """Rewrite a shot prompt for I2V by prepending the language-matched first-frame sentence. |
| 661 | |
| 662 | Follows ``pe/v7_cinematic_full/skills/i2v-tail-frame-prompt-rewriter/SKILL.md``. |
| 663 | Only the opening sentence is added; the rest of the prompt is preserved unchanged. |
| 664 | """ |
| 665 | caption_language = (caption_language or "").strip().lower() |
| 666 | is_chinese = caption_language in {"simplified chinese", "zh", "chinese", "mandarin chinese"} |
| 667 | |
| 668 | first_frame_zh = ( |
| 669 | "以当前图片作为视频首帧，并基于首帧中已有的人物、物体、环境、构图、机位、光线和动作状态自然延续。" |
| 670 | ) |
| 671 | first_frame_en = ( |
| 672 | "Use the current image as the first frame of the video, and continue naturally " |
| 673 | "from the characters, objects, environment, composition, camera position, lighting, " |
| 674 | "and action state already shown in it." |
| 675 | ) |
| 676 | |
| 677 | first_frame_sentence = first_frame_zh if is_chinese else first_frame_en |
| 678 | |
| 679 | trimmed = original_prompt.strip() |
| 680 | |
| 681 | # Detect if the prompt has a cut-count style opening (e.g. "1 cut" / "1个镜头"). |
| 682 | # Insert the first-frame sentence before the cut-count sentence. |
| 683 | cut_count_pattern = re.compile( |
| 684 | r"^(\d+)\s*(?:cuts?|个镜头|个景别)", |
| 685 | re.IGNORECASE, |
| 686 | ) |
| 687 | match = cut_count_pattern.match(trimmed) |
| 688 | if match: |
| 689 | prefix = trimmed[: match.end()] |
| 690 | rest = trimmed[match.end() :] |
| 691 | return f"{first_frame_sentence}\n{prefix}{rest}" |
| 692 | |
| 693 | return f"{first_frame_sentence}\n{trimmed}" |
| 694 | |
| 695 | |
| 696 | class DirectorTool(Tool): |
| 697 | """Shared helpers for director-state tools.""" |
| 698 | |
| 699 | _DEFAULT_STAGE = "story_discussion" |
| 700 | _FINAL_STAGES = frozenset({"done", "cancelled"}) |
| 701 | |
| 702 | @property |
| 703 | def description(self) -> str: |
| 704 | """Pull each tool's description from the active PE set, keyed by tool name.""" |
| 705 | return prompts.text(f"director.tool.{self.name}.description") |
| 706 | |
| 707 | def __init__( |
| 708 | self, |
| 709 | workspace: Path, |
| 710 | *, |
| 711 | tools_config: Any | None = None, |
| 712 | callback_base_url: str | None = None, |
| 713 | ): |
| 714 | from nanobot.config.schema import ToolsConfig |
| 715 | |
| 716 | self.workspace = workspace |
| 717 | self._tools_config = tools_config or ToolsConfig() |
| 718 | self._callback_base_url = ( |
| 719 | callback_base_url.rstrip("/") |
| 720 | if isinstance(callback_base_url, str) and callback_base_url.strip() |
| 721 | else None |
| 722 | ) |
| 723 | self._channel: ContextVar[str] = ContextVar("director_channel", default="cli") |
| 724 | self._chat_id: ContextVar[str] = ContextVar("director_chat_id", default="direct") |
| 725 | self._session_key: ContextVar[str] = ContextVar( |
| 726 | "director_session_key", |
| 727 | default="cli:direct", |
| 728 | ) |
| 729 | |
| 730 | def set_context( |
| 731 | self, |
| 732 | channel: str, |
| 733 | chat_id: str, |
| 734 | effective_key: str | None = None, |
| 735 | *, |
| 736 | injected_event: str | None | object = _WORKFLOW_CONTEXT_UNSET, |
| 737 | ) -> None: |
| 738 | self._channel.set(channel) |
| 739 | self._chat_id.set(chat_id) |
| 740 | self._session_key.set(effective_key or f"{channel}:{chat_id}") |
| 741 | if injected_event is not _WORKFLOW_CONTEXT_UNSET: |
| 742 | _WORKFLOW_INJECTED_EVENT.set(injected_event) |
| 743 | |
| 744 | @staticmethod |
| 745 | def allow_workflow_gate_bypass() -> None: |
| 746 | """Test helper: allow gated workflow tools without a workplace injection.""" |
| 747 | _WORKFLOW_INJECTED_EVENT.set(WORKFLOW_GATE_BYPASS) |
| 748 | |
| 749 | @property |
| 750 | def director_root(self) -> Path: |
| 751 | return self.workspace / "director" |
| 752 | |
| 753 | @property |
| 754 | def works_root(self) -> Path: |
| 755 | return self.director_root / "works" |
| 756 | |
| 757 | @property |
| 758 | def active_work_path(self) -> Path: |
| 759 | return self.director_root / "active_work.json" |
| 760 | |
| 761 | @property |
| 762 | def session_map_path(self) -> Path: |
| 763 | return self.director_root / "session_map.json" |
| 764 | |
| 765 | def _ensure_root(self) -> None: |
| 766 | self.works_root.mkdir(parents=True, exist_ok=True) |
| 767 | |
| 768 | def _read_reference_image_from_session(self) -> dict[str, Any] | None: |
| 769 | """从 session metadata 读取首帧参考图信息。""" |
| 770 | try: |
| 771 | from nanobot.session.manager import SessionManager |
| 772 | |
| 773 | session_key = self._session_key.get() |
| 774 | if not session_key: |
| 775 | return None |
| 776 | session = SessionManager(self.workspace).get_or_create(session_key) |
| 777 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 778 | return normalize_reference_image(metadata.get("reference_image")) |
| 779 | except Exception: |
| 780 | logger.exception("director: failed to read session reference_image") |
| 781 | return None |
| 782 | |
| 783 | def _session_auto_generate(self) -> bool: |
| 784 | try: |
| 785 | from nanobot.session.manager import SessionManager |
| 786 | |
| 787 | session_key = self._session_key.get() |
| 788 | if not session_key: |
| 789 | return False |
| 790 | session = SessionManager(self.workspace).get_or_create(session_key) |
| 791 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 792 | return get_auto_generate(metadata) |
| 793 | except Exception: |
| 794 | logger.exception("director: failed to read session auto_generate") |
| 795 | return False |
| 796 | |
| 797 | def _consume_shot_count_next_step_hint( |
| 798 | self, |
| 799 | *, |
| 800 | auto_generate: bool = False, |
| 801 | emit: bool = True, |
| 802 | ) -> str | None: |
| 803 | work_id = self._active_work_id() |
| 804 | if not work_id: |
| 805 | return None |
| 806 | state = self._load_state(work_id) |
| 807 | pending = bool(state.pop(SHOT_COUNT_NEXT_STEP_HINT_PENDING_KEY, False)) |
| 808 | if not pending: |
| 809 | return None |
| 810 | stage = str(state.get("stage") or "") |
| 811 | should_emit = ( |
| 812 | emit |
| 813 | and not auto_generate |
| 814 | and not bool(state.get("auto_generate")) |
| 815 | and stage not in _STAGES_PAST_SHOT_COUNT_HINT |
| 816 | ) |
| 817 | self._save_state(work_id, state) |
| 818 | return SHOT_COUNT_NEXT_STEP_HINT if should_emit else None |
| 819 | |
| 820 | def _lock_session_reference_image(self) -> None: |
| 821 | try: |
| 822 | from nanobot.session.manager import SessionManager |
| 823 | |
| 824 | session_key = self._session_key.get() |
| 825 | if not session_key: |
| 826 | return |
| 827 | manager = SessionManager(self.workspace) |
| 828 | session = manager.get_or_create(session_key) |
| 829 | if not isinstance(session.metadata, dict): |
| 830 | session.metadata = {} |
| 831 | if session.metadata.get("reference_image_locked") is True: |
| 832 | return |
| 833 | session.metadata["reference_image_locked"] = True |
| 834 | manager.save(session) |
| 835 | except Exception: |
| 836 | logger.exception("director: failed to lock session reference_image") |
| 837 | |
| 838 | def _clear_reference_image_story_rewrite_flag(self) -> None: |
| 839 | try: |
| 840 | from nanobot.session.manager import SessionManager |
| 841 | |
| 842 | session_key = self._session_key.get() |
| 843 | if not session_key: |
| 844 | return |
| 845 | manager = SessionManager(self.workspace) |
| 846 | session = manager.get_or_create(session_key) |
| 847 | if not isinstance(session.metadata, dict): |
| 848 | return |
| 849 | if not session.metadata.get("reference_image_needs_story_rewrite"): |
| 850 | return |
| 851 | clear_reference_image_needs_story_rewrite(session.metadata) |
| 852 | manager.save(session) |
| 853 | except Exception: |
| 854 | logger.exception( |
| 855 | "director: failed to clear reference_image_needs_story_rewrite" |
| 856 | ) |
| 857 | |
| 858 | def _session_reference_inject_failed(self) -> bool: |
| 859 | try: |
| 860 | from nanobot.session.manager import SessionManager |
| 861 | |
| 862 | session_key = self._session_key.get() |
| 863 | if not session_key: |
| 864 | return False |
| 865 | session = SessionManager(self.workspace).get_or_create(session_key) |
| 866 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 867 | return bool(metadata.get("reference_image_inject_failed")) |
| 868 | except Exception: |
| 869 | logger.exception("director: failed to read reference_image_inject_failed") |
| 870 | return False |
| 871 | |
| 872 | def _session_reference_needs_rewrite(self) -> bool: |
| 873 | try: |
| 874 | from nanobot.session.manager import SessionManager |
| 875 | |
| 876 | session_key = self._session_key.get() |
| 877 | if not session_key: |
| 878 | return False |
| 879 | session = SessionManager(self.workspace).get_or_create(session_key) |
| 880 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 881 | return reference_image_needs_story_rewrite(metadata) |
| 882 | except Exception: |
| 883 | logger.exception("director: failed to read reference_image_needs_story_rewrite") |
| 884 | return False |
| 885 | |
| 886 | def _effective_auto_generate_shot_count(self, goal: dict[str, Any] | None) -> int | None: |
| 887 | try: |
| 888 | from nanobot.session.manager import SessionManager |
| 889 | |
| 890 | session_key = self._session_key.get() |
| 891 | metadata: dict[str, Any] | None = None |
| 892 | if session_key: |
| 893 | session = SessionManager(self.workspace).get_or_create(session_key) |
| 894 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 895 | return effective_auto_generate_shot_count(goal=goal, metadata=metadata) |
| 896 | except Exception: |
| 897 | logger.exception("director: failed to resolve auto_generate shot_count") |
| 898 | return locked_shot_count_from_goal(goal) |
| 899 | |
| 900 | def _state_first_frame_url(self, state: dict[str, Any]) -> str | None: |
| 901 | ref = normalize_reference_image(state.get("reference_image")) |
| 902 | if not ref: |
| 903 | return None |
| 904 | url = ref.get("url") |
| 905 | return url if isinstance(url, str) and url.strip() else None |
| 906 | |
| 907 | # ── tail-frame extraction pipeline (shared by agent + REST paths) ── |
| 908 | |
| 909 | @staticmethod |
| 910 | def _extract_tail_frame(video_path: Path, output_path: Path) -> bool: |
| 911 | """Extract the last frame of *video_path* as a PNG using ffmpeg.""" |
| 912 | from nanobot.director.memory_coordinator import _resolve_media_binary |
| 913 | |
| 914 | ffmpeg = _resolve_media_binary("ffmpeg") |
| 915 | cmd = [ |
| 916 | ffmpeg, "-sseof", "-1", "-i", str(video_path), |
| 917 | "-update", "1", "-q:v", "1", str(output_path), "-y", |
| 918 | ] |
| 919 | import subprocess |
| 920 | |
| 921 | try: |
| 922 | subprocess.run(cmd, check=True, capture_output=True, timeout=60) |
| 923 | return output_path.is_file() and output_path.stat().st_size > 0 |
| 924 | except Exception: |
| 925 | return False |
| 926 | |
| 927 | def _publish_tail_frame( |
| 928 | self, image_path: Path, work_id: str, shot_id: int |
| 929 | ) -> str | None: |
| 930 | """Persist tail frame locally. Returns the public URL or None on failure.""" |
| 931 | from nanobot.storage.files import configured_file_publisher |
| 932 | |
| 933 | name = f"tail_frames/shot_{shot_id:03d}.png" |
| 934 | try: |
| 935 | publisher = configured_file_publisher( |
| 936 | work_id, |
| 937 | storage=self._tools_config.file_storage, |
| 938 | workspace=self.workspace, |
| 939 | ) |
| 940 | return publisher(str(image_path), name) |
| 941 | except Exception: |
| 942 | return None |
| 943 | |
| 944 | def _extract_and_publish_tail_frame( |
| 945 | self, work_id: str, shot_id: int, video_url: str, |
| 946 | ) -> str | None: |
| 947 | """Download video, extract last frame, publish locally. Returns public URL.""" |
| 948 | import shutil |
| 949 | import tempfile |
| 950 | import urllib.request |
| 951 | |
| 952 | tmp_dir = Path(tempfile.mkdtemp(prefix="tail_frame_")) |
| 953 | try: |
| 954 | video_path = tmp_dir / "source.mp4" |
| 955 | frame_path = tmp_dir / "tail.png" |
| 956 | |
| 957 | # download |
| 958 | if video_url.startswith(("http://", "https://")): |
| 959 | req = urllib.request.Request(video_url, headers={"Accept": "video/mp4,*/*"}) |
| 960 | with urllib.request.urlopen(req, timeout=120) as resp: |
| 961 | with open(video_path, "wb") as f: |
| 962 | shutil.copyfileobj(resp, f) |
| 963 | else: |
| 964 | src = Path(video_url) |
| 965 | if not src.is_file(): |
| 966 | return None |
| 967 | shutil.copyfile(str(src), str(video_path)) |
| 968 | |
| 969 | if not video_path.is_file() or video_path.stat().st_size <= 0: |
| 970 | return None |
| 971 | |
| 972 | # extract |
| 973 | if not DirectorTool._extract_tail_frame(video_path, frame_path): |
| 974 | return None |
| 975 | |
| 976 | # upload |
| 977 | return self._publish_tail_frame(frame_path, work_id, shot_id) |
| 978 | except Exception: |
| 979 | return None |
| 980 | finally: |
| 981 | shutil.rmtree(tmp_dir, ignore_errors=True) |
| 982 | |
| 983 | @staticmethod |
| 984 | def _read_json(path: Path, default: Any) -> Any: |
| 985 | if not path.exists(): |
| 986 | return default |
| 987 | try: |
| 988 | return json.loads(path.read_text(encoding="utf-8")) |
| 989 | except (json.JSONDecodeError, OSError): |
| 990 | return default |
| 991 | |
| 992 | @staticmethod |
| 993 | def _write_json(path: Path, data: Any) -> None: |
| 994 | write_json_atomic(path, data) |
| 995 | |
| 996 | @staticmethod |
| 997 | def _write_text(path: Path, content: str) -> None: |
| 998 | path.parent.mkdir(parents=True, exist_ok=True) |
| 999 | path.write_text(content.rstrip() + "\n", encoding="utf-8") |
| 1000 | |
| 1001 | def _remote_http_base_url(self) -> str | None: |
| 1002 | echo_generator = getattr(self._tools_config, "echo_generator", None) |
| 1003 | base = "" |
| 1004 | if echo_generator is not None: |
| 1005 | base = str(getattr(echo_generator, "base_url", "") or "").strip() |
| 1006 | return base.rstrip("/") if base else None |
| 1007 | |
| 1008 | def _remote_callback_base_url(self) -> str | None: |
| 1009 | if self._callback_base_url: |
| 1010 | return self._callback_base_url |
| 1011 | echo_generator = getattr(self._tools_config, "echo_generator", None) |
| 1012 | base = "" |
| 1013 | if echo_generator is not None: |
| 1014 | base = str(getattr(echo_generator, "callback_base_url", "") or "").strip() |
| 1015 | return base.rstrip("/") if base else None |
| 1016 | |
| 1017 | def _remote_http_timeout_sec(self) -> float: |
| 1018 | echo_generator = getattr(self._tools_config, "echo_generator", None) |
| 1019 | raw_timeout = ( |
| 1020 | getattr(echo_generator, "http_timeout_sec", 30.0) if echo_generator is not None else 30.0 |
| 1021 | ) |
| 1022 | try: |
| 1023 | return max(1.0, float(raw_timeout)) |
| 1024 | except (TypeError, ValueError): |
| 1025 | return 30.0 |
| 1026 | |
| 1027 | def _remote_endpoint_path(self, operation: str) -> str: |
| 1028 | endpoint = _REMOTE_ENDPOINT_PATHS.get(operation) |
| 1029 | if not endpoint: |
| 1030 | raise RuntimeError(f"No remote endpoint mapping exists for operation '{operation}'.") |
| 1031 | return endpoint |
| 1032 | |
| 1033 | def _remote_callback_path(self, operation: str) -> str | None: |
| 1034 | return _REMOTE_CALLBACK_PATHS.get(operation) |
| 1035 | |
| 1036 | def _remote_callback_url(self, operation: str) -> str | None: |
| 1037 | base_url = self._remote_callback_base_url() |
| 1038 | callback_path = self._remote_callback_path(operation) |
| 1039 | if not base_url or not callback_path: |
| 1040 | return None |
| 1041 | return f"{base_url}{callback_path}" |
| 1042 | |
| 1043 | def _build_remote_callback_contract( |
| 1044 | self, |
| 1045 | work_id: str, |
| 1046 | job_id: str, |
| 1047 | operation: str, |
| 1048 | target: str | list[str], |
| 1049 | ) -> dict[str, Any]: |
| 1050 | contract = { |
| 1051 | "event_type": "director_remote_result", |
| 1052 | "protocol_version": _REMOTE_PROTOCOL_VERSION, |
| 1053 | "operation": operation, |
| 1054 | "work_id": work_id, |
| 1055 | "job_id": job_id, |
| 1056 | "target": target, |
| 1057 | "channel": self._channel.get(), |
| 1058 | "chat_id": self._chat_id.get(), |
| 1059 | "session_key": self._session_key.get(), |
| 1060 | "inject_back_to_agent": True, |
| 1061 | "note": ( |
| 1062 | "When the backend finishes, your client-side callback handler should " |
| 1063 | "clear the pending_remote_jobs entry, update the director workspace, " |
| 1064 | "and publish an InboundMessage for this session." |
| 1065 | ), |
| 1066 | } |
| 1067 | callback_url = self._remote_callback_url(operation) |
| 1068 | if callback_url: |
| 1069 | contract["url"] = callback_url |
| 1070 | return contract |
| 1071 | |
| 1072 | def _build_remote_request_envelope( |
| 1073 | self, |
| 1074 | operation: str, |
| 1075 | work_id: str, |
| 1076 | job_id: str, |
| 1077 | target: str | list[str], |
| 1078 | payload: dict[str, Any], |
| 1079 | ) -> dict[str, Any]: |
| 1080 | return { |
| 1081 | "protocol_version": _REMOTE_PROTOCOL_VERSION, |
| 1082 | "operation": operation, |
| 1083 | "job": { |
| 1084 | "job_id": job_id, |
| 1085 | "work_id": work_id, |
| 1086 | "target": target, |
| 1087 | "created_at": _now_iso(), |
| 1088 | }, |
| 1089 | "callback": self._build_remote_callback_contract(work_id, job_id, operation, target), |
| 1090 | "payload": payload, |
| 1091 | } |
| 1092 | |
| 1093 | def _post_remote_http_request( |
| 1094 | self, |
| 1095 | endpoint_url: str, |
| 1096 | envelope: dict[str, Any], |
| 1097 | ) -> dict[str, Any]: |
| 1098 | headers = { |
| 1099 | "Content-Type": "application/json", |
| 1100 | "Accept": "application/json", |
| 1101 | } |
| 1102 | callback = envelope.get("callback") |
| 1103 | if isinstance(callback, dict): |
| 1104 | callback_url = callback.get("url") |
| 1105 | if isinstance(callback_url, str) and callback_url.strip(): |
| 1106 | headers["X-Nanobot-Director-Callback-Url"] = callback_url.strip() |
| 1107 | body = json.dumps(envelope, ensure_ascii=False).encode("utf-8") |
| 1108 | request = urllib_request.Request( |
| 1109 | endpoint_url, |
| 1110 | data=body, |
| 1111 | headers=headers, |
| 1112 | method="POST", |
| 1113 | ) |
| 1114 | try: |
| 1115 | with urllib_request.urlopen( |
| 1116 | request, timeout=self._remote_http_timeout_sec() |
| 1117 | ) as response: |
| 1118 | raw = response.read().decode("utf-8") |
| 1119 | except urllib_error.URLError as exc: |
| 1120 | raise RuntimeError(f"Remote HTTP request failed: {exc}") from exc |
| 1121 | if not raw.strip(): |
| 1122 | return {} |
| 1123 | try: |
| 1124 | parsed = json.loads(raw) |
| 1125 | except json.JSONDecodeError: |
| 1126 | return {"raw_response": raw} |
| 1127 | return parsed if isinstance(parsed, dict) else {"response": parsed} |
| 1128 | |
| 1129 | def _active_work_id(self) -> str | None: |
| 1130 | session_map = self._read_json(self.session_map_path, {}) |
| 1131 | if not isinstance(session_map, dict): |
| 1132 | return None |
| 1133 | entry = session_map.get(self._session_key.get()) |
| 1134 | if isinstance(entry, dict): |
| 1135 | return entry.get("active") |
| 1136 | # Backwards compat: old format stored bare work_id string |
| 1137 | if isinstance(entry, str): |
| 1138 | return entry |
| 1139 | return None |
| 1140 | |
| 1141 | def _session_work_history(self) -> list[str]: |
| 1142 | session_map = self._read_json(self.session_map_path, {}) |
| 1143 | if not isinstance(session_map, dict): |
| 1144 | return [] |
| 1145 | entry = session_map.get(self._session_key.get()) |
| 1146 | if isinstance(entry, dict): |
| 1147 | history = entry.get("history", []) |
| 1148 | return history if isinstance(history, list) else [] |
| 1149 | # Backwards compat: old format stored bare work_id string |
| 1150 | if isinstance(entry, str): |
| 1151 | return [entry] |
| 1152 | return [] |
| 1153 | |
| 1154 | def _set_active_work(self, work_id: str) -> None: |
| 1155 | self._ensure_root() |
| 1156 | session_key = self._session_key.get() |
| 1157 | session_map = self._read_json(self.session_map_path, {}) |
| 1158 | if not isinstance(session_map, dict): |
| 1159 | session_map = {} |
| 1160 | entry = session_map.get(session_key) |
| 1161 | # Migrate old bare-string entries |
| 1162 | if isinstance(entry, str): |
| 1163 | entry = {"active": entry, "history": [entry]} |
| 1164 | elif not isinstance(entry, dict): |
| 1165 | entry = {"active": None, "history": []} |
| 1166 | history = entry.get("history", []) |
| 1167 | if not isinstance(history, list): |
| 1168 | history = [] |
| 1169 | if work_id not in history: |
| 1170 | history.append(work_id) |
| 1171 | entry["active"] = work_id |
| 1172 | entry["history"] = history |
| 1173 | session_map[session_key] = entry |
| 1174 | self._write_json(self.session_map_path, session_map) |
| 1175 | self._write_json( |
| 1176 | self.active_work_path, |
| 1177 | { |
| 1178 | "work_id": work_id, |
| 1179 | "channel": self._channel.get(), |
| 1180 | "chat_id": self._chat_id.get(), |
| 1181 | "session_key": session_key, |
| 1182 | "updated_at": _now_iso(), |
| 1183 | }, |
| 1184 | ) |
| 1185 | |
| 1186 | def _resolve_work_id(self, work_id: str | None = None) -> tuple[str | None, Path | None]: |
| 1187 | self._ensure_root() |
| 1188 | candidate = work_id or self._active_work_id() |
| 1189 | if not candidate: |
| 1190 | return None, None |
| 1191 | work_dir = self.works_root / candidate |
| 1192 | if not work_dir.exists(): |
| 1193 | return None, None |
| 1194 | return candidate, work_dir |
| 1195 | |
| 1196 | def _paths(self, work_id: str) -> dict[str, Path]: |
| 1197 | work_dir = self.works_root / work_id |
| 1198 | return { |
| 1199 | "work_dir": work_dir, |
| 1200 | "state": work_dir / "state.json", |
| 1201 | "fact": work_dir / "fact.md", |
| 1202 | "work_memory": work_dir / "work_memory_lite.md", |
| 1203 | "story": work_dir / "story.md", |
| 1204 | "story_profile": work_dir / "story_profile.json", |
| 1205 | "shots": work_dir / "shots", |
| 1206 | "jobs": work_dir / "jobs", |
| 1207 | "outputs": work_dir / "outputs", |
| 1208 | "memory_bank": work_dir / "memory" / "memory_bank.json", |
| 1209 | "previous_shot_memory": work_dir / "memory" / "previous_shot.json", |
| 1210 | "manual_memory_workspace": work_dir / "memory" / "manual" / "workspace.json", |
| 1211 | "memory_asset_profiles": work_dir / "memory" / "asset_profiles.json", |
| 1212 | } |
| 1213 | |
| 1214 | @staticmethod |
| 1215 | def _automatic_memory_asset_id(raw: dict[str, Any], memory_id: str, kind: str) -> str: |
| 1216 | fingerprint = json.dumps( |
| 1217 | [ |
| 1218 | memory_id, |
| 1219 | int(raw.get("source_shot_id") or 0), |
| 1220 | int(raw.get("frame_index") or 0), |
| 1221 | kind, |
| 1222 | ], |
| 1223 | ensure_ascii=False, |
| 1224 | separators=(",", ":"), |
| 1225 | ) |
| 1226 | return "auto_" + hashlib.sha256(fingerprint.encode("utf-8")).hexdigest()[:20] |
| 1227 | |
| 1228 | def _memory_asset_catalog(self, work_id: str) -> list[dict[str, Any]]: |
| 1229 | """Return profile-bearing assets safe for the agent to reason over.""" |
| 1230 | paths = self._paths(work_id) |
| 1231 | overrides = self._read_json(paths["memory_asset_profiles"], {}) |
| 1232 | overrides = overrides if isinstance(overrides, dict) else {} |
| 1233 | assets: list[dict[str, Any]] = [] |
| 1234 | |
| 1235 | def add_automatic(raw: Any, memory_id: str, kind: str) -> None: |
| 1236 | if not isinstance(raw, dict): |
| 1237 | return |
| 1238 | asset_id = self._automatic_memory_asset_id(raw, memory_id, kind) |
| 1239 | override = overrides.get(asset_id) |
| 1240 | override = override if isinstance(override, dict) else {} |
| 1241 | profile_text = str( |
| 1242 | override.get("profile_text") |
| 1243 | or raw.get("profile_text") |
| 1244 | or raw.get("reasoning") |
| 1245 | or "" |
| 1246 | ).strip() |
| 1247 | if not profile_text: |
| 1248 | return |
| 1249 | identities = override.get("identity_ids") or raw.get("visible_character_ids") |
| 1250 | if not isinstance(identities, list): |
| 1251 | identities = [memory_id] if memory_id.startswith("ID_") else [] |
| 1252 | reference_type = str( |
| 1253 | override.get("reference_type") |
| 1254 | if "reference_type" in override |
| 1255 | else raw.get("reference_type") or "" |
| 1256 | ).strip() |
| 1257 | reference_label = str( |
| 1258 | override.get("reference_label") |
| 1259 | if "reference_label" in override |
| 1260 | else raw.get("reference_label") or "" |
| 1261 | ).strip() |
| 1262 | assets.append({ |
| 1263 | "asset_id": asset_id, |
| 1264 | "media_type": ( |
| 1265 | "image_audio" if raw.get("image_path") and raw.get("audio_path") |
| 1266 | else "audio" if raw.get("audio_path") |
| 1267 | else "image" |
| 1268 | ), |
| 1269 | "profile_text": profile_text, |
| 1270 | "identity_ids": [str(value) for value in identities if str(value).strip()], |
| 1271 | **({"reference_type": reference_type} if reference_type else {}), |
| 1272 | **({"reference_label": reference_label} if reference_label else {}), |
| 1273 | "source": { |
| 1274 | "type": "generated_shot", |
| 1275 | "shot_id": int(raw.get("source_shot_id") or 0), |
| 1276 | "timestamp_sec": float(raw.get("timestamp_sec") or 0), |
| 1277 | }, |
| 1278 | }) |
| 1279 | |
| 1280 | bank = self._read_json(paths["memory_bank"], {}) |
| 1281 | if isinstance(bank, dict): |
| 1282 | for memory_id, raw in bank.items(): |
| 1283 | add_automatic(raw, str(memory_id), "character") |
| 1284 | previous = self._read_json(paths["previous_shot_memory"], None) |
| 1285 | add_automatic(previous, "PREVIOUS_SHOT", "previous_shot") |
| 1286 | |
| 1287 | manual = self._read_json(paths["manual_memory_workspace"], {}) |
| 1288 | rows = manual.get("assets") if isinstance(manual, dict) else [] |
| 1289 | for raw in rows if isinstance(rows, list) else []: |
| 1290 | if not isinstance(raw, dict): |
| 1291 | continue |
| 1292 | profile_text = str(raw.get("profile_text") or "").strip() |
| 1293 | asset_id = str(raw.get("asset_id") or "").strip() |
| 1294 | if not asset_id or not profile_text: |
| 1295 | continue |
| 1296 | has_image = bool(raw.get("image_path")) |
| 1297 | has_audio = bool(raw.get("audio_path")) |
| 1298 | reference_type = str(raw.get("reference_type") or "").strip() |
| 1299 | reference_label = str(raw.get("reference_label") or "").strip() |
| 1300 | source_shot_id = int(raw.get("source_shot_id") or 0) |
| 1301 | source = ( |
| 1302 | { |
| 1303 | "type": "generated_shot", |
| 1304 | "shot_id": source_shot_id, |
| 1305 | "timestamp_sec": float(raw.get("timestamp_sec") or 0), |
| 1306 | **( |
| 1307 | {"audio_start_sec": float(raw["audio_start_sec"])} |
| 1308 | if raw.get("audio_start_sec") is not None |
| 1309 | else {} |
| 1310 | ), |
| 1311 | **( |
| 1312 | {"audio_end_sec": float(raw["audio_end_sec"])} |
| 1313 | if raw.get("audio_end_sec") is not None |
| 1314 | else {} |
| 1315 | ), |
| 1316 | } |
| 1317 | if source_shot_id > 0 |
| 1318 | else {"type": "local_upload"} |
| 1319 | ) |
| 1320 | assets.append({ |
| 1321 | "asset_id": asset_id, |
| 1322 | "media_type": ( |
| 1323 | "image_audio" if has_image and has_audio |
| 1324 | else "audio" if has_audio |
| 1325 | else "image" |
| 1326 | ), |
| 1327 | "profile_text": profile_text, |
| 1328 | "identity_ids": [ |
| 1329 | str(value) for value in raw.get("identity_ids", []) if str(value).strip() |
| 1330 | ], |
| 1331 | **({"reference_type": reference_type} if reference_type else {}), |
| 1332 | **({"reference_label": reference_label} if reference_label else {}), |
| 1333 | "source": source, |
| 1334 | }) |
| 1335 | return assets |
| 1336 | |
| 1337 | def _load_state(self, work_id: str) -> dict[str, Any]: |
| 1338 | state = self._read_json(self._paths(work_id)["state"], {}) |
| 1339 | return state if isinstance(state, dict) else {} |
| 1340 | |
| 1341 | def _save_state(self, work_id: str, state: dict[str, Any]) -> None: |
| 1342 | state["story_profile"] = self._load_story_profile(work_id) |
| 1343 | state["updated_at"] = _now_iso() |
| 1344 | self._write_json(self._paths(work_id)["state"], state) |
| 1345 | |
| 1346 | def _shot_path(self, work_id: str, shot_id: int) -> Path: |
| 1347 | return self._paths(work_id)["shots"] / f"{_shot_key(shot_id)}.json" |
| 1348 | |
| 1349 | def _job_path(self, work_id: str, job_id: str) -> Path: |
| 1350 | return self._paths(work_id)["jobs"] / f"{job_id}.json" |
| 1351 | |
| 1352 | def _load_job(self, work_id: str, job_id: str) -> dict[str, Any]: |
| 1353 | data = self._read_json(self._job_path(work_id, job_id), {}) |
| 1354 | return data if isinstance(data, dict) else {} |
| 1355 | |
| 1356 | def _save_job(self, work_id: str, job_id: str, job: dict[str, Any]) -> None: |
| 1357 | self._write_json(self._job_path(work_id, job_id), job) |
| 1358 | |
| 1359 | def _load_shot(self, work_id: str, shot_id: int) -> dict[str, Any]: |
| 1360 | data = self._read_json(self._shot_path(work_id, shot_id), {}) |
| 1361 | return data if isinstance(data, dict) else {} |
| 1362 | |
| 1363 | def _save_shot(self, work_id: str, shot_id: int, shot: dict[str, Any]) -> None: |
| 1364 | shot["updated_at"] = _now_iso() |
| 1365 | self._write_json(self._shot_path(work_id, shot_id), shot) |
| 1366 | |
| 1367 | def _load_story_profile(self, work_id: str) -> dict[str, Any]: |
| 1368 | data = self._read_json(self._paths(work_id)["story_profile"], {}) |
| 1369 | return data if isinstance(data, dict) else {} |
| 1370 | |
| 1371 | def _save_story_profile(self, work_id: str, story_profile: dict[str, Any]) -> None: |
| 1372 | previous = self._load_story_profile(work_id) |
| 1373 | profile = dict(story_profile) |
| 1374 | _preserve_story_profile_language(profile, previous) |
| 1375 | # Prefer session UI language when profile still has none. |
| 1376 | if "language" not in profile: |
| 1377 | from nanobot.session.generation_settings import get_generation_settings |
| 1378 | from nanobot.session.manager import SessionManager |
| 1379 | |
| 1380 | session = SessionManager(self.workspace).get_or_create(self._session_key.get()) |
| 1381 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 1382 | settings = get_generation_settings(metadata) |
| 1383 | _apply_story_profile_language(profile, str(settings.get("language") or "")) |
| 1384 | _normalize_story_profile(profile) |
| 1385 | self._write_json(self._paths(work_id)["story_profile"], profile) |
| 1386 | |
| 1387 | def _default_state(self, work_id: str, *, title: str | None, goal: str) -> dict[str, Any]: |
| 1388 | return { |
| 1389 | "work_id": work_id, |
| 1390 | "title": title or "", |
| 1391 | "goal_brief": goal, |
| 1392 | "stage": self._DEFAULT_STAGE, |
| 1393 | "story_confirmed": False, |
| 1394 | "goal": { |
| 1395 | "shot_count": None, |
| 1396 | "shot_duration_sec": None, |
| 1397 | "generation_mode": "sequential", |
| 1398 | }, |
| 1399 | "shots": {}, |
| 1400 | "pending_remote_jobs": {}, |
| 1401 | "latest_story_summary": "", |
| 1402 | "story_profile": {}, |
| 1403 | "latest_merge_job_id": None, |
| 1404 | "final_output_path": None, |
| 1405 | "final_output_url": None, |
| 1406 | "reference_image": None, |
| 1407 | "reference_image_locked": False, |
| 1408 | "auto_generate": False, |
| 1409 | "created_at": _now_iso(), |
| 1410 | "updated_at": _now_iso(), |
| 1411 | } |
| 1412 | |
| 1413 | def _ensure_work_files(self, work_id: str, *, title: str | None, goal: str) -> dict[str, Any]: |
| 1414 | self._ensure_root() |
| 1415 | paths = self._paths(work_id) |
| 1416 | for key in ("work_dir", "shots", "jobs", "outputs"): |
| 1417 | paths[key].mkdir(parents=True, exist_ok=True) |
| 1418 | if not paths["story"].exists(): |
| 1419 | self._write_text(paths["story"], "") |
| 1420 | if not paths["work_memory"].exists(): |
| 1421 | self._write_text(paths["work_memory"], "# Work Memory Lite\n") |
| 1422 | if not paths["story_profile"].exists(): |
| 1423 | from nanobot.session.generation_settings import get_generation_settings |
| 1424 | from nanobot.session.manager import SessionManager |
| 1425 | |
| 1426 | profile: dict[str, Any] = {} |
| 1427 | session = SessionManager(self.workspace).get_or_create(self._session_key.get()) |
| 1428 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 1429 | settings = get_generation_settings(metadata) |
| 1430 | _apply_story_profile_language(profile, str(settings.get("language") or "")) |
| 1431 | self._write_json(paths["story_profile"], profile) |
| 1432 | if not paths["state"].exists(): |
| 1433 | self._save_state(work_id, self._default_state(work_id, title=title, goal=goal)) |
| 1434 | state = self._load_state(work_id) |
| 1435 | changed = False |
| 1436 | if not is_reference_image_locked(state): |
| 1437 | ref = self._read_reference_image_from_session() |
| 1438 | if ref and state.get("reference_image") != ref: |
| 1439 | state["reference_image"] = ref |
| 1440 | changed = True |
| 1441 | auto = self._session_auto_generate() |
| 1442 | if auto and not bool(state.get("auto_generate")): |
| 1443 | state["auto_generate"] = True |
| 1444 | changed = True |
| 1445 | if changed: |
| 1446 | self._save_state(work_id, state) |
| 1447 | self._refresh_fact(work_id, state) |
| 1448 | return state |
| 1449 | |
| 1450 | def _is_unfinished(self, state: dict[str, Any]) -> bool: |
| 1451 | stage = str(state.get("stage") or self._DEFAULT_STAGE) |
| 1452 | return stage not in self._FINAL_STAGES |
| 1453 | |
| 1454 | def _shot_entries(self, state: dict[str, Any]) -> list[dict[str, Any]]: |
| 1455 | shots = state.get("shots", {}) |
| 1456 | if not isinstance(shots, dict): |
| 1457 | return [] |
| 1458 | items = [] |
| 1459 | for shot_key, payload in shots.items(): |
| 1460 | if isinstance(payload, dict): |
| 1461 | items.append({"shot_key": shot_key, **payload}) |
| 1462 | return sorted(items, key=lambda item: int(item.get("shot_id", 0))) |
| 1463 | |
| 1464 | def _status_counts(self, state: dict[str, Any]) -> dict[str, int]: |
| 1465 | counts: dict[str, int] = {} |
| 1466 | for item in self._shot_entries(state): |
| 1467 | status = str(item.get("status") or "planned") |
| 1468 | counts[status] = counts.get(status, 0) + 1 |
| 1469 | return counts |
| 1470 | |
| 1471 | def _pending_remote_jobs(self, state: dict[str, Any]) -> dict[str, Any]: |
| 1472 | pending = state.setdefault("pending_remote_jobs", {}) |
| 1473 | if not isinstance(pending, dict): |
| 1474 | pending = {} |
| 1475 | state["pending_remote_jobs"] = pending |
| 1476 | return pending |
| 1477 | |
| 1478 | def _register_pending_remote_job(self, state: dict[str, Any], job: dict[str, Any]) -> None: |
| 1479 | if job.get("status") != "queued": |
| 1480 | return |
| 1481 | pending = self._pending_remote_jobs(state) |
| 1482 | pending[str(job["job_id"])] = { |
| 1483 | "kind": job.get("kind"), |
| 1484 | "target": job.get("target"), |
| 1485 | "created_at": job.get("created_at"), |
| 1486 | } |
| 1487 | |
| 1488 | def _clear_pending_remote_job(self, state: dict[str, Any], job_id: str) -> None: |
| 1489 | pending = self._pending_remote_jobs(state) |
| 1490 | pending.pop(job_id, None) |
| 1491 | |
| 1492 | def _clear_pending_remote_jobs_for_target( |
| 1493 | self, |
| 1494 | state: dict[str, Any], |
| 1495 | kind: str, |
| 1496 | target: str | list[str], |
| 1497 | ) -> None: |
| 1498 | pending = self._pending_remote_jobs(state) |
| 1499 | target_key = json.dumps(target, ensure_ascii=False, sort_keys=True) |
| 1500 | stale_job_ids = [ |
| 1501 | job_id |
| 1502 | for job_id, item in pending.items() |
| 1503 | if isinstance(item, dict) |
| 1504 | and item.get("kind") == kind |
| 1505 | and json.dumps(item.get("target"), ensure_ascii=False, sort_keys=True) == target_key |
| 1506 | ] |
| 1507 | for job_id in stale_job_ids: |
| 1508 | pending.pop(job_id, None) |
| 1509 | |
| 1510 | def _clear_pending_remote_jobs_for_shot( |
| 1511 | self, |
| 1512 | state: dict[str, Any], |
| 1513 | shot_id: int, |
| 1514 | *, |
| 1515 | kinds: set[str] | None = None, |
| 1516 | ) -> None: |
| 1517 | pending = self._pending_remote_jobs(state) |
| 1518 | shot_key = _shot_key(shot_id) |
| 1519 | stale_job_ids = [] |
| 1520 | for job_id, item in pending.items(): |
| 1521 | if not isinstance(item, dict): |
| 1522 | continue |
| 1523 | if kinds is not None and str(item.get("kind")) not in kinds: |
| 1524 | continue |
| 1525 | target = item.get("target") |
| 1526 | if target == shot_key or ( |
| 1527 | isinstance(target, list) and any(str(value) == shot_key for value in target) |
| 1528 | ): |
| 1529 | stale_job_ids.append(job_id) |
| 1530 | for job_id in stale_job_ids: |
| 1531 | pending.pop(job_id, None) |
| 1532 | |
| 1533 | @staticmethod |
| 1534 | def _final_output_is_playable(locator: str) -> bool: |
| 1535 | raw = locator.strip() |
| 1536 | if not raw: |
| 1537 | return False |
| 1538 | parsed = urlparse(raw) |
| 1539 | if parsed.scheme in {"http", "https"}: |
| 1540 | return True |
| 1541 | suffix = Path(unquote(parsed.path or raw)).suffix.lower() |
| 1542 | return suffix in {".mp4", ".webm", ".mov", ".m4v", ".mkv"} |
| 1543 | |
| 1544 | def _sync_stage_from_state(self, state: dict[str, Any]) -> None: |
| 1545 | final_output = state.get("final_output_url") or state.get("final_output_path") |
| 1546 | if final_output and self._final_output_is_playable(str(final_output)): |
| 1547 | state["stage"] = "done" |
| 1548 | return |
| 1549 | if final_output: |
| 1550 | state["stage"] = "merging" |
| 1551 | return |
| 1552 | pending_remote_jobs = self._pending_remote_jobs(state) |
| 1553 | pending_kinds = { |
| 1554 | str(item.get("kind")) for item in pending_remote_jobs.values() if isinstance(item, dict) |
| 1555 | } |
| 1556 | if "merge_shot" in pending_kinds: |
| 1557 | state["stage"] = "merging" |
| 1558 | return |
| 1559 | if "generate_echo_shot" in pending_kinds: |
| 1560 | state["stage"] = "shot_generating" |
| 1561 | return |
| 1562 | current_stage = str(state.get("stage") or "") |
| 1563 | # Keep Memory review / generate-fail on 03 even if shot rows look idle. |
| 1564 | if current_stage in {"awaiting_memory_review", "failed"}: |
| 1565 | return |
| 1566 | shots = self._shot_entries(state) |
| 1567 | if any(item.get("status") in {"review_fail", "error"} for item in shots): |
| 1568 | state["stage"] = "shot_revising" |
| 1569 | return |
| 1570 | if any(item.get("status") in {"generated", "review_pass", "approved"} for item in shots): |
| 1571 | state["stage"] = "shot_reviewing" |
| 1572 | return |
| 1573 | if any(item.get("status") == "queued" for item in shots): |
| 1574 | state["stage"] = "shot_generating" |
| 1575 | return |
| 1576 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 1577 | try: |
| 1578 | shot_count = int(goal.get("shot_count") or 0) |
| 1579 | except (TypeError, ValueError): |
| 1580 | shot_count = 0 |
| 1581 | if current_stage in { |
| 1582 | "shot_generating", |
| 1583 | "shot_reviewing", |
| 1584 | "shot_revising", |
| 1585 | "merging", |
| 1586 | }: |
| 1587 | return |
| 1588 | # 02 分镜脚本 only after workplace confirm_story (「下一步」). |
| 1589 | # Chat set_director_goal / early shot files must not skip 策划剧本. |
| 1590 | if shot_count <= 0: |
| 1591 | if state.get("story_confirmed"): |
| 1592 | state["stage"] = "story_confirmed" |
| 1593 | else: |
| 1594 | state["stage"] = self._DEFAULT_STAGE |
| 1595 | return |
| 1596 | if current_stage == "shot_planning": |
| 1597 | return |
| 1598 | if state.get("story_confirmed"): |
| 1599 | state["stage"] = "story_confirmed" |
| 1600 | return |
| 1601 | state["stage"] = self._DEFAULT_STAGE |
| 1602 | |
| 1603 | def _refresh_fact(self, work_id: str, state: dict[str, Any]) -> str: |
| 1604 | paths = self._paths(work_id) |
| 1605 | story_exists = ( |
| 1606 | paths["story"].exists() and paths["story"].read_text(encoding="utf-8").strip() != "" |
| 1607 | ) |
| 1608 | story_profile = self._load_story_profile(work_id) |
| 1609 | goal = state.get("goal", {}) if isinstance(state.get("goal"), dict) else {} |
| 1610 | shot_items = self._shot_entries(state) |
| 1611 | counts = self._status_counts(state) |
| 1612 | pending_remote = self._pending_remote_jobs(state) |
| 1613 | lines = [ |
| 1614 | "# Director Fact", |
| 1615 | "", |
| 1616 | f"- work_id: `{work_id}`", |
| 1617 | f"- work_dir: `{paths['work_dir']}`", |
| 1618 | f"- stage: `{state.get('stage', self._DEFAULT_STAGE)}`", |
| 1619 | f"- story_confirmed: `{bool(state.get('story_confirmed'))}`", |
| 1620 | f"- story_exists: `{story_exists}`", |
| 1621 | f"- story_profile_exists: `{bool(story_profile)}`", |
| 1622 | f"- goal_brief: {state.get('goal_brief') or '(empty)'}", |
| 1623 | f"- reference_image_present: `{reference_image_present(state.get('reference_image'))}`", |
| 1624 | f"- reference_image_locked: `{is_reference_image_locked(state)}`", |
| 1625 | f"- auto_generate: `{bool(state.get('auto_generate'))}`", |
| 1626 | f"- auto_generate_shot_count: `{self._effective_auto_generate_shot_count(goal)}`", |
| 1627 | f"- reference_image_needs_story_rewrite: `{self._session_reference_needs_rewrite()}`", |
| 1628 | f"- reference_image_inject_failed: `{self._session_reference_inject_failed()}`", |
| 1629 | "", |
| 1630 | "## Goal", |
| 1631 | "", |
| 1632 | f"- shot_count: `{goal.get('shot_count')}`", |
| 1633 | f"- shot_duration_sec: `{goal.get('shot_duration_sec')}`", |
| 1634 | f"- generation_mode: `{goal.get('generation_mode', 'sequential')}`", |
| 1635 | "", |
| 1636 | "## Progress", |
| 1637 | "", |
| 1638 | f"- total_shots: `{len(shot_items)}`", |
| 1639 | f"- pending_remote_jobs: `{len(pending_remote)}`", |
| 1640 | ] |
| 1641 | if counts: |
| 1642 | for status, count in sorted(counts.items()): |
| 1643 | lines.append(f"- {status}: `{count}`") |
| 1644 | else: |
| 1645 | lines.append("- shot_statuses: `(none yet)`") |
| 1646 | lines += [ |
| 1647 | "", |
| 1648 | "## Paths", |
| 1649 | "", |
| 1650 | f"- state_json: `{paths['state']}`", |
| 1651 | f"- story_md: `{paths['story']}`", |
| 1652 | f"- story_profile_json: `{paths['story_profile']}`", |
| 1653 | f"- shots_dir: `{paths['shots']}`", |
| 1654 | f"- jobs_dir: `{paths['jobs']}`", |
| 1655 | f"- outputs_dir: `{paths['outputs']}`", |
| 1656 | "", |
| 1657 | "## Tool Ownership", |
| 1658 | "", |
| 1659 | "- Director state files are tool-owned. Use director tools instead of raw file edits whenever possible.", |
| 1660 | ] |
| 1661 | content = "\n".join(lines) |
| 1662 | self._write_text(paths["fact"], content) |
| 1663 | return content |
| 1664 | |
| 1665 | def _summary_from_shot(self, shot: dict[str, Any]) -> str: |
| 1666 | if isinstance(shot.get("summary"), str) and shot["summary"].strip(): |
| 1667 | return shot["summary"].strip() |
| 1668 | caption = shot.get("caption") |
| 1669 | if isinstance(caption, str) and caption.strip(): |
| 1670 | return caption.strip()[:160] |
| 1671 | prompt = shot.get("prompt") |
| 1672 | if isinstance(prompt, str) and prompt.strip(): |
| 1673 | return prompt.strip()[:160] |
| 1674 | return "" |
| 1675 | |
| 1676 | def _shot_artifact_locator(self, shot: dict[str, Any]) -> str | None: |
| 1677 | artifact_url = shot.get("artifact_url") |
| 1678 | if isinstance(artifact_url, str) and artifact_url.strip(): |
| 1679 | return artifact_url.strip() |
| 1680 | echo = shot.get("echo") |
| 1681 | if isinstance(echo, dict): |
| 1682 | result_url = echo.get("result_url") |
| 1683 | if isinstance(result_url, str) and result_url.strip(): |
| 1684 | return result_url.strip() |
| 1685 | artifact_path = shot.get("artifact_path") |
| 1686 | if isinstance(artifact_path, str) and artifact_path.strip(): |
| 1687 | return artifact_path.strip() |
| 1688 | remote_result = shot.get("remote_result") |
| 1689 | if isinstance(remote_result, dict): |
| 1690 | video_path = remote_result.get("video_path") |
| 1691 | if isinstance(video_path, str) and video_path.strip(): |
| 1692 | return video_path.strip() |
| 1693 | return None |
| 1694 | |
| 1695 | def _state_shot_entry(self, shot: dict[str, Any]) -> dict[str, Any]: |
| 1696 | return { |
| 1697 | "shot_id": int(shot["shot_id"]), |
| 1698 | "status": shot.get("status"), |
| 1699 | "summary": self._summary_from_shot(shot), |
| 1700 | "cut": bool(shot.get("cut", True)), |
| 1701 | "has_shot_spec": bool(shot.get("caption")), |
| 1702 | "has_artifact": bool(self._shot_artifact_locator(shot)), |
| 1703 | "artifact_path": shot.get("artifact_path"), |
| 1704 | "artifact_url": shot.get("artifact_url"), |
| 1705 | "last_review": shot.get("last_review"), |
| 1706 | "review_notes": shot.get("review_notes") or "", |
| 1707 | "generation_error": shot.get("generation_error") or "", |
| 1708 | "updated_at": _now_iso(), |
| 1709 | } |
| 1710 | |
| 1711 | def _mark_shot_generation_error( |
| 1712 | self, |
| 1713 | work_id: str, |
| 1714 | shot_id: int, |
| 1715 | *, |
| 1716 | error_message: str, |
| 1717 | job_id: str | None = None, |
| 1718 | ) -> dict[str, Any]: |
| 1719 | shot = self._load_shot(work_id, shot_id) |
| 1720 | if not shot: |
| 1721 | raise ValueError(f"Shot {shot_id} does not exist in work {work_id}.") |
| 1722 | shot["status"] = "error" |
| 1723 | shot["generation_error"] = error_message |
| 1724 | shot["last_review"] = "error" |
| 1725 | if job_id: |
| 1726 | shot["last_job_id"] = job_id |
| 1727 | echo = shot.get("echo") |
| 1728 | if isinstance(echo, dict): |
| 1729 | echo.update( |
| 1730 | { |
| 1731 | "status": "failed", |
| 1732 | "last_error": error_message, |
| 1733 | } |
| 1734 | ) |
| 1735 | self._save_shot(work_id, shot_id, shot) |
| 1736 | return shot |
| 1737 | |
| 1738 | @staticmethod |
| 1739 | def _review_history(shot: dict[str, Any]) -> list[dict[str, Any]]: |
| 1740 | history = shot.get("review_history") |
| 1741 | if isinstance(history, list): |
| 1742 | return [item for item in history if isinstance(item, dict)] |
| 1743 | return [] |
| 1744 | |
| 1745 | @staticmethod |
| 1746 | def _is_revised_prompt_update(shot: dict[str, Any]) -> bool: |
| 1747 | if str(shot.get("status") or "") == "review_fail": |
| 1748 | return True |
| 1749 | if shot.get("last_review") == "revise": |
| 1750 | return True |
| 1751 | return any(item.get("verdict") == "revise" for item in DirectorTool._review_history(shot)) |
| 1752 | |
| 1753 | def _append_review_history( |
| 1754 | self, |
| 1755 | shot: dict[str, Any], |
| 1756 | *, |
| 1757 | verdict: str, |
| 1758 | review_source: str, |
| 1759 | feedback: str | None, |
| 1760 | ) -> None: |
| 1761 | history = self._review_history(shot) |
| 1762 | history.append( |
| 1763 | { |
| 1764 | "verdict": verdict, |
| 1765 | "source": review_source, |
| 1766 | "feedback": feedback or "", |
| 1767 | "created_at": _now_iso(), |
| 1768 | } |
| 1769 | ) |
| 1770 | shot["review_history"] = history |
| 1771 | |
| 1772 | def _apply_shot_review( |
| 1773 | self, |
| 1774 | work_id: str, |
| 1775 | shot_id: int, |
| 1776 | *, |
| 1777 | verdict: str, |
| 1778 | review_source: str, |
| 1779 | feedback: str | None, |
| 1780 | ) -> tuple[dict[str, Any], dict[str, Any]]: |
| 1781 | shot = self._load_shot(work_id, shot_id) |
| 1782 | if not shot: |
| 1783 | raise ValueError(f"Shot {shot_id} does not exist in work {work_id}.") |
| 1784 | |
| 1785 | normalized_feedback = ( |
| 1786 | feedback.strip() if isinstance(feedback, str) and feedback.strip() else None |
| 1787 | ) |
| 1788 | if verdict == "revise" and not normalized_feedback: |
| 1789 | raise ValueError("feedback is required when verdict='revise'.") |
| 1790 | |
| 1791 | if verdict == "accept": |
| 1792 | shot["status"] = "approved" |
| 1793 | shot["last_review"] = "accepted" |
| 1794 | shot["review_notes"] = "" |
| 1795 | shot["approved_at"] = _now_iso() |
| 1796 | else: |
| 1797 | shot["status"] = "review_fail" |
| 1798 | shot["last_review"] = "revise" |
| 1799 | shot["review_notes"] = normalized_feedback or "" |
| 1800 | |
| 1801 | self._append_review_history( |
| 1802 | shot, |
| 1803 | verdict=verdict, |
| 1804 | review_source=review_source, |
| 1805 | feedback=normalized_feedback, |
| 1806 | ) |
| 1807 | self._save_shot(work_id, shot_id, shot) |
| 1808 | |
| 1809 | state = self._load_state(work_id) |
| 1810 | if verdict == "revise": |
| 1811 | self._clear_pending_remote_jobs_for_shot( |
| 1812 | state, |
| 1813 | shot_id, |
| 1814 | kinds={"generate_echo_shot"}, |
| 1815 | ) |
| 1816 | state.pop("merge_confirmation_requested_at", None) |
| 1817 | shots = state.setdefault("shots", {}) |
| 1818 | if isinstance(shots, dict): |
| 1819 | shots[_shot_key(shot_id)] = self._state_shot_entry(shot) |
| 1820 | self._sync_stage_from_state(state) |
| 1821 | self._save_state(work_id, state) |
| 1822 | self._refresh_fact(work_id, state) |
| 1823 | return shot, state |
| 1824 | |
| 1825 | def _normalize_reference_shot_ids( |
| 1826 | self, |
| 1827 | shot_id: int, |
| 1828 | reference_shot_ids: list[int], |
| 1829 | *, |
| 1830 | cut: bool, |
| 1831 | ) -> list[int]: |
| 1832 | normalized_reference_ids = sorted({int(item) for item in reference_shot_ids}) |
| 1833 | if len(normalized_reference_ids) != len(reference_shot_ids): |
| 1834 | raise ValueError("reference_shot_ids must not contain duplicates.") |
| 1835 | if any(ref_id <= 0 for ref_id in normalized_reference_ids): |
| 1836 | raise ValueError("reference_shot_ids must contain positive shot IDs only.") |
| 1837 | if any(ref_id >= shot_id for ref_id in normalized_reference_ids): |
| 1838 | raise ValueError("reference_shot_ids must refer only to earlier shots.") |
| 1839 | if not cut and shot_id > 1 and (shot_id - 1) not in normalized_reference_ids: |
| 1840 | raise ValueError( |
| 1841 | "Shots with cut=false must include the immediately previous shot as a reference." |
| 1842 | ) |
| 1843 | return normalized_reference_ids |
| 1844 | |
| 1845 | def _build_echo_payload( |
| 1846 | self, |
| 1847 | work_id: str, |
| 1848 | shot_id: int, |
| 1849 | reference_shot_ids: list[int], |
| 1850 | selection_note: str | None, |
| 1851 | condition_image_url: str | None = None, |
| 1852 | i2v_prompt: str | None = None, |
| 1853 | ) -> tuple[dict[str, Any], dict[str, Any], list[int]]: |
| 1854 | shot = self._load_shot(work_id, shot_id) |
| 1855 | if not shot: |
| 1856 | raise ValueError(f"Shot {shot_id} does not exist in work {work_id}.") |
| 1857 | caption = shot.get("caption") |
| 1858 | if not isinstance(caption, str) or not caption.strip(): |
| 1859 | raise ValueError(f"Shot {shot_id} has no caption yet. Call create_shot_prompt first.") |
| 1860 | |
| 1861 | normalized_reference_ids = self._normalize_reference_shot_ids( |
| 1862 | shot_id, |
| 1863 | reference_shot_ids, |
| 1864 | cut=bool(shot.get("cut", True)), |
| 1865 | ) |
| 1866 | |
| 1867 | reference_shots: list[dict[str, Any]] = [] |
| 1868 | for ref_id in normalized_reference_ids: |
| 1869 | reference_shot = self._load_shot(work_id, ref_id) |
| 1870 | if not reference_shot: |
| 1871 | raise ValueError(f"Reference shot {ref_id} does not exist in work {work_id}.") |
| 1872 | reference_shots.append( |
| 1873 | { |
| 1874 | "shot_id": ref_id, |
| 1875 | "shot_key": _shot_key(ref_id), |
| 1876 | "summary": self._summary_from_shot(reference_shot), |
| 1877 | "cut": bool(reference_shot.get("cut", True)), |
| 1878 | "artifact_url": reference_shot.get("artifact_url"), |
| 1879 | "artifact_path": reference_shot.get("artifact_path"), |
| 1880 | } |
| 1881 | ) |
| 1882 | |
| 1883 | state = self._load_state(work_id) |
| 1884 | story_profile = self._load_story_profile(work_id) |
| 1885 | duration_value = resolve_echo_duration_seconds(shot, state) |
| 1886 | num_frames = sync_shot_echo_duration(shot, duration_value) |
| 1887 | shot_payload: dict[str, Any] = { |
| 1888 | "shot_id": shot_id, |
| 1889 | "shot_key": _shot_key(shot_id), |
| 1890 | "cut": bool(shot.get("cut", True)), |
| 1891 | "summary": self._summary_from_shot(shot), |
| 1892 | "text": i2v_prompt.strip() if i2v_prompt else caption.strip(), |
| 1893 | "num_frames": num_frames, |
| 1894 | # Exact seconds matching num_frames (e.g. 177 → 7.08). A rounded |
| 1895 | # 7.0 would let the Echo service re-snap downward to 169 frames. |
| 1896 | "duration_sec": num_frames_to_exact_duration_sec(num_frames), |
| 1897 | } |
| 1898 | if condition_image_url: |
| 1899 | shot_payload["condition_image_url"] = condition_image_url |
| 1900 | shot_payload["generation_mode"] = "i2v" |
| 1901 | memory_slots = shot.get("approved_memory_slots") |
| 1902 | if isinstance(memory_slots, list) and memory_slots: |
| 1903 | shot_payload["memory_slots"] = memory_slots |
| 1904 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 1905 | width = goal.get("width") |
| 1906 | height = goal.get("height") |
| 1907 | if width is not None: |
| 1908 | shot_payload["width"] = int(width) |
| 1909 | if height is not None: |
| 1910 | shot_payload["height"] = int(height) |
| 1911 | payload = { |
| 1912 | "work_id": work_id, |
| 1913 | "shot": shot_payload, |
| 1914 | "reference_shot_ids": normalized_reference_ids, |
| 1915 | "reference_shots": reference_shots, |
| 1916 | "selection_note": selection_note.strip() |
| 1917 | if isinstance(selection_note, str) and selection_note.strip() |
| 1918 | else None, |
| 1919 | "story_context": { |
| 1920 | "latest_story_summary": state.get("latest_story_summary") or None, |
| 1921 | "story_profile_summary": story_profile.get("summary") |
| 1922 | if isinstance(story_profile, dict) |
| 1923 | else None, |
| 1924 | }, |
| 1925 | } |
| 1926 | return payload, shot, normalized_reference_ids |
| 1927 | |
| 1928 | def _build_r2v_payload( |
| 1929 | self, |
| 1930 | work_id: str, |
| 1931 | shot_id: int, |
| 1932 | *, |
| 1933 | prompt: str, |
| 1934 | num_frames: int, |
| 1935 | width: int | None = None, |
| 1936 | height: int | None = None, |
| 1937 | condition_image_url: str | None = None, |
| 1938 | memory_slots: list[dict[str, Any]] | None = None, |
| 1939 | ) -> dict[str, Any]: |
| 1940 | """Build request body for POST /r2v (unified T2V/I2V/R2V).""" |
| 1941 | payload: dict[str, Any] = { |
| 1942 | "work_id": work_id, |
| 1943 | "shot_id": _shot_key(shot_id), |
| 1944 | "prompt": prompt.strip(), |
| 1945 | "num_frames": num_frames, |
| 1946 | "memory_slots": memory_slots if isinstance(memory_slots, list) else [], |
| 1947 | } |
| 1948 | if condition_image_url: |
| 1949 | from nanobot.storage.files import outbound_file_url |
| 1950 | |
| 1951 | payload["condition_img"] = outbound_file_url( |
| 1952 | condition_image_url, |
| 1953 | workspace=self.workspace, |
| 1954 | work_id=work_id, |
| 1955 | name=f"request_assets/shot_{shot_id:03d}_condition.jpg", |
| 1956 | storage=self._tools_config.file_storage, |
| 1957 | ) |
| 1958 | if width is not None: |
| 1959 | payload["width"] = int(width) |
| 1960 | if height is not None: |
| 1961 | payload["height"] = int(height) |
| 1962 | return payload |
| 1963 | |
| 1964 | def _build_memory_slots( |
| 1965 | self, |
| 1966 | approved_memory_slots: Any, |
| 1967 | reference_shot_ids: list[int], |
| 1968 | *, |
| 1969 | work_id: str | None = None, |
| 1970 | ) -> list[dict[str, Any]]: |
| 1971 | """Resolve only human/auto-approved Memory slots for R2V. |
| 1972 | |
| 1973 | ``reference_shot_ids`` remains screenplay context and request metadata; |
| 1974 | it must never be converted into a media slot behind the user's back. |
| 1975 | """ |
| 1976 | slots: list[dict[str, Any]] = ( |
| 1977 | [dict(slot) for slot in approved_memory_slots if isinstance(slot, dict)] |
| 1978 | if isinstance(approved_memory_slots, list) |
| 1979 | else [] |
| 1980 | ) |
| 1981 | from nanobot.storage.files import outbound_file_url |
| 1982 | |
| 1983 | resolved_work_id = work_id or self._active_work_id() or "work" |
| 1984 | for index, slot in enumerate(slots): |
| 1985 | for key in ("image_url", "audio_url"): |
| 1986 | value = slot.get(key) |
| 1987 | if isinstance(value, str) and value.strip(): |
| 1988 | suffix = Path(value.split("?", 1)[0]).suffix |
| 1989 | if not suffix: |
| 1990 | suffix = ".jpg" if key == "image_url" else ".wav" |
| 1991 | slot[key] = outbound_file_url( |
| 1992 | value, |
| 1993 | workspace=self.workspace, |
| 1994 | work_id=resolved_work_id, |
| 1995 | name=f"request_assets/slot_{index:02d}_{key}{suffix}", |
| 1996 | storage=self._tools_config.file_storage, |
| 1997 | ) |
| 1998 | if len(slots) > 7: |
| 1999 | raise ValueError("approved memory slots cannot exceed 7") |
| 2000 | return slots |
| 2001 | |
| 2002 | def _build_merge_payload( |
| 2003 | self, |
| 2004 | work_id: str, |
| 2005 | shot_ids: list[int], |
| 2006 | selected_shots: list[dict[str, Any]], |
| 2007 | ) -> dict[str, Any]: |
| 2008 | from nanobot.storage.files import LocalFilePublisher, resolve_local_asset_path |
| 2009 | |
| 2010 | inputs: list[dict[str, Any]] = [] |
| 2011 | for shot in selected_shots: |
| 2012 | shot_id = int(shot["shot_id"]) |
| 2013 | echo = shot.get("echo") |
| 2014 | version_id = echo.get("version_id") if isinstance(echo, dict) else None |
| 2015 | item: dict[str, Any] = {"shot_id": shot_id} |
| 2016 | source = None |
| 2017 | if isinstance(echo, dict): |
| 2018 | source = echo.get("result_url") or echo.get("base_result_url") |
| 2019 | source = source or shot.get("artifact_url") or shot.get("artifact_path") |
| 2020 | if isinstance(source, str) and source.strip(): |
| 2021 | source = source.strip() |
| 2022 | if source.startswith(("http://", "https://")): |
| 2023 | item["video_url"] = source |
| 2024 | else: |
| 2025 | local_config = self._tools_config.file_storage.local |
| 2026 | local_path = resolve_local_asset_path( |
| 2027 | source, |
| 2028 | workspace=self.workspace, |
| 2029 | config=local_config, |
| 2030 | ) or Path(source).expanduser() |
| 2031 | if not str(local_config.base_url).strip(): |
| 2032 | raise ValueError( |
| 2033 | "tools.fileStorage.local.baseUrl is required to merge a local video." |
| 2034 | ) |
| 2035 | else: |
| 2036 | item["video_url"] = LocalFilePublisher( |
| 2037 | local_config, |
| 2038 | workspace=self.workspace, |
| 2039 | work_id=work_id, |
| 2040 | )(str(local_path), f"merge_inputs/shot_{shot_id:03d}.mp4") |
| 2041 | elif isinstance(version_id, str) and version_id.strip(): |
| 2042 | # Version records are process-local on the public Echo server and |
| 2043 | # disappear after a restart. Use them only when no durable artifact |
| 2044 | # locator was saved with the completed shot. |
| 2045 | item["version_id"] = version_id.strip() |
| 2046 | else: |
| 2047 | raise ValueError( |
| 2048 | f"Shot {shot_id} has no Echo version or playable video artifact." |
| 2049 | ) |
| 2050 | inputs.append(item) |
| 2051 | return { |
| 2052 | "work_id": work_id, |
| 2053 | "shot_ids": shot_ids, |
| 2054 | "shots": inputs, |
| 2055 | } |
| 2056 | |
| 2057 | def _submit_remote_request( |
| 2058 | self, |
| 2059 | work_id: str, |
| 2060 | job_id: str, |
| 2061 | request_payload: dict[str, Any], |
| 2062 | *, |
| 2063 | target: str | list[str], |
| 2064 | operation: str, |
| 2065 | ) -> dict[str, Any]: |
| 2066 | payload_path = self._paths(work_id)["outputs"] / f"{job_id}.payload.json" |
| 2067 | envelope = self._build_remote_request_envelope( |
| 2068 | operation, |
| 2069 | work_id, |
| 2070 | job_id, |
| 2071 | target, |
| 2072 | request_payload, |
| 2073 | ) |
| 2074 | request_path = self._paths(work_id)["outputs"] / f"{job_id}.request.json" |
| 2075 | self._write_json(payload_path, request_payload) |
| 2076 | self._write_json(request_path, envelope) |
| 2077 | |
| 2078 | job: dict[str, Any] = { |
| 2079 | "job_id": job_id, |
| 2080 | "kind": operation, |
| 2081 | "status": "queued", |
| 2082 | "work_id": work_id, |
| 2083 | "target": target, |
| 2084 | "created_at": _now_iso(), |
| 2085 | "completed_at": None, |
| 2086 | "request_payload_path": str(payload_path), |
| 2087 | "request_envelope_path": str(request_path), |
| 2088 | "remote": { |
| 2089 | "transport": "http", |
| 2090 | "protocol_version": _REMOTE_PROTOCOL_VERSION, |
| 2091 | "endpoint_path": self._remote_endpoint_path(operation), |
| 2092 | "remote_task_id": None, |
| 2093 | "callback_expected": True, |
| 2094 | "callback_contract": envelope["callback"], |
| 2095 | }, |
| 2096 | } |
| 2097 | |
| 2098 | callback_url = envelope.get("callback", {}).get("url") |
| 2099 | if not isinstance(callback_url, str) or not callback_url.strip(): |
| 2100 | raise RuntimeError( |
| 2101 | "Echo callback URL is not configured. " |
| 2102 | "Set tools.echoGenerator.callbackBaseUrl to the local Agent URL." |
| 2103 | ) |
| 2104 | |
| 2105 | EchoAdmissionController.from_tools_config(self._tools_config).ensure_allowed( |
| 2106 | operation=operation, |
| 2107 | ) |
| 2108 | |
| 2109 | base_url = self._remote_http_base_url() |
| 2110 | if not base_url: |
| 2111 | raise RuntimeError( |
| 2112 | "No Echo generator base URL configured. " |
| 2113 | "Set tools.echoGenerator.baseUrl for real HTTP submission." |
| 2114 | ) |
| 2115 | endpoint_url = f"{base_url}{self._remote_endpoint_path(operation)}" |
| 2116 | remote_ack = self._post_remote_http_request(endpoint_url, envelope) |
| 2117 | remote_task_id = remote_ack.get("remote_task_id") or remote_ack.get("task_id") |
| 2118 | job["remote"].update( |
| 2119 | { |
| 2120 | "endpoint_url": endpoint_url, |
| 2121 | "remote_task_id": remote_task_id, |
| 2122 | "version_id": remote_ack.get("version_id"), |
| 2123 | "status_url": remote_ack.get("status_url"), |
| 2124 | "ack": remote_ack, |
| 2125 | } |
| 2126 | ) |
| 2127 | return job |
| 2128 | |
| 2129 | def _submit_r2v_request( |
| 2130 | self, |
| 2131 | work_id: str, |
| 2132 | job_id: str, |
| 2133 | request_payload: dict[str, Any], |
| 2134 | *, |
| 2135 | target: str, |
| 2136 | ) -> dict[str, Any]: |
| 2137 | """Submit a generation job via POST /r2v (no director envelope).""" |
| 2138 | payload_path = self._paths(work_id)["outputs"] / f"{job_id}.payload.json" |
| 2139 | request_path = self._paths(work_id)["outputs"] / f"{job_id}.request.json" |
| 2140 | session = { |
| 2141 | "session_key": self._session_key.get() if self._session_key.get() else None, |
| 2142 | "channel": self._channel.get() if self._channel.get() else None, |
| 2143 | "chat_id": self._chat_id.get() if self._chat_id.get() else None, |
| 2144 | } |
| 2145 | callback_url = self._remote_callback_url("generate_echo_shot") |
| 2146 | outbound_payload = dict(request_payload) |
| 2147 | outbound_payload.update( |
| 2148 | { |
| 2149 | "job_id": job_id, |
| 2150 | "callback_context": session, |
| 2151 | } |
| 2152 | ) |
| 2153 | if callback_url: |
| 2154 | outbound_payload["callback_url"] = callback_url |
| 2155 | self._write_json(payload_path, request_payload) |
| 2156 | self._write_json(request_path, outbound_payload) |
| 2157 | |
| 2158 | job: dict[str, Any] = { |
| 2159 | "job_id": job_id, |
| 2160 | "kind": "generate_echo_shot", |
| 2161 | "status": "queued", |
| 2162 | "work_id": work_id, |
| 2163 | "target": target, |
| 2164 | "created_at": _now_iso(), |
| 2165 | "completed_at": None, |
| 2166 | "request_payload_path": str(payload_path), |
| 2167 | "request_envelope_path": str(request_path), |
| 2168 | "remote": { |
| 2169 | "transport": "http", |
| 2170 | "endpoint_path": self._remote_endpoint_path("r2v_generate"), |
| 2171 | "remote_task_id": None, |
| 2172 | "r2v": True, |
| 2173 | "callback_expected": True, |
| 2174 | "callback_url": callback_url, |
| 2175 | }, |
| 2176 | "session": session, |
| 2177 | } |
| 2178 | if not callback_url: |
| 2179 | raise RuntimeError( |
| 2180 | "Echo callback URL is not configured. " |
| 2181 | "Set tools.echoGenerator.callbackBaseUrl to the local Agent URL." |
| 2182 | ) |
| 2183 | |
| 2184 | EchoAdmissionController.from_tools_config(self._tools_config).ensure_allowed( |
| 2185 | operation="r2v_generate", |
| 2186 | ) |
| 2187 | |
| 2188 | base_url = self._remote_http_base_url() |
| 2189 | if not base_url: |
| 2190 | raise RuntimeError( |
| 2191 | "No Echo generator base URL configured. " |
| 2192 | "Set tools.echoGenerator.baseUrl in config." |
| 2193 | ) |
| 2194 | endpoint_url = f"{base_url}{self._remote_endpoint_path('r2v_generate')}" |
| 2195 | |
| 2196 | headers: dict[str, str] = { |
| 2197 | "Content-Type": "application/json", |
| 2198 | "Accept": "application/json", |
| 2199 | } |
| 2200 | |
| 2201 | body = json.dumps(outbound_payload, ensure_ascii=False).encode("utf-8") |
| 2202 | # This request only submits the job. Rendering completes through the |
| 2203 | # callback, so it must not inherit a generation-length timeout. |
| 2204 | timeout = self._remote_http_timeout_sec() |
| 2205 | last_exc: Exception | None = None |
| 2206 | raw = "" |
| 2207 | for attempt in range(1, _R2V_SUBMIT_ATTEMPTS + 1): |
| 2208 | req = urllib_request.Request( |
| 2209 | endpoint_url, data=body, headers=headers, method="POST" |
| 2210 | ) |
| 2211 | try: |
| 2212 | with urllib_request.urlopen(req, timeout=timeout) as resp: |
| 2213 | raw = resp.read().decode("utf-8") |
| 2214 | last_exc = None |
| 2215 | break |
| 2216 | except urllib_error.HTTPError as exc: |
| 2217 | last_exc = exc |
| 2218 | transient = exc.code in _R2V_TRANSIENT_HTTP_CODES |
| 2219 | try: |
| 2220 | response_detail = exc.read(4096).decode("utf-8", errors="replace").strip() |
| 2221 | except Exception: # noqa: BLE001 - HTTPError may have no response stream. |
| 2222 | response_detail = "" |
| 2223 | logger.error( |
| 2224 | "R2V submit HTTP {} work_id={} job_id={} attempt={}/{} " |
| 2225 | "transient={} error={} response={}", |
| 2226 | exc.code, |
| 2227 | work_id, |
| 2228 | job_id, |
| 2229 | attempt, |
| 2230 | _R2V_SUBMIT_ATTEMPTS, |
| 2231 | transient, |
| 2232 | exc, |
| 2233 | response_detail or "-", |
| 2234 | ) |
| 2235 | if not transient or attempt >= _R2V_SUBMIT_ATTEMPTS: |
| 2236 | suffix = f"; response: {response_detail}" if response_detail else "" |
| 2237 | raise RuntimeError( |
| 2238 | f"R2V submit failed with HTTP {exc.code}: {exc}{suffix}" |
| 2239 | ) from exc |
| 2240 | time.sleep(min(2 * attempt, 6)) |
| 2241 | except (urllib_error.URLError, TimeoutError, OSError) as exc: |
| 2242 | last_exc = exc |
| 2243 | if is_connection_refused(exc): |
| 2244 | logger.error( |
| 2245 | "R2V submit unreachable work_id={} job_id={} error={}", |
| 2246 | work_id, |
| 2247 | job_id, |
| 2248 | exc, |
| 2249 | ) |
| 2250 | raise EchoGeneratorUnavailableError(UNAVAILABLE_MESSAGE) from exc |
| 2251 | logger.error( |
| 2252 | "R2V submit failed work_id={} job_id={} attempt={}/{} " |
| 2253 | "retryable={} timeout_s={} error={}", |
| 2254 | work_id, |
| 2255 | job_id, |
| 2256 | attempt, |
| 2257 | _R2V_SUBMIT_ATTEMPTS, |
| 2258 | True, |
| 2259 | timeout, |
| 2260 | exc, |
| 2261 | ) |
| 2262 | if attempt >= _R2V_SUBMIT_ATTEMPTS: |
| 2263 | raise RuntimeError(f"R2V submit failed: {exc}") from exc |
| 2264 | time.sleep(min(2 * attempt, 6)) |
| 2265 | if last_exc is not None: |
| 2266 | raise RuntimeError(f"R2V submit failed: {last_exc}") from last_exc |
| 2267 | |
| 2268 | try: |
| 2269 | remote_ack = json.loads(raw) if raw.strip() else {} |
| 2270 | except (json.JSONDecodeError, UnicodeDecodeError) as exc: |
| 2271 | raise RuntimeError(f"R2V submit returned invalid JSON: {exc}") from exc |
| 2272 | |
| 2273 | if not isinstance(remote_ack, dict): |
| 2274 | raise RuntimeError( |
| 2275 | f"R2V submit returned unexpected payload type: {type(remote_ack).__name__}" |
| 2276 | ) |
| 2277 | |
| 2278 | version_id = remote_ack.get("version_id") |
| 2279 | if not isinstance(version_id, str) or not version_id.strip(): |
| 2280 | raise RuntimeError( |
| 2281 | "R2V submit response is missing a non-empty 'version_id'" |
| 2282 | ) |
| 2283 | |
| 2284 | task_id = remote_ack.get("task_id") |
| 2285 | status_url = remote_ack.get("status_url") or f"/version/{version_id}" |
| 2286 | job["remote"].update( |
| 2287 | { |
| 2288 | "endpoint_url": endpoint_url, |
| 2289 | "remote_task_id": task_id, |
| 2290 | "version_id": version_id, |
| 2291 | "status_url": status_url, |
| 2292 | "ack": remote_ack, |
| 2293 | } |
| 2294 | ) |
| 2295 | return job |
| 2296 | |
| 2297 | @staticmethod |
| 2298 | def _target_shot_ids(target: Any) -> list[int]: |
| 2299 | if isinstance(target, list): |
| 2300 | shot_ids: list[int] = [] |
| 2301 | for item in target: |
| 2302 | if isinstance(item, str): |
| 2303 | shot_ids.append(_shot_id_from_key(item)) |
| 2304 | elif isinstance(item, int): |
| 2305 | shot_ids.append(item) |
| 2306 | return shot_ids |
| 2307 | if isinstance(target, str): |
| 2308 | return [_shot_id_from_key(target)] |
| 2309 | if isinstance(target, int): |
| 2310 | return [target] |
| 2311 | return [] |
| 2312 | |
| 2313 | @staticmethod |
| 2314 | def _extract_remote_stored_locator(*sources: dict[str, Any]) -> str | None: |
| 2315 | """Resolve the canonical stored locator from a remote callback payload. |
| 2316 | |
| 2317 | Preference order: configured ``asset_urls`` entries, then direct HTTP-style URLs. |
| 2318 | PFS/local paths are ignored because the browser cannot play them. |
| 2319 | """ |
| 2320 | from nanobot.integrations.remote_video_url import resolve_public_video_url |
| 2321 | |
| 2322 | return resolve_public_video_url(*sources) |
| 2323 | |
| 2324 | def _normalize_echo_callback_result( |
| 2325 | self, |
| 2326 | callback_payload: dict[str, Any], |
| 2327 | job: dict[str, Any], |
| 2328 | ) -> dict[str, Any]: |
| 2329 | default_shot_ids = self._target_shot_ids(job.get("target")) |
| 2330 | default_shot_id = default_shot_ids[0] if default_shot_ids else None |
| 2331 | raw_result = callback_payload.get("result") |
| 2332 | sources: list[dict[str, Any]] = [callback_payload] |
| 2333 | if isinstance(raw_result, dict): |
| 2334 | sources.insert(0, raw_result) |
| 2335 | shot_id = raw_result.get("shot_id", default_shot_id) |
| 2336 | else: |
| 2337 | shot_id = callback_payload.get("shot_id", default_shot_id) |
| 2338 | result_url = self._extract_remote_stored_locator(*sources) |
| 2339 | if shot_id is None or not result_url: |
| 2340 | raise ValueError( |
| 2341 | "Echo callback requires shot_id plus a public asset_urls URL or result_url, " |
| 2342 | "either at the top level or inside result={...}." |
| 2343 | ) |
| 2344 | normalized_shot_id = ( |
| 2345 | _shot_id_from_key(shot_id) |
| 2346 | if isinstance(shot_id, str) and shot_id.startswith("shot_") |
| 2347 | else int(shot_id) |
| 2348 | ) |
| 2349 | normalized: dict[str, Any] = { |
| 2350 | "shot_id": normalized_shot_id, |
| 2351 | "result_url": result_url, |
| 2352 | } |
| 2353 | for key in ("video_id",): |
| 2354 | value = next((source.get(key) for source in sources if source.get(key)), None) |
| 2355 | if value is not None: |
| 2356 | normalized[key] = value |
| 2357 | return normalized |
| 2358 | |
| 2359 | def apply_echo_callback_payload(self, callback_payload: dict[str, Any]) -> dict[str, Any]: |
| 2360 | work_id = callback_payload.get("work_id") |
| 2361 | job_id = callback_payload.get("job_id") |
| 2362 | if not isinstance(work_id, str) or not work_id.strip(): |
| 2363 | raise ValueError("Echo callback missing work_id.") |
| 2364 | if not isinstance(job_id, str) or not job_id.strip(): |
| 2365 | raise ValueError("Echo callback missing job_id.") |
| 2366 | |
| 2367 | job = self._load_job(work_id, job_id) |
| 2368 | if not job: |
| 2369 | raise ValueError(f"Director job '{job_id}' was not found for work '{work_id}'.") |
| 2370 | if job.get("kind") != "generate_echo_shot": |
| 2371 | raise ValueError(f"Director job '{job_id}' is not a generate_echo_shot job.") |
| 2372 | |
| 2373 | existing_status = str(job.get("status") or "") |
| 2374 | if existing_status in {"completed", "failed"}: |
| 2375 | result_url = job.get("result_url") |
| 2376 | return { |
| 2377 | "status": existing_status, |
| 2378 | "operation": "generate_echo_shot", |
| 2379 | "work_id": work_id, |
| 2380 | "job_id": job_id, |
| 2381 | "duplicate": True, |
| 2382 | "result_urls": [result_url] |
| 2383 | if isinstance(result_url, str) and result_url.strip() |
| 2384 | else [], |
| 2385 | "updated_shots": [], |
| 2386 | } |
| 2387 | |
| 2388 | status = str(callback_payload.get("status") or "completed") |
| 2389 | if status not in {"completed", "failed"}: |
| 2390 | raise ValueError("Echo callback status must be 'completed' or 'failed'.") |
| 2391 | |
| 2392 | state = self._load_state(work_id) |
| 2393 | result = ( |
| 2394 | self._normalize_echo_callback_result(callback_payload, job) |
| 2395 | if status == "completed" |
| 2396 | else None |
| 2397 | ) |
| 2398 | default_shot_ids = self._target_shot_ids(job.get("target")) |
| 2399 | target_shot_id = ( |
| 2400 | int(result["shot_id"]) |
| 2401 | if isinstance(result, dict) |
| 2402 | else (default_shot_ids[0] if default_shot_ids else None) |
| 2403 | ) |
| 2404 | if target_shot_id is None: |
| 2405 | raise ValueError( |
| 2406 | "Echo callback could not determine shot_id from callback or job target." |
| 2407 | ) |
| 2408 | |
| 2409 | shot = self._load_shot(work_id, target_shot_id) |
| 2410 | if not shot: |
| 2411 | raise ValueError(f"Shot {target_shot_id} does not exist in work {work_id}.") |
| 2412 | |
| 2413 | current_status = str(shot.get("status") or "") |
| 2414 | if current_status in {"prompt_ready", "revised_prompt_ready", "planned"}: |
| 2415 | job["status"] = status |
| 2416 | job["completed_at"] = callback_payload.get("completed_at") or _now_iso() |
| 2417 | remote = job.get("remote") |
| 2418 | if not isinstance(remote, dict): |
| 2419 | remote = {} |
| 2420 | job["remote"] = remote |
| 2421 | remote["callback_received_at"] = _now_iso() |
| 2422 | remote["ignored_stale_callback"] = True |
| 2423 | if callback_payload.get("remote_task_id"): |
| 2424 | remote["remote_task_id"] = callback_payload.get("remote_task_id") |
| 2425 | job["callback_payload"] = callback_payload |
| 2426 | self._save_job(work_id, job_id, job) |
| 2427 | self._clear_pending_remote_job(state, job_id) |
| 2428 | self._save_state(work_id, state) |
| 2429 | return { |
| 2430 | "status": "ignored", |
| 2431 | "operation": "generate_echo_shot", |
| 2432 | "work_id": work_id, |
| 2433 | "job_id": job_id, |
| 2434 | "shot_id": target_shot_id, |
| 2435 | "reason": f"shot is {current_status} after replan; callback ignored", |
| 2436 | } |
| 2437 | |
| 2438 | job["status"] = status |
| 2439 | job["completed_at"] = callback_payload.get("completed_at") or _now_iso() |
| 2440 | remote = job.get("remote") |
| 2441 | if not isinstance(remote, dict): |
| 2442 | remote = {} |
| 2443 | job["remote"] = remote |
| 2444 | remote["callback_received_at"] = _now_iso() |
| 2445 | if callback_payload.get("remote_task_id"): |
| 2446 | remote["remote_task_id"] = callback_payload.get("remote_task_id") |
| 2447 | job["callback_payload"] = callback_payload |
| 2448 | |
| 2449 | echo = shot.get("echo") |
| 2450 | if not isinstance(echo, dict): |
| 2451 | echo = {} |
| 2452 | shot["echo"] = echo |
| 2453 | |
| 2454 | updated_shots: list[dict[str, Any]] = [] |
| 2455 | result_urls: list[str] = [] |
| 2456 | if status == "completed" and isinstance(result, dict): |
| 2457 | result_url = result["result_url"] |
| 2458 | shot["status"] = "generated" |
| 2459 | shot.pop("generation_error", None) |
| 2460 | shot["artifact_url"] = result_url |
| 2461 | shot["last_job_id"] = job_id |
| 2462 | echo.update( |
| 2463 | { |
| 2464 | "status": "completed", |
| 2465 | "result_url": result_url, |
| 2466 | "completed_at": job.get("completed_at"), |
| 2467 | "callback_received_at": remote.get("callback_received_at"), |
| 2468 | "remote_task_id": remote.get("remote_task_id"), |
| 2469 | } |
| 2470 | ) |
| 2471 | for key in ("video_id",): |
| 2472 | if result.get(key) is not None: |
| 2473 | echo[key] = result[key] |
| 2474 | job["result_url"] = result_url |
| 2475 | updated_shots.append( |
| 2476 | { |
| 2477 | "shot_id": target_shot_id, |
| 2478 | "shot_key": _shot_key(target_shot_id), |
| 2479 | "result_url": result_url, |
| 2480 | } |
| 2481 | ) |
| 2482 | result_urls.append(result_url) |
| 2483 | else: |
| 2484 | error_message = callback_payload.get("error") or "Remote Echo generation failed." |
| 2485 | job["error"] = error_message |
| 2486 | shot = self._mark_shot_generation_error( |
| 2487 | work_id, |
| 2488 | target_shot_id, |
| 2489 | error_message=error_message, |
| 2490 | job_id=job_id, |
| 2491 | ) |
| 2492 | echo = shot.get("echo") |
| 2493 | if isinstance(echo, dict): |
| 2494 | echo.update( |
| 2495 | { |
| 2496 | "callback_received_at": remote.get("callback_received_at"), |
| 2497 | "remote_task_id": remote.get("remote_task_id"), |
| 2498 | } |
| 2499 | ) |
| 2500 | shot["echo"] = echo |
| 2501 | self._save_shot(work_id, target_shot_id, shot) |
| 2502 | |
| 2503 | if status == "completed": |
| 2504 | self._save_shot(work_id, target_shot_id, shot) |
| 2505 | shots = state.setdefault("shots", {}) |
| 2506 | if isinstance(shots, dict): |
| 2507 | shots[_shot_key(target_shot_id)] = self._state_shot_entry(shot) |
| 2508 | self._save_job(work_id, job_id, job) |
| 2509 | self._clear_pending_remote_job(state, job_id) |
| 2510 | self._sync_stage_from_state(state) |
| 2511 | self._save_state(work_id, state) |
| 2512 | self._refresh_fact(work_id, state) |
| 2513 | |
| 2514 | if status == "completed": |
| 2515 | message = prompts.text( |
| 2516 | "director.callback.generate_echo_shot.completed", |
| 2517 | work_id=work_id, |
| 2518 | shot_key=_shot_key(target_shot_id), |
| 2519 | result_url=result_urls[0], |
| 2520 | ) |
| 2521 | else: |
| 2522 | message = prompts.text( |
| 2523 | "director.callback.generate_echo_shot.failed", |
| 2524 | work_id=work_id, |
| 2525 | shot_key=_shot_key(target_shot_id), |
| 2526 | error=job.get("error"), |
| 2527 | ) |
| 2528 | return { |
| 2529 | "status": status, |
| 2530 | "operation": "generate_echo_shot", |
| 2531 | "work_id": work_id, |
| 2532 | "job_id": job_id, |
| 2533 | "result_urls": result_urls, |
| 2534 | "updated_shots": updated_shots, |
| 2535 | "injection_message": message, |
| 2536 | "session_key": callback_payload.get("session_key"), |
| 2537 | "channel": callback_payload.get("channel"), |
| 2538 | "chat_id": callback_payload.get("chat_id"), |
| 2539 | } |
| 2540 | |
| 2541 | @staticmethod |
| 2542 | def _normalize_merge_callback_result(callback_payload: dict[str, Any]) -> dict[str, str | None]: |
| 2543 | raw_result = callback_payload.get("result") |
| 2544 | sources: list[dict[str, Any]] = [callback_payload] |
| 2545 | if isinstance(raw_result, dict): |
| 2546 | sources.insert(0, raw_result) |
| 2547 | |
| 2548 | stored_locator = DirectorTool._extract_remote_stored_locator(*sources) |
| 2549 | |
| 2550 | artifact_path = None |
| 2551 | for source in sources: |
| 2552 | candidate = source.get("artifact_path") or source.get("output_path") |
| 2553 | if isinstance(candidate, str) and candidate.strip(): |
| 2554 | artifact_path = candidate.strip() |
| 2555 | break |
| 2556 | |
| 2557 | normalized_path = artifact_path |
| 2558 | normalized_url = stored_locator |
| 2559 | if not normalized_path and not normalized_url: |
| 2560 | raise ValueError( |
| 2561 | "Merge callback requires a public asset_urls URL, artifact_path, " |
| 2562 | "or artifact_url/result_url, " |
| 2563 | "either at the top level or inside result={...}." |
| 2564 | ) |
| 2565 | return { |
| 2566 | "artifact_path": normalized_path, |
| 2567 | "artifact_url": normalized_url, |
| 2568 | } |
| 2569 | |
| 2570 | def apply_merge_callback_payload(self, callback_payload: dict[str, Any]) -> dict[str, Any]: |
| 2571 | work_id = callback_payload.get("work_id") |
| 2572 | job_id = callback_payload.get("job_id") |
| 2573 | if not isinstance(work_id, str) or not work_id.strip(): |
| 2574 | raise ValueError("Merge callback missing work_id.") |
| 2575 | if not isinstance(job_id, str) or not job_id.strip(): |
| 2576 | raise ValueError("Merge callback missing job_id.") |
| 2577 | |
| 2578 | job = self._load_job(work_id, job_id) |
| 2579 | if not job: |
| 2580 | raise ValueError(f"Director job '{job_id}' was not found for work '{work_id}'.") |
| 2581 | if job.get("kind") != "merge_shot": |
| 2582 | raise ValueError(f"Director job '{job_id}' is not a merge_shot job.") |
| 2583 | |
| 2584 | existing_status = str(job.get("status") or "") |
| 2585 | if existing_status in {"completed", "failed"}: |
| 2586 | final_output = job.get("artifact_url") or job.get("artifact_path") |
| 2587 | return { |
| 2588 | "status": existing_status, |
| 2589 | "operation": "merge_shot", |
| 2590 | "work_id": work_id, |
| 2591 | "job_id": job_id, |
| 2592 | "duplicate": True, |
| 2593 | "final_output": final_output, |
| 2594 | "final_output_path": job.get("artifact_path"), |
| 2595 | "final_output_url": job.get("artifact_url"), |
| 2596 | "media": [final_output] |
| 2597 | if isinstance(final_output, str) and final_output.strip() |
| 2598 | else [], |
| 2599 | } |
| 2600 | |
| 2601 | status = str(callback_payload.get("status") or "completed") |
| 2602 | if status not in {"completed", "failed"}: |
| 2603 | raise ValueError("Merge callback status must be 'completed' or 'failed'.") |
| 2604 | |
| 2605 | state = self._load_state(work_id) |
| 2606 | result = ( |
| 2607 | self._normalize_merge_callback_result(callback_payload) |
| 2608 | if status == "completed" |
| 2609 | else None |
| 2610 | ) |
| 2611 | |
| 2612 | job["status"] = status |
| 2613 | job["completed_at"] = callback_payload.get("completed_at") or _now_iso() |
| 2614 | remote = job.get("remote") |
| 2615 | if not isinstance(remote, dict): |
| 2616 | remote = {} |
| 2617 | job["remote"] = remote |
| 2618 | remote["callback_received_at"] = _now_iso() |
| 2619 | if callback_payload.get("remote_task_id"): |
| 2620 | remote["remote_task_id"] = callback_payload.get("remote_task_id") |
| 2621 | job["callback_payload"] = callback_payload |
| 2622 | |
| 2623 | final_output_path = None |
| 2624 | final_output_url = None |
| 2625 | if status == "completed" and isinstance(result, dict): |
| 2626 | state.pop("generation_error", None) |
| 2627 | final_output_path = result["artifact_path"] |
| 2628 | final_output_url = result["artifact_url"] |
| 2629 | state["final_output_path"] = final_output_path |
| 2630 | state["final_output_url"] = final_output_url |
| 2631 | if final_output_path is not None: |
| 2632 | job["artifact_path"] = final_output_path |
| 2633 | if final_output_url is not None: |
| 2634 | job["artifact_url"] = final_output_url |
| 2635 | else: |
| 2636 | error_message = callback_payload.get("error") or "Remote merge failed." |
| 2637 | job["error"] = error_message |
| 2638 | state["generation_error"] = error_message |
| 2639 | state["stage"] = "failed" |
| 2640 | |
| 2641 | self._save_job(work_id, job_id, job) |
| 2642 | self._clear_pending_remote_job(state, job_id) |
| 2643 | self._sync_stage_from_state(state) |
| 2644 | self._save_state(work_id, state) |
| 2645 | self._refresh_fact(work_id, state) |
| 2646 | |
| 2647 | final_output = final_output_url or final_output_path |
| 2648 | if status == "completed": |
| 2649 | message = prompts.text( |
| 2650 | "director.callback.merge_shot.completed", |
| 2651 | work_id=work_id, |
| 2652 | final_output=final_output, |
| 2653 | ) |
| 2654 | else: |
| 2655 | message = prompts.text( |
| 2656 | "director.callback.merge_shot.failed", |
| 2657 | work_id=work_id, |
| 2658 | job_id=job_id, |
| 2659 | error=job.get("error"), |
| 2660 | ) |
| 2661 | return { |
| 2662 | "status": status, |
| 2663 | "operation": "merge_shot", |
| 2664 | "work_id": work_id, |
| 2665 | "job_id": job_id, |
| 2666 | "final_output": final_output, |
| 2667 | "final_output_path": final_output_path, |
| 2668 | "final_output_url": final_output_url, |
| 2669 | "media": [final_output] |
| 2670 | if isinstance(final_output, str) and final_output.strip() |
| 2671 | else [], |
| 2672 | "injection_message": message, |
| 2673 | "session_key": callback_payload.get("session_key"), |
| 2674 | "channel": callback_payload.get("channel"), |
| 2675 | "chat_id": callback_payload.get("chat_id"), |
| 2676 | } |
| 2677 | |
| 2678 | def _next_action(self, state: dict[str, Any]) -> str: |
| 2679 | if not state.get("story_confirmed"): |
| 2680 | return "" |
| 2681 | goal = state.get("goal", {}) if isinstance(state.get("goal"), dict) else {} |
| 2682 | if goal.get("shot_count") in (None, 0): |
| 2683 | return "" |
| 2684 | shot_count = int(goal.get("shot_count") or 0) |
| 2685 | shots = self._shot_entries(state) |
| 2686 | if len(shots) < shot_count: |
| 2687 | if self._session_auto_generate() or bool(state.get("auto_generate")): |
| 2688 | return "" |
| 2689 | return prompts.text("director.next_action.storyboard_ready") |
| 2690 | if any( |
| 2691 | item.get("status") in {"planned", "prompt_ready", "revised_prompt_ready", "queued"} |
| 2692 | for item in shots |
| 2693 | ): |
| 2694 | return prompts.text("director.next_action.start_generation") |
| 2695 | if state.get("review_completed_at"): |
| 2696 | return prompts.text("director.next_action.review_complete") |
| 2697 | if not state.get("final_output_path"): |
| 2698 | return prompts.text("director.next_action.merge_ready") |
| 2699 | return prompts.text("director.next_action.work_complete") |
| 2700 | |
| 2701 | def _confirm(self, work_id: str) -> dict[str, Any]: |
| 2702 | state = self._load_state(work_id) |
| 2703 | self._sync_stage_from_state(state) |
| 2704 | self._save_state(work_id, state) |
| 2705 | fact = self._refresh_fact(work_id, state) |
| 2706 | goal = state.get("goal", {}) if isinstance(state.get("goal"), dict) else {} |
| 2707 | shots = self._shot_entries(state) |
| 2708 | approved = sum(1 for item in shots if item.get("status") in {"review_pass", "approved"}) |
| 2709 | generated = sum(1 for item in shots if item.get("status") == "generated") |
| 2710 | return { |
| 2711 | "work_id": work_id, |
| 2712 | "stage": state.get("stage", self._DEFAULT_STAGE), |
| 2713 | "story_confirmed": bool(state.get("story_confirmed")), |
| 2714 | "goal": goal, |
| 2715 | "story_exists": self._paths(work_id)["story"].read_text(encoding="utf-8").strip() != "", |
| 2716 | "shot_total": len(shots), |
| 2717 | "shot_generated": generated, |
| 2718 | "shot_approved": approved, |
| 2719 | "next_recommended_action": self._next_action(state), |
| 2720 | "fact_md": fact, |
| 2721 | } |
| 2722 | |
| 2723 | |
| 2724 | @tool_parameters( |
| 2725 | tool_parameters_schema( |
| 2726 | goal=StringSchema("Brief description of the user's intended video or story work"), |
| 2727 | title=StringSchema("Optional short title for the work"), |
| 2728 | continue_policy=StringSchema( |
| 2729 | "How to handle an unfinished existing work: ask, resume, or new", |
| 2730 | enum=["ask", "resume", "new"], |
| 2731 | ), |
| 2732 | required=["goal"], |
| 2733 | ) |
| 2734 | ) |
| 2735 | class StartDirectorTool(DirectorTool): |
| 2736 | @property |
| 2737 | def name(self) -> str: |
| 2738 | return "start_director" |
| 2739 | |
| 2740 | async def execute( |
| 2741 | self, |
| 2742 | goal: str, |
| 2743 | title: str | None = None, |
| 2744 | continue_policy: str = "ask", |
| 2745 | **kwargs: Any, |
| 2746 | ) -> str: |
| 2747 | # Check current active work first, then scan history for unfinished works |
| 2748 | existing_work_id: str | None = None |
| 2749 | existing_state: dict[str, Any] | None = None |
| 2750 | |
| 2751 | active_id = self._active_work_id() |
| 2752 | if active_id: |
| 2753 | state = self._load_state(active_id) |
| 2754 | if state and self._is_unfinished(state): |
| 2755 | existing_work_id = active_id |
| 2756 | existing_state = state |
| 2757 | |
| 2758 | if not existing_work_id: |
| 2759 | for hist_id in reversed(self._session_work_history()): |
| 2760 | if hist_id == active_id: |
| 2761 | continue |
| 2762 | state = self._load_state(hist_id) |
| 2763 | if state and self._is_unfinished(state): |
| 2764 | existing_work_id = hist_id |
| 2765 | existing_state = state |
| 2766 | break |
| 2767 | |
| 2768 | if existing_work_id and continue_policy == "ask": |
| 2769 | return _json_dump( |
| 2770 | { |
| 2771 | "status": "needs_confirmation", |
| 2772 | "message": "An unfinished director work already exists for this session.", |
| 2773 | "existing_work_id": existing_work_id, |
| 2774 | "stage": existing_state.get("stage") if existing_state else None, |
| 2775 | "goal_brief": existing_state.get("goal_brief") if existing_state else None, |
| 2776 | "next_step": "Ask the user whether to continue the existing work or create a new one.", |
| 2777 | } |
| 2778 | ) |
| 2779 | if existing_work_id and continue_policy == "resume": |
| 2780 | self._set_active_work(existing_work_id) |
| 2781 | return _json_dump( |
| 2782 | { |
| 2783 | "status": "resumed", |
| 2784 | "work_id": existing_work_id, |
| 2785 | "stage": existing_state.get("stage") if existing_state else None, |
| 2786 | "goal_brief": existing_state.get("goal_brief") if existing_state else None, |
| 2787 | } |
| 2788 | ) |
| 2789 | |
| 2790 | slug_source = title or goal[:48] |
| 2791 | stamp = datetime.now(timezone.utc).strftime("%Y%m%d-%H%M%S") |
| 2792 | work_id = f"work-{stamp}-{_slugify(slug_source, fallback='video')}" |
| 2793 | state = self._ensure_work_files(work_id, title=title, goal=goal) |
| 2794 | self._set_active_work(work_id) |
| 2795 | return _json_dump( |
| 2796 | { |
| 2797 | "status": "created", |
| 2798 | "work_id": work_id, |
| 2799 | "work_dir": str(self._paths(work_id)["work_dir"]), |
| 2800 | "stage": state.get("stage"), |
| 2801 | "goal_brief": goal, |
| 2802 | } |
| 2803 | ) |
| 2804 | |
| 2805 | |
| 2806 | @tool_parameters( |
| 2807 | tool_parameters_schema( |
| 2808 | work_id=StringSchema( |
| 2809 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 2810 | ), |
| 2811 | shot_count=IntegerSchema(description="Target number of shots", minimum=1, nullable=True), |
| 2812 | shot_duration_sec=IntegerSchema( |
| 2813 | description="Nominal duration per shot in seconds", |
| 2814 | minimum=1, |
| 2815 | nullable=True, |
| 2816 | ), |
| 2817 | generation_mode=StringSchema( |
| 2818 | "Shot generation mode", |
| 2819 | enum=["sequential", "parallel"], |
| 2820 | nullable=True, |
| 2821 | ), |
| 2822 | ) |
| 2823 | ) |
| 2824 | class SetDirectorGoalTool(DirectorTool): |
| 2825 | @property |
| 2826 | def name(self) -> str: |
| 2827 | return "set_director_goal" |
| 2828 | |
| 2829 | async def execute( |
| 2830 | self, |
| 2831 | work_id: str | None = None, |
| 2832 | shot_count: int | None = None, |
| 2833 | shot_duration_sec: int | None = None, |
| 2834 | generation_mode: str | None = None, |
| 2835 | **kwargs: Any, |
| 2836 | ) -> str: |
| 2837 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 2838 | if not resolved_work_id: |
| 2839 | return "Error: No active director work. Call start_director first." |
| 2840 | if all( |
| 2841 | value is None |
| 2842 | for value in (shot_count, shot_duration_sec, generation_mode) |
| 2843 | ): |
| 2844 | return "Error: At least one goal field must be provided." |
| 2845 | state = self._load_state(resolved_work_id) |
| 2846 | goal = state.setdefault("goal", {}) |
| 2847 | if not isinstance(goal, dict): |
| 2848 | goal = {} |
| 2849 | state["goal"] = goal |
| 2850 | try: |
| 2851 | previous_shot_count = int(goal.get("shot_count") or 0) |
| 2852 | except (TypeError, ValueError): |
| 2853 | previous_shot_count = 0 |
| 2854 | if shot_count is not None: |
| 2855 | goal["shot_count"] = shot_count |
| 2856 | lock_reference_image(state) |
| 2857 | self._lock_session_reference_image() |
| 2858 | try: |
| 2859 | new_shot_count = int(shot_count) |
| 2860 | except (TypeError, ValueError): |
| 2861 | new_shot_count = 0 |
| 2862 | if ( |
| 2863 | previous_shot_count <= 0 |
| 2864 | and new_shot_count > 0 |
| 2865 | and not self._session_auto_generate() |
| 2866 | and not bool(state.get("auto_generate")) |
| 2867 | ): |
| 2868 | state[SHOT_COUNT_NEXT_STEP_HINT_PENDING_KEY] = True |
| 2869 | if shot_duration_sec is not None: |
| 2870 | goal["shot_duration_sec"] = shot_duration_sec |
| 2871 | if generation_mode is not None: |
| 2872 | goal["generation_mode"] = generation_mode |
| 2873 | self._sync_stage_from_state(state) |
| 2874 | self._save_state(resolved_work_id, state) |
| 2875 | self._refresh_fact(resolved_work_id, state) |
| 2876 | return _json_dump( |
| 2877 | { |
| 2878 | "status": "ok", |
| 2879 | "work_id": resolved_work_id, |
| 2880 | "goal": goal, |
| 2881 | "stage": state.get("stage"), |
| 2882 | } |
| 2883 | ) |
| 2884 | |
| 2885 | |
| 2886 | @tool_parameters( |
| 2887 | tool_parameters_schema( |
| 2888 | work_id=StringSchema( |
| 2889 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 2890 | ), |
| 2891 | include_shots=BooleanSchema(description="Include per-shot summary rows", default=True), |
| 2892 | include_jobs=BooleanSchema(description="Include recent job rows", default=True), |
| 2893 | limit=IntegerSchema(description="Maximum shots/jobs to return", minimum=1, maximum=200), |
| 2894 | ) |
| 2895 | ) |
| 2896 | class GetWorkplaceStatusTool(DirectorTool): |
| 2897 | @property |
| 2898 | def name(self) -> str: |
| 2899 | return "get_workplace_status" |
| 2900 | |
| 2901 | async def execute( |
| 2902 | self, |
| 2903 | work_id: str | None = None, |
| 2904 | include_shots: bool = True, |
| 2905 | include_jobs: bool = True, |
| 2906 | limit: int = 20, |
| 2907 | **kwargs: Any, |
| 2908 | ) -> str: |
| 2909 | resolved_work_id, work_dir = self._resolve_work_id(work_id) |
| 2910 | if not resolved_work_id or not work_dir: |
| 2911 | return "Error: No active director work. Call start_director first." |
| 2912 | state = self._load_state(resolved_work_id) |
| 2913 | self._sync_stage_from_state(state) |
| 2914 | shots = self._shot_entries(state) |
| 2915 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 2916 | payload: dict[str, Any] = { |
| 2917 | "work_id": resolved_work_id, |
| 2918 | "work_dir": str(work_dir), |
| 2919 | "stage": state.get("stage"), |
| 2920 | "story_confirmed": bool(state.get("story_confirmed")), |
| 2921 | "goal_brief": state.get("goal_brief"), |
| 2922 | "goal": state.get("goal", {}), |
| 2923 | "final_output_path": state.get("final_output_path"), |
| 2924 | "final_output_url": state.get("final_output_url"), |
| 2925 | "reference_image_present": reference_image_present(state.get("reference_image")), |
| 2926 | "reference_image_locked": is_reference_image_locked(state), |
| 2927 | "auto_generate": bool(state.get("auto_generate")), |
| 2928 | "auto_generate_shot_count": self._effective_auto_generate_shot_count(goal), |
| 2929 | "reference_image_needs_story_rewrite": self._session_reference_needs_rewrite(), |
| 2930 | "counts": self._status_counts(state), |
| 2931 | "pending_remote_jobs": self._pending_remote_jobs(state), |
| 2932 | "story_path": str(self._paths(resolved_work_id)["story"]), |
| 2933 | "story_profile_path": str(self._paths(resolved_work_id)["story_profile"]), |
| 2934 | "fact_path": str(self._paths(resolved_work_id)["fact"]), |
| 2935 | "next_recommended_action": self._next_action(state), |
| 2936 | # Only assets with a textual profile are exposed to the agent. |
| 2937 | # Binary media stays in the local workspace and is resolved only |
| 2938 | # after a human approves the recommendation. |
| 2939 | "memory_assets": self._memory_asset_catalog(resolved_work_id), |
| 2940 | } |
| 2941 | if include_shots: |
| 2942 | payload["shots"] = [ |
| 2943 | { |
| 2944 | "shot_id": item.get("shot_id"), |
| 2945 | "shot_key": item.get("shot_key"), |
| 2946 | "status": item.get("status"), |
| 2947 | "summary": item.get("summary"), |
| 2948 | "cut": bool(item.get("cut", True)), |
| 2949 | "has_shot_spec": bool(item.get("has_shot_spec")), |
| 2950 | "has_artifact": bool(item.get("artifact_path") or item.get("artifact_url")), |
| 2951 | "artifact_path": item.get("artifact_path"), |
| 2952 | "artifact_url": item.get("artifact_url"), |
| 2953 | "last_review": item.get("last_review"), |
| 2954 | "review_notes": item.get("review_notes") or "", |
| 2955 | } |
| 2956 | for item in shots[:limit] |
| 2957 | ] |
| 2958 | if include_jobs: |
| 2959 | jobs_dir = self._paths(resolved_work_id)["jobs"] |
| 2960 | jobs: list[dict[str, Any]] = [] |
| 2961 | for path in sorted(jobs_dir.glob("*.json"), reverse=True)[:limit]: |
| 2962 | data = self._read_json(path, {}) |
| 2963 | if isinstance(data, dict): |
| 2964 | jobs.append( |
| 2965 | { |
| 2966 | "job_id": data.get("job_id"), |
| 2967 | "kind": data.get("kind"), |
| 2968 | "status": data.get("status"), |
| 2969 | "target": data.get("target"), |
| 2970 | "created_at": data.get("created_at"), |
| 2971 | } |
| 2972 | ) |
| 2973 | payload["jobs"] = jobs |
| 2974 | return _json_dump(payload) |
| 2975 | |
| 2976 | |
| 2977 | @tool_parameters( |
| 2978 | tool_parameters_schema( |
| 2979 | work_id=StringSchema( |
| 2980 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 2981 | ), |
| 2982 | ) |
| 2983 | ) |
| 2984 | class GetStoryTool(DirectorTool): |
| 2985 | @property |
| 2986 | def name(self) -> str: |
| 2987 | return "get_story" |
| 2988 | |
| 2989 | async def execute(self, work_id: str | None = None, **kwargs: Any) -> str: |
| 2990 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 2991 | if not resolved_work_id: |
| 2992 | return "Error: No active director work. Call start_director first." |
| 2993 | paths = self._paths(resolved_work_id) |
| 2994 | return _json_dump( |
| 2995 | { |
| 2996 | "work_id": resolved_work_id, |
| 2997 | "story_md": paths["story"].read_text(encoding="utf-8"), |
| 2998 | "story_profile": self._load_story_profile(resolved_work_id), |
| 2999 | } |
| 3000 | ) |
| 3001 | |
| 3002 | |
| 3003 | @tool_parameters( |
| 3004 | tool_parameters_schema( |
| 3005 | topic=StringSchema( |
| 3006 | "Guidance topic to load, e.g. 'shot-sequence-patterns' or 'shot-prompt-writer'" |
| 3007 | ), |
| 3008 | required=["topic"], |
| 3009 | ) |
| 3010 | ) |
| 3011 | class GetGuidanceTool(DirectorTool): |
| 3012 | @property |
| 3013 | def name(self) -> str: |
| 3014 | return "get_guidance" |
| 3015 | |
| 3016 | async def execute(self, topic: str = "", **kwargs: Any) -> str: |
| 3017 | manager = PEManager.instance() |
| 3018 | active = manager.active_for_session(self._session_key.get()) |
| 3019 | path = manager.resolve_reference(topic, name=active) |
| 3020 | if path is None: |
| 3021 | available = ", ".join(manager.list_references(name=active)) or "(none)" |
| 3022 | return prompts.text( |
| 3023 | "director.guidance.not_found", topic=topic, available=available |
| 3024 | ) |
| 3025 | return path.read_text(encoding="utf-8") |
| 3026 | |
| 3027 | |
| 3028 | @tool_parameters( |
| 3029 | tool_parameters_schema( |
| 3030 | work_id=StringSchema( |
| 3031 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3032 | ), |
| 3033 | story_md=StringSchema("Story markdown or screenplay text"), |
| 3034 | story_profile=ObjectSchema( |
| 3035 | description=( |
| 3036 | "Structured story profile JSON; use for shot mapping and beat lookup. " |
| 3037 | "When provided, must include a non-empty summary and beats " |
| 3038 | "(array of {shot_id, summary})." |
| 3039 | ), |
| 3040 | additional_properties=True, |
| 3041 | nullable=True, |
| 3042 | ), |
| 3043 | confirmed=BooleanSchema( |
| 3044 | description=( |
| 3045 | "Whether the screenplay is locked for generation. " |
| 3046 | "Set to true only after the user explicitly confirms the screenplay in chat." |
| 3047 | ), |
| 3048 | default=False, |
| 3049 | ), |
| 3050 | summary=StringSchema("Optional short story summary to cache in state", nullable=True), |
| 3051 | required=["story_md"], |
| 3052 | ) |
| 3053 | ) |
| 3054 | class WriteStoryTool(DirectorTool): |
| 3055 | @property |
| 3056 | def name(self) -> str: |
| 3057 | return "write_story" |
| 3058 | |
| 3059 | async def execute( |
| 3060 | self, |
| 3061 | story_md: str, |
| 3062 | work_id: str | None = None, |
| 3063 | story_profile: dict[str, Any] | None = None, |
| 3064 | confirmed: bool = False, |
| 3065 | summary: str | None = None, |
| 3066 | **kwargs: Any, |
| 3067 | ) -> str: |
| 3068 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3069 | if not resolved_work_id: |
| 3070 | return "Error: No active director work. Call start_director first." |
| 3071 | paths = self._paths(resolved_work_id) |
| 3072 | if isinstance(story_profile, dict): |
| 3073 | profile_error = _story_profile_validation_error(story_profile) |
| 3074 | if profile_error: |
| 3075 | return profile_error |
| 3076 | previous_profile = self._load_story_profile(resolved_work_id) |
| 3077 | prepared_profile = dict(story_profile) |
| 3078 | _preserve_story_profile_language(prepared_profile, previous_profile) |
| 3079 | if "language" not in prepared_profile: |
| 3080 | from nanobot.session.generation_settings import get_generation_settings |
| 3081 | from nanobot.session.manager import SessionManager |
| 3082 | |
| 3083 | session = SessionManager(self.workspace).get_or_create(self._session_key.get()) |
| 3084 | metadata = session.metadata if isinstance(session.metadata, dict) else {} |
| 3085 | settings = get_generation_settings(metadata) |
| 3086 | _apply_story_profile_language( |
| 3087 | prepared_profile, |
| 3088 | str(settings.get("language") or ""), |
| 3089 | ) |
| 3090 | language_error = _story_profile_language_validation_error(prepared_profile) |
| 3091 | if language_error: |
| 3092 | return language_error |
| 3093 | screenplay_error = _story_md_language_validation_error(story_md, prepared_profile) |
| 3094 | if screenplay_error: |
| 3095 | return screenplay_error |
| 3096 | self._save_story_profile(resolved_work_id, prepared_profile) |
| 3097 | story_profile = prepared_profile |
| 3098 | else: |
| 3099 | screenplay_error = _story_md_language_validation_error( |
| 3100 | story_md, |
| 3101 | self._load_story_profile(resolved_work_id), |
| 3102 | ) |
| 3103 | if screenplay_error: |
| 3104 | return screenplay_error |
| 3105 | self._write_text(paths["story"], story_md) |
| 3106 | self._clear_reference_image_story_rewrite_flag() |
| 3107 | if confirmed: |
| 3108 | profile_error = _story_profile_validation_error( |
| 3109 | self._load_story_profile(resolved_work_id), |
| 3110 | ) |
| 3111 | if profile_error: |
| 3112 | return ( |
| 3113 | "Error: Cannot confirm story without a valid story_profile. " |
| 3114 | "Call write_story with story_profile including a non-empty summary " |
| 3115 | "and at least one beat in beats." |
| 3116 | ) |
| 3117 | state = self._load_state(resolved_work_id) |
| 3118 | if summary: |
| 3119 | state["latest_story_summary"] = summary |
| 3120 | elif isinstance(story_profile, dict): |
| 3121 | state["latest_story_summary"] = story_profile["summary"].strip() |
| 3122 | if confirmed: |
| 3123 | state["story_confirmed"] = True |
| 3124 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 3125 | try: |
| 3126 | if int(goal.get("shot_count") or 0) > 0: |
| 3127 | lock_reference_image(state) |
| 3128 | self._lock_session_reference_image() |
| 3129 | except (TypeError, ValueError): |
| 3130 | pass |
| 3131 | # Agent has reconciled the user's story edit — clear the pending flag. |
| 3132 | state.pop("story_pending_agent_review", None) |
| 3133 | self._sync_stage_from_state(state) |
| 3134 | self._save_state(resolved_work_id, state) |
| 3135 | confirm = self._confirm(resolved_work_id) |
| 3136 | return _json_dump( |
| 3137 | { |
| 3138 | "status": "ok", |
| 3139 | "work_id": resolved_work_id, |
| 3140 | "story_confirmed": bool(state.get("story_confirmed")), |
| 3141 | "stage": state.get("stage"), |
| 3142 | "confirmation": confirm, |
| 3143 | } |
| 3144 | ) |
| 3145 | |
| 3146 | |
| 3147 | @tool_parameters( |
| 3148 | tool_parameters_schema( |
| 3149 | work_id=StringSchema( |
| 3150 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3151 | ), |
| 3152 | ) |
| 3153 | ) |
| 3154 | class GetFactTool(DirectorTool): |
| 3155 | @property |
| 3156 | def name(self) -> str: |
| 3157 | return "get_fact" |
| 3158 | |
| 3159 | async def execute(self, work_id: str | None = None, **kwargs: Any) -> str: |
| 3160 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3161 | if not resolved_work_id: |
| 3162 | return "Error: No active director work. Call start_director first." |
| 3163 | return self._paths(resolved_work_id)["fact"].read_text(encoding="utf-8") |
| 3164 | |
| 3165 | |
| 3166 | @tool_parameters( |
| 3167 | tool_parameters_schema( |
| 3168 | work_id=StringSchema( |
| 3169 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3170 | ), |
| 3171 | ) |
| 3172 | ) |
| 3173 | class ConfirmFactTool(DirectorTool): |
| 3174 | @property |
| 3175 | def name(self) -> str: |
| 3176 | return "confirm_fact" |
| 3177 | |
| 3178 | async def execute(self, work_id: str | None = None, **kwargs: Any) -> str: |
| 3179 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3180 | if not resolved_work_id: |
| 3181 | return "Error: No active director work. Call start_director first." |
| 3182 | return _json_dump(self._confirm(resolved_work_id)) |
| 3183 | |
| 3184 | |
| 3185 | @tool_parameters( |
| 3186 | tool_parameters_schema( |
| 3187 | work_id=StringSchema( |
| 3188 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3189 | ), |
| 3190 | shot_id=IntegerSchema(description="1-based shot number", minimum=1), |
| 3191 | required=["shot_id"], |
| 3192 | ) |
| 3193 | ) |
| 3194 | class GetShotTool(DirectorTool): |
| 3195 | @property |
| 3196 | def name(self) -> str: |
| 3197 | return "get_shot" |
| 3198 | |
| 3199 | async def execute(self, shot_id: int, work_id: str | None = None, **kwargs: Any) -> str: |
| 3200 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3201 | if not resolved_work_id: |
| 3202 | return "Error: No active director work. Call start_director first." |
| 3203 | shot = self._load_shot(resolved_work_id, shot_id) |
| 3204 | if not shot: |
| 3205 | return f"Error: Shot {shot_id} does not exist in work {resolved_work_id}." |
| 3206 | return _json_dump(shot) |
| 3207 | |
| 3208 | |
| 3209 | def build_create_shot_prompt_parameters() -> dict[str, Any]: |
| 3210 | """Build create_shot_prompt's JSON schema at call time so PE switches take effect.""" |
| 3211 | return tool_parameters_schema( |
| 3212 | work_id=StringSchema( |
| 3213 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3214 | ), |
| 3215 | shot_id=IntegerSchema(description="1-based shot number", minimum=1), |
| 3216 | cut=BooleanSchema( |
| 3217 | description="Whether this shot starts a fresh cut. false means generate as a continuation from the previous shot tail frame." |
| 3218 | ), |
| 3219 | caption=StringSchema(prompts.text("director.shot_caption.description")), |
| 3220 | status=StringSchema( |
| 3221 | "Optional explicit shot status", |
| 3222 | enum=[ |
| 3223 | "planned", |
| 3224 | "prompt_ready", |
| 3225 | "revised_prompt_ready", |
| 3226 | "queued", |
| 3227 | "generated", |
| 3228 | "error", |
| 3229 | "review_pass", |
| 3230 | "review_fail", |
| 3231 | "approved", |
| 3232 | ], |
| 3233 | nullable=True, |
| 3234 | ), |
| 3235 | required=["shot_id", "cut", "caption"], |
| 3236 | ) |
| 3237 | |
| 3238 | |
| 3239 | class CreateShotPromptTool(DirectorTool): |
| 3240 | @property |
| 3241 | def name(self) -> str: |
| 3242 | return "create_shot_prompt" |
| 3243 | |
| 3244 | @property |
| 3245 | def parameters(self) -> dict[str, Any]: |
| 3246 | return build_create_shot_prompt_parameters() |
| 3247 | |
| 3248 | async def execute( |
| 3249 | self, |
| 3250 | shot_id: int, |
| 3251 | work_id: str | None = None, |
| 3252 | cut: bool = True, |
| 3253 | caption: str | None = None, |
| 3254 | status: str | None = None, |
| 3255 | **kwargs: Any, |
| 3256 | ) -> str: |
| 3257 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3258 | if not resolved_work_id: |
| 3259 | return "Error: No active director work. Call start_director first." |
| 3260 | if not _allow_workflow_operation("create_shot_prompt"): |
| 3261 | return _workflow_gate_error("create_shot_prompt") |
| 3262 | if not isinstance(caption, str) or not caption.strip(): |
| 3263 | return "Error: caption is required." |
| 3264 | story_profile = self._load_story_profile(resolved_work_id) |
| 3265 | language_error = _caption_language_validation_error(caption, story_profile) |
| 3266 | if language_error: |
| 3267 | return language_error |
| 3268 | shot = self._load_shot(resolved_work_id, shot_id) |
| 3269 | is_revised_prompt = self._is_revised_prompt_update(shot) |
| 3270 | shot["shot_id"] = shot_id |
| 3271 | shot["shot_key"] = _shot_key(shot_id) |
| 3272 | shot["cut"] = cut |
| 3273 | shot["caption"] = caption.strip() |
| 3274 | shot.pop("shot_spec", None) |
| 3275 | shot.pop("summary", None) |
| 3276 | shot["summary"] = self._summary_from_shot(shot) |
| 3277 | shot.pop("prompt", None) |
| 3278 | shot.pop("negative_prompt", None) |
| 3279 | for key in ( |
| 3280 | "artifact_url", |
| 3281 | "artifact_path", |
| 3282 | "generation_error", |
| 3283 | "last_job_id", |
| 3284 | "echo", |
| 3285 | "remote_result", |
| 3286 | "last_review", |
| 3287 | "review_notes", |
| 3288 | ): |
| 3289 | shot.pop(key, None) |
| 3290 | if status is not None: |
| 3291 | shot["status"] = status |
| 3292 | else: |
| 3293 | shot["status"] = "revised_prompt_ready" if is_revised_prompt else "prompt_ready" |
| 3294 | state = self._load_state(resolved_work_id) |
| 3295 | sync_shot_echo_duration(shot, resolve_echo_duration_seconds(shot, state)) |
| 3296 | self._save_shot(resolved_work_id, shot_id, shot) |
| 3297 | |
| 3298 | shots = state.setdefault("shots", {}) |
| 3299 | if not isinstance(shots, dict): |
| 3300 | shots = {} |
| 3301 | state["shots"] = shots |
| 3302 | shots[shot["shot_key"]] = self._state_shot_entry(shot) |
| 3303 | self._sync_stage_from_state(state) |
| 3304 | self._save_state(resolved_work_id, state) |
| 3305 | confirm = self._confirm(resolved_work_id) |
| 3306 | return _json_dump( |
| 3307 | { |
| 3308 | "status": "ok", |
| 3309 | "work_id": resolved_work_id, |
| 3310 | "shot_id": shot_id, |
| 3311 | "shot_key": shot["shot_key"], |
| 3312 | "shot_status": shot["status"], |
| 3313 | "confirmation": confirm, |
| 3314 | } |
| 3315 | ) |
| 3316 | |
| 3317 | |
| 3318 | @tool_parameters( |
| 3319 | tool_parameters_schema( |
| 3320 | work_id=StringSchema( |
| 3321 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3322 | ), |
| 3323 | shot_id=IntegerSchema(description="1-based shot number", minimum=1), |
| 3324 | verdict=StringSchema( |
| 3325 | "Review result for the shot", |
| 3326 | enum=["accept", "revise"], |
| 3327 | ), |
| 3328 | review_source=StringSchema( |
| 3329 | "Who provided the review result", |
| 3330 | enum=["human", "vlm"], |
| 3331 | ), |
| 3332 | feedback=StringSchema( |
| 3333 | "Required when verdict='revise'; concise feedback for the next revision round", |
| 3334 | nullable=True, |
| 3335 | ), |
| 3336 | required=["shot_id", "verdict"], |
| 3337 | ) |
| 3338 | ) |
| 3339 | class ReviewShotTool(DirectorTool): |
| 3340 | @property |
| 3341 | def name(self) -> str: |
| 3342 | return "review_shot" |
| 3343 | |
| 3344 | async def execute( |
| 3345 | self, |
| 3346 | shot_id: int, |
| 3347 | verdict: str, |
| 3348 | work_id: str | None = None, |
| 3349 | review_source: str = "human", |
| 3350 | feedback: str | None = None, |
| 3351 | **kwargs: Any, |
| 3352 | ) -> str: |
| 3353 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3354 | if not resolved_work_id: |
| 3355 | return "Error: No active director work. Call start_director first." |
| 3356 | if not _allow_workflow_operation("review_shot"): |
| 3357 | return _workflow_gate_error("review_shot") |
| 3358 | try: |
| 3359 | shot, state = self._apply_shot_review( |
| 3360 | resolved_work_id, |
| 3361 | shot_id, |
| 3362 | verdict=verdict, |
| 3363 | review_source=review_source, |
| 3364 | feedback=feedback, |
| 3365 | ) |
| 3366 | except ValueError as exc: |
| 3367 | return f"Error: {exc}" |
| 3368 | return _json_dump( |
| 3369 | { |
| 3370 | "status": "ok", |
| 3371 | "work_id": resolved_work_id, |
| 3372 | "shot_id": shot_id, |
| 3373 | "shot_status": shot.get("status"), |
| 3374 | "stage": state.get("stage"), |
| 3375 | "review_notes": shot.get("review_notes") or "", |
| 3376 | "last_review": shot.get("last_review"), |
| 3377 | } |
| 3378 | ) |
| 3379 | |
| 3380 | |
| 3381 | @tool_parameters( |
| 3382 | tool_parameters_schema( |
| 3383 | work_id=StringSchema( |
| 3384 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3385 | ), |
| 3386 | shot_id=IntegerSchema(description="1-based shot number", minimum=1), |
| 3387 | reference_shot_ids=ArraySchema( |
| 3388 | IntegerSchema( |
| 3389 | description="Earlier logical shot ID used as visual reference", minimum=1 |
| 3390 | ), |
| 3391 | description="Logical prior shot IDs selected as Echo references for this shot.", |
| 3392 | ), |
| 3393 | selection_note=StringSchema( |
| 3394 | "Optional short note explaining why these references were selected", |
| 3395 | nullable=True, |
| 3396 | ), |
| 3397 | required=["shot_id", "reference_shot_ids"], |
| 3398 | ) |
| 3399 | ) |
| 3400 | class SetShotReferencesTool(DirectorTool): |
| 3401 | @property |
| 3402 | def name(self) -> str: |
| 3403 | return "set_shot_references" |
| 3404 | |
| 3405 | def apply_set_references( |
| 3406 | self, |
| 3407 | work_id: str, |
| 3408 | shot_id: int, |
| 3409 | reference_shot_ids: list[int], |
| 3410 | selection_note: str | None = None, |
| 3411 | ) -> dict[str, Any]: |
| 3412 | shot = self._load_shot(work_id, shot_id) |
| 3413 | if not shot: |
| 3414 | raise ValueError(f"Shot {shot_id} does not exist in work {work_id}.") |
| 3415 | if not isinstance(shot.get("caption"), str) or not shot.get("caption", "").strip(): |
| 3416 | raise ValueError(f"Shot {shot_id} has no caption yet. Call create_shot_prompt first.") |
| 3417 | normalized = self._normalize_reference_shot_ids( |
| 3418 | shot_id, |
| 3419 | reference_shot_ids, |
| 3420 | cut=bool(shot.get("cut", True)), |
| 3421 | ) |
| 3422 | shot["planned_reference_shot_ids"] = normalized |
| 3423 | if isinstance(selection_note, str) and selection_note.strip(): |
| 3424 | shot["reference_selection_note"] = selection_note.strip() |
| 3425 | self._save_shot(work_id, shot_id, shot) |
| 3426 | state = self._load_state(work_id) |
| 3427 | shots = state.setdefault("shots", {}) |
| 3428 | if isinstance(shots, dict): |
| 3429 | shots[_shot_key(shot_id)] = self._state_shot_entry(shot) |
| 3430 | self._save_state(work_id, state) |
| 3431 | return shot |
| 3432 | |
| 3433 | async def execute( |
| 3434 | self, |
| 3435 | shot_id: int, |
| 3436 | reference_shot_ids: list[int], |
| 3437 | work_id: str | None = None, |
| 3438 | selection_note: str | None = None, |
| 3439 | **kwargs: Any, |
| 3440 | ) -> str: |
| 3441 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3442 | if not resolved_work_id: |
| 3443 | return "Error: No active director work. Call start_director first." |
| 3444 | if not _allow_workflow_operation("set_shot_references"): |
| 3445 | return _workflow_gate_error("set_shot_references") |
| 3446 | try: |
| 3447 | shot = self.apply_set_references( |
| 3448 | resolved_work_id, |
| 3449 | shot_id, |
| 3450 | reference_shot_ids, |
| 3451 | selection_note=selection_note, |
| 3452 | ) |
| 3453 | except ValueError as exc: |
| 3454 | return f"Error: {exc}" |
| 3455 | return _json_dump( |
| 3456 | { |
| 3457 | "status": "ok", |
| 3458 | "work_id": resolved_work_id, |
| 3459 | "shot_id": shot_id, |
| 3460 | "planned_reference_shot_ids": shot.get("planned_reference_shot_ids") or [], |
| 3461 | "reference_selection_note": shot.get("reference_selection_note") or "", |
| 3462 | } |
| 3463 | ) |
| 3464 | |
| 3465 | |
| 3466 | @tool_parameters( |
| 3467 | tool_parameters_schema( |
| 3468 | work_id=StringSchema( |
| 3469 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3470 | ), |
| 3471 | shot_id=IntegerSchema(description="Target 1-based shot number", minimum=1), |
| 3472 | recommendations=ArraySchema( |
| 3473 | ObjectSchema( |
| 3474 | image_asset_id=StringSchema( |
| 3475 | "Profile-bearing Memory Workspace asset used for the slot image" |
| 3476 | ), |
| 3477 | audio_asset_id=StringSchema( |
| 3478 | "Optional profile-bearing asset used for slot audio", nullable=True |
| 3479 | ), |
| 3480 | reason=StringSchema("Short reason this slot helps the target shot"), |
| 3481 | required=["image_asset_id", "reason"], |
| 3482 | additional_properties=False, |
| 3483 | ), |
| 3484 | description="Ordered recommendation draft; zero to seven slots.", |
| 3485 | max_items=7, |
| 3486 | ), |
| 3487 | required=["shot_id", "recommendations"], |
| 3488 | ) |
| 3489 | ) |
| 3490 | class SetShotMemoryRecommendationsTool(DirectorTool): |
| 3491 | """Let the agent propose slots without granting generation approval.""" |
| 3492 | |
| 3493 | @property |
| 3494 | def name(self) -> str: |
| 3495 | return "set_shot_memory_recommendations" |
| 3496 | |
| 3497 | async def execute( |
| 3498 | self, |
| 3499 | shot_id: int, |
| 3500 | recommendations: list[dict[str, Any]], |
| 3501 | work_id: str | None = None, |
| 3502 | **kwargs: Any, |
| 3503 | ) -> str: |
| 3504 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3505 | if not resolved_work_id: |
| 3506 | return "Error: No active director work. Call start_director first." |
| 3507 | if not _allow_workflow_operation("set_shot_memory_recommendations"): |
| 3508 | return _workflow_gate_error("set_shot_memory_recommendations") |
| 3509 | if len(recommendations) > 7: |
| 3510 | return "Error: recommendations cannot exceed 7 slots." |
| 3511 | shot = self._load_shot(resolved_work_id, shot_id) |
| 3512 | if not shot: |
| 3513 | return f"Error: Shot {shot_id} does not exist in work {resolved_work_id}." |
| 3514 | catalog = { |
| 3515 | item["asset_id"]: item for item in self._memory_asset_catalog(resolved_work_id) |
| 3516 | } |
| 3517 | normalized: list[dict[str, Any]] = [] |
| 3518 | seen_images: set[str] = set() |
| 3519 | for raw in recommendations: |
| 3520 | if not isinstance(raw, dict): |
| 3521 | return "Error: each recommendation must be an object." |
| 3522 | image_id = str(raw.get("image_asset_id") or "").strip() |
| 3523 | audio_id = str(raw.get("audio_asset_id") or "").strip() or None |
| 3524 | reason = str(raw.get("reason") or "").strip() |
| 3525 | if not image_id or image_id not in catalog: |
| 3526 | return f"Error: unknown or unprofiled image asset '{image_id}'." |
| 3527 | if catalog[image_id].get("media_type") == "audio": |
| 3528 | return f"Error: asset '{image_id}' has no image." |
| 3529 | if image_id in seen_images: |
| 3530 | return f"Error: duplicate image asset '{image_id}'." |
| 3531 | if audio_id is not None and audio_id not in catalog: |
| 3532 | return f"Error: unknown or unprofiled audio asset '{audio_id}'." |
| 3533 | if audio_id is not None and catalog[audio_id].get("media_type") == "image": |
| 3534 | return f"Error: asset '{audio_id}' has no audio." |
| 3535 | if not reason: |
| 3536 | return "Error: every recommendation needs a reason." |
| 3537 | seen_images.add(image_id) |
| 3538 | normalized.append({ |
| 3539 | "image_asset_id": image_id, |
| 3540 | **({"audio_asset_id": audio_id} if audio_id else {}), |
| 3541 | "reason": reason[:500], |
| 3542 | }) |
| 3543 | shot["recommended_memory_slot_refs"] = normalized |
| 3544 | shot["memory_recommendation_source"] = "agent" |
| 3545 | shot["memory_recommendation_updated_at"] = _now_iso() |
| 3546 | self._save_shot(resolved_work_id, shot_id, shot) |
| 3547 | return _json_dump({ |
| 3548 | "status": "ok", |
| 3549 | "work_id": resolved_work_id, |
| 3550 | "shot_id": shot_id, |
| 3551 | "recommended_memory_slot_refs": normalized, |
| 3552 | "approval": "pending_human", |
| 3553 | }) |
| 3554 | |
| 3555 | |
| 3556 | @tool_parameters( |
| 3557 | tool_parameters_schema( |
| 3558 | work_id=StringSchema( |
| 3559 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3560 | ), |
| 3561 | shot_id=IntegerSchema(description="1-based shot number", minimum=1), |
| 3562 | reference_shot_ids=ArraySchema( |
| 3563 | IntegerSchema( |
| 3564 | description="Earlier logical shot ID used as visual reference", minimum=1 |
| 3565 | ), |
| 3566 | description=( |
| 3567 | "Logical prior shot IDs the agent selected as Echo references. " |
| 3568 | "Almost every shot MUST reference at least one earlier shot for visual continuity — " |
| 3569 | "an empty list is allowed ONLY for shot_id=1, or when the shot introduces a completely new scene " |
| 3570 | "with entirely new characters that have zero visual overlap with any previous shot. " |
| 3571 | "When in doubt, include at least the most recent shot that shares a character, environment, or prop." |
| 3572 | ), |
| 3573 | ), |
| 3574 | selection_note=StringSchema( |
| 3575 | "Optional short note explaining why these references were selected", |
| 3576 | nullable=True, |
| 3577 | ), |
| 3578 | required=["shot_id", "reference_shot_ids"], |
| 3579 | ) |
| 3580 | ) |
| 3581 | class GenerateEchoShotTool(DirectorTool): |
| 3582 | @property |
| 3583 | def name(self) -> str: |
| 3584 | return "generate_echo_shot" |
| 3585 | |
| 3586 | def apply_generate( |
| 3587 | self, |
| 3588 | work_id: str, |
| 3589 | shot_id: int, |
| 3590 | reference_shot_ids: list[int], |
| 3591 | selection_note: str | None = None, |
| 3592 | condition_image_url: str | None = None, |
| 3593 | i2v_prompt: str | None = None, |
| 3594 | ) -> dict[str, Any]: |
| 3595 | state = self._load_state(work_id) |
| 3596 | existing_shot = self._load_shot(work_id, shot_id) |
| 3597 | if ( |
| 3598 | shot_id > 1 |
| 3599 | and str(state.get("stage") or "") == "awaiting_memory_build" |
| 3600 | and not existing_shot.get("memory_slots_user_configured") |
| 3601 | and not state.get("auto_generate") |
| 3602 | ): |
| 3603 | raise ValueError( |
| 3604 | "Build Memory must be reviewed and applied before generating this shot." |
| 3605 | ) |
| 3606 | caption = existing_shot.get("caption") |
| 3607 | # Skip language validation when an I2V prompt is supplied — the prompt |
| 3608 | # has already been rewritten by rewrite_prompt_for_i2v with the correct |
| 3609 | # first-frame contract sentence, and the underlying caption may contain |
| 3610 | # technical tokens from PE re-captioning that the validator flags. |
| 3611 | if isinstance(caption, str) and not i2v_prompt: |
| 3612 | language_error = _caption_language_validation_error( |
| 3613 | caption, |
| 3614 | self._load_story_profile(work_id), |
| 3615 | ) |
| 3616 | if language_error: |
| 3617 | raise ValueError(language_error.removeprefix("Error: ")) |
| 3618 | |
| 3619 | duration_value = resolve_echo_duration_seconds(existing_shot, state) |
| 3620 | num_frames = sync_shot_echo_duration(existing_shot, duration_value) |
| 3621 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 3622 | width = goal.get("width") |
| 3623 | height = goal.get("height") |
| 3624 | prompt_text = i2v_prompt.strip() if i2v_prompt else caption.strip() |
| 3625 | |
| 3626 | # Normalize references early so invalid refs fail before submission. |
| 3627 | normalized_reference_ids = self._normalize_reference_shot_ids( |
| 3628 | shot_id, reference_shot_ids, cut=bool(existing_shot.get("cut", True)) |
| 3629 | ) |
| 3630 | |
| 3631 | memory_slots = self._build_memory_slots( |
| 3632 | existing_shot.get("approved_memory_slots"), normalized_reference_ids, |
| 3633 | work_id=work_id, |
| 3634 | ) |
| 3635 | |
| 3636 | request_payload = self._build_r2v_payload( |
| 3637 | work_id, |
| 3638 | shot_id, |
| 3639 | prompt=prompt_text, |
| 3640 | num_frames=num_frames, |
| 3641 | width=int(width) if width is not None else None, |
| 3642 | height=int(height) if height is not None else None, |
| 3643 | condition_image_url=condition_image_url, |
| 3644 | memory_slots=memory_slots, |
| 3645 | ) |
| 3646 | |
| 3647 | state["stage"] = "shot_generating" |
| 3648 | job_id = _job_id("echo", work_id, _shot_key(shot_id)) |
| 3649 | try: |
| 3650 | job = self._submit_r2v_request( |
| 3651 | work_id, |
| 3652 | job_id, |
| 3653 | request_payload, |
| 3654 | target=_shot_key(shot_id), |
| 3655 | ) |
| 3656 | except EchoGeneratorBusyError: |
| 3657 | raise |
| 3658 | except EchoGeneratorUnavailableError: |
| 3659 | raise |
| 3660 | except RuntimeError as exc: |
| 3661 | raise ValueError(str(exc)) from exc |
| 3662 | |
| 3663 | self._write_json(self._job_path(work_id, job_id), job) |
| 3664 | self._clear_pending_remote_jobs_for_target(state, "generate_echo_shot", _shot_key(shot_id)) |
| 3665 | self._register_pending_remote_job(state, job) |
| 3666 | |
| 3667 | existing_shot["status"] = "queued" |
| 3668 | existing_shot.pop("generation_error", None) |
| 3669 | existing_shot["last_job_id"] = job_id |
| 3670 | existing_shot["reference_shot_ids"] = normalized_reference_ids |
| 3671 | if selection_note is not None: |
| 3672 | existing_shot["reference_selection_note"] = selection_note |
| 3673 | echo = existing_shot.get("echo") |
| 3674 | if not isinstance(echo, dict): |
| 3675 | echo = {} |
| 3676 | existing_shot["echo"] = echo |
| 3677 | echo.update( |
| 3678 | { |
| 3679 | "status": "queued", |
| 3680 | "reference_shot_ids": normalized_reference_ids, |
| 3681 | "selection_note": selection_note, |
| 3682 | "request_payload_path": job.get("request_payload_path"), |
| 3683 | "request_envelope_path": job.get("request_envelope_path"), |
| 3684 | "remote_task_id": ( |
| 3685 | job.get("remote", {}).get("remote_task_id") |
| 3686 | if isinstance(job.get("remote"), dict) |
| 3687 | else None |
| 3688 | ), |
| 3689 | "version_id": ( |
| 3690 | job.get("remote", {}).get("version_id") |
| 3691 | if isinstance(job.get("remote"), dict) |
| 3692 | else None |
| 3693 | ), |
| 3694 | } |
| 3695 | ) |
| 3696 | self._save_shot(work_id, shot_id, existing_shot) |
| 3697 | |
| 3698 | shots = state.setdefault("shots", {}) |
| 3699 | if isinstance(shots, dict): |
| 3700 | shots[_shot_key(shot_id)] = self._state_shot_entry(existing_shot) |
| 3701 | self._save_state(work_id, state) |
| 3702 | self._refresh_fact(work_id, state) |
| 3703 | return job |
| 3704 | |
| 3705 | def apply_generate_continuous( |
| 3706 | self, |
| 3707 | work_id: str, |
| 3708 | shot_id: int, |
| 3709 | condition_image_url: str, |
| 3710 | reference_shot_ids: list[int], |
| 3711 | selection_note: str | None = None, |
| 3712 | i2v_prompt: str | None = None, |
| 3713 | ) -> dict[str, Any]: |
| 3714 | """Submit a shot for I2V generation using the previous shot's tail frame as condition image. |
| 3715 | |
| 3716 | Rewrites the current shot's prompt for I2V format, then submits to the Echo |
| 3717 | backend with ``condition_image_url`` pointing to the tail-frame image. |
| 3718 | """ |
| 3719 | state = self._load_state(work_id) |
| 3720 | existing_shot = self._load_shot(work_id, shot_id) |
| 3721 | if ( |
| 3722 | shot_id > 1 |
| 3723 | and str(state.get("stage") or "") == "awaiting_memory_build" |
| 3724 | and not existing_shot.get("memory_slots_user_configured") |
| 3725 | and not state.get("auto_generate") |
| 3726 | ): |
| 3727 | raise ValueError( |
| 3728 | "Build Memory must be reviewed and applied before generating this shot." |
| 3729 | ) |
| 3730 | caption = existing_shot.get("caption") |
| 3731 | if not isinstance(caption, str) or not caption.strip(): |
| 3732 | raise ValueError(f"Shot {shot_id} has no caption yet. Call create_shot_prompt first.") |
| 3733 | |
| 3734 | # Defensive caption-language check (mirrors apply_generate). |
| 3735 | language_error = _caption_language_validation_error( |
| 3736 | caption, |
| 3737 | self._load_story_profile(work_id), |
| 3738 | ) |
| 3739 | if language_error: |
| 3740 | raise ValueError(language_error.removeprefix("Error: ")) |
| 3741 | |
| 3742 | story_profile = self._load_story_profile(work_id) |
| 3743 | caption_language = str( |
| 3744 | story_profile.get("caption_language") |
| 3745 | or story_profile.get("language") |
| 3746 | or "" |
| 3747 | ) |
| 3748 | |
| 3749 | # The WebSocket continuation path performs the multimodal rewrite |
| 3750 | # (ordinary PE + I2V skill + the extracted tail frame). Keep the |
| 3751 | # deterministic helper as a fallback for existing internal callers. |
| 3752 | rewritten_prompt = ( |
| 3753 | i2v_prompt.strip() |
| 3754 | if isinstance(i2v_prompt, str) and i2v_prompt.strip() |
| 3755 | else rewrite_prompt_for_i2v(caption.strip(), caption_language) |
| 3756 | ) |
| 3757 | |
| 3758 | # Persist the I2V prompt on the shot record that will be saved. |
| 3759 | existing_shot["i2v_prompt"] = rewritten_prompt |
| 3760 | |
| 3761 | duration_value = resolve_echo_duration_seconds(existing_shot, state) |
| 3762 | num_frames = sync_shot_echo_duration(existing_shot, duration_value) |
| 3763 | goal = state.get("goal") if isinstance(state.get("goal"), dict) else {} |
| 3764 | width = goal.get("width") |
| 3765 | height = goal.get("height") |
| 3766 | |
| 3767 | # Normalize references early so invalid refs fail before submission. |
| 3768 | normalized_reference_ids = self._normalize_reference_shot_ids( |
| 3769 | shot_id, reference_shot_ids, cut=bool(existing_shot.get("cut", True)) |
| 3770 | ) |
| 3771 | |
| 3772 | memory_slots = self._build_memory_slots( |
| 3773 | existing_shot.get("approved_memory_slots"), normalized_reference_ids, |
| 3774 | work_id=work_id, |
| 3775 | ) |
| 3776 | |
| 3777 | request_payload = self._build_r2v_payload( |
| 3778 | work_id, |
| 3779 | shot_id, |
| 3780 | prompt=rewritten_prompt, |
| 3781 | num_frames=num_frames, |
| 3782 | width=int(width) if width is not None else None, |
| 3783 | height=int(height) if height is not None else None, |
| 3784 | condition_image_url=condition_image_url, |
| 3785 | memory_slots=memory_slots, |
| 3786 | ) |
| 3787 | |
| 3788 | state["stage"] = "shot_generating" |
| 3789 | job_id = _job_id("echo", work_id, _shot_key(shot_id)) |
| 3790 | try: |
| 3791 | job = self._submit_r2v_request( |
| 3792 | work_id, |
| 3793 | job_id, |
| 3794 | request_payload, |
| 3795 | target=_shot_key(shot_id), |
| 3796 | ) |
| 3797 | except EchoGeneratorBusyError: |
| 3798 | raise |
| 3799 | except EchoGeneratorUnavailableError: |
| 3800 | raise |
| 3801 | except RuntimeError as exc: |
| 3802 | raise ValueError(str(exc)) from exc |
| 3803 | |
| 3804 | self._write_json(self._job_path(work_id, job_id), job) |
| 3805 | self._clear_pending_remote_jobs_for_target(state, "generate_echo_shot", _shot_key(shot_id)) |
| 3806 | self._register_pending_remote_job(state, job) |
| 3807 | |
| 3808 | existing_shot["status"] = "queued" |
| 3809 | existing_shot.pop("generation_error", None) |
| 3810 | existing_shot["last_job_id"] = job_id |
| 3811 | existing_shot["reference_shot_ids"] = normalized_reference_ids |
| 3812 | if selection_note is not None: |
| 3813 | existing_shot["reference_selection_note"] = selection_note |
| 3814 | echo = existing_shot.get("echo") |
| 3815 | if not isinstance(echo, dict): |
| 3816 | echo = {} |
| 3817 | existing_shot["echo"] = echo |
| 3818 | echo.update( |
| 3819 | { |
| 3820 | "status": "queued", |
| 3821 | "reference_shot_ids": normalized_reference_ids, |
| 3822 | "selection_note": selection_note, |
| 3823 | "request_payload_path": job.get("request_payload_path"), |
| 3824 | "request_envelope_path": job.get("request_envelope_path"), |
| 3825 | "remote_task_id": ( |
| 3826 | job.get("remote", {}).get("remote_task_id") |
| 3827 | if isinstance(job.get("remote"), dict) |
| 3828 | else None |
| 3829 | ), |
| 3830 | "version_id": ( |
| 3831 | job.get("remote", {}).get("version_id") |
| 3832 | if isinstance(job.get("remote"), dict) |
| 3833 | else None |
| 3834 | ), |
| 3835 | } |
| 3836 | ) |
| 3837 | self._save_shot(work_id, shot_id, existing_shot) |
| 3838 | |
| 3839 | shots = state.setdefault("shots", {}) |
| 3840 | if isinstance(shots, dict): |
| 3841 | shots[_shot_key(shot_id)] = self._state_shot_entry(existing_shot) |
| 3842 | self._save_state(work_id, state) |
| 3843 | self._refresh_fact(work_id, state) |
| 3844 | return job |
| 3845 | |
| 3846 | async def execute( |
| 3847 | self, |
| 3848 | shot_id: int, |
| 3849 | reference_shot_ids: list[int], |
| 3850 | work_id: str | None = None, |
| 3851 | selection_note: str | None = None, |
| 3852 | **kwargs: Any, |
| 3853 | ) -> str: |
| 3854 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 3855 | if not resolved_work_id: |
| 3856 | return "Error: No active director work. Call start_director first." |
| 3857 | if not _allow_workflow_operation("generate_echo_shot"): |
| 3858 | return _workflow_gate_error("generate_echo_shot") |
| 3859 | |
| 3860 | # 检查是否开启首尾衔接 → I2V 生成 |
| 3861 | shot = self._load_shot(resolved_work_id, shot_id) |
| 3862 | use_continuous = ( |
| 3863 | bool(shot.get("continuous_enabled")) |
| 3864 | and shot_id > 1 |
| 3865 | and not bool(self._load_state(resolved_work_id).get("auto_generate")) |
| 3866 | ) |
| 3867 | |
| 3868 | try: |
| 3869 | if use_continuous: |
| 3870 | previous_shot_id = shot_id - 1 |
| 3871 | prev_shot = self._load_shot(resolved_work_id, previous_shot_id) |
| 3872 | video_url = prev_shot.get("artifact_url") or ( |
| 3873 | prev_shot.get("echo") or {} |
| 3874 | ).get("result_url") |
| 3875 | if not video_url: |
| 3876 | return ( |
| 3877 | f"Error: previous shot {previous_shot_id} has no video " |
| 3878 | f"artifact; cannot extract tail frame for continuous generation" |
| 3879 | ) |
| 3880 | logger.info( |
| 3881 | "agent continuous-generate: shot_id={} using previous shot {} " |
| 3882 | "tail frame, extracting from video_url={}", |
| 3883 | shot_id, previous_shot_id, video_url, |
| 3884 | ) |
| 3885 | condition_image_url = await asyncio.to_thread( |
| 3886 | self._extract_and_publish_tail_frame, |
| 3887 | resolved_work_id, |
| 3888 | previous_shot_id, |
| 3889 | video_url, |
| 3890 | ) |
| 3891 | if not condition_image_url: |
| 3892 | return ( |
| 3893 | f"Error: failed to extract tail frame from shot " |
| 3894 | f"{previous_shot_id}" |
| 3895 | ) |
| 3896 | logger.info( |
| 3897 | "agent continuous-generate: tail frame ready, " |
| 3898 | "shot_id={} condition_image_url={}", |
| 3899 | shot_id, condition_image_url, |
| 3900 | ) |
| 3901 | # 确保 previous_shot_id 在 reference_shot_ids 中 |
| 3902 | if previous_shot_id not in reference_shot_ids: |
| 3903 | reference_shot_ids = sorted( |
| 3904 | set(reference_shot_ids) | {previous_shot_id} |
| 3905 | ) |
| 3906 | job = await asyncio.to_thread( |
| 3907 | self.apply_generate_continuous, |
| 3908 | resolved_work_id, |
| 3909 | shot_id, |
| 3910 | condition_image_url, |
| 3911 | reference_shot_ids, |
| 3912 | selection_note=selection_note, |
| 3913 | ) |
| 3914 | else: |
| 3915 | first_frame_url = None |
| 3916 | if shot_id == 1: |
| 3917 | first_frame_url = self._state_first_frame_url( |
| 3918 | self._load_state(resolved_work_id) |
| 3919 | ) |
| 3920 | if first_frame_url: |
| 3921 | logger.info( |
| 3922 | "agent generate_echo_shot: shot_id=1 using state.reference_image " |
| 3923 | "url={}", |
| 3924 | first_frame_url, |
| 3925 | ) |
| 3926 | profile = self._load_story_profile(resolved_work_id) |
| 3927 | language = "" |
| 3928 | if isinstance(profile, dict): |
| 3929 | language = str( |
| 3930 | profile.get("caption_language") or profile.get("language") or "" |
| 3931 | ) |
| 3932 | caption = str(shot.get("caption") or "").strip() |
| 3933 | i2v_prompt = ( |
| 3934 | rewrite_prompt_for_i2v(caption, language) if caption else None |
| 3935 | ) |
| 3936 | job = await asyncio.to_thread( |
| 3937 | self.apply_generate, |
| 3938 | resolved_work_id, |
| 3939 | shot_id, |
| 3940 | reference_shot_ids, |
| 3941 | selection_note=selection_note, |
| 3942 | condition_image_url=first_frame_url, |
| 3943 | i2v_prompt=i2v_prompt, |
| 3944 | ) |
| 3945 | else: |
| 3946 | job = await asyncio.to_thread( |
| 3947 | self.apply_generate, |
| 3948 | resolved_work_id, |
| 3949 | shot_id, |
| 3950 | reference_shot_ids, |
| 3951 | selection_note=selection_note, |
| 3952 | ) |
| 3953 | except EchoGeneratorBusyError as exc: |
| 3954 | return f"Error: {exc}" |
| 3955 | except EchoGeneratorUnavailableError as exc: |
| 3956 | return f"Error: {exc}" |
| 3957 | except ValueError as exc: |
| 3958 | return f"Error: {exc}" |
| 3959 | return _json_dump(job) |
| 3960 | |
| 3961 | |
| 3962 | @tool_parameters( |
| 3963 | tool_parameters_schema( |
| 3964 | work_id=StringSchema( |
| 3965 | "Optional explicit work ID; defaults to the active work", nullable=True |
| 3966 | ), |
| 3967 | shot_ids=ArraySchema( |
| 3968 | IntegerSchema(description="1-based shot number"), |
| 3969 | description="Optional explicit shot list; defaults to all known shots in order", |
| 3970 | nullable=True, |
| 3971 | ), |
| 3972 | ) |
| 3973 | ) |
| 3974 | class MergeShotTool(DirectorTool): |
| 3975 | @property |
| 3976 | def name(self) -> str: |
| 3977 | return "merge_shot" |
| 3978 | |
| 3979 | async def execute( |
| 3980 | self, |
| 3981 | work_id: str | None = None, |
| 3982 | shot_ids: list[int] | None = None, |
| 3983 | **kwargs: Any, |
| 3984 | ) -> str: |
| 3985 | try: |
| 3986 | job = await asyncio.to_thread( |
| 3987 | self.apply_merge, |
| 3988 | work_id=work_id, |
| 3989 | shot_ids=shot_ids, |
| 3990 | ) |
| 3991 | except (RuntimeError, ValueError) as exc: |
| 3992 | return f"Error: {exc}" |
| 3993 | return _json_dump(job) |
| 3994 | |
| 3995 | def apply_merge( |
| 3996 | self, |
| 3997 | work_id: str | None = None, |
| 3998 | shot_ids: list[int] | None = None, |
| 3999 | ) -> dict[str, Any]: |
| 4000 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 4001 | if not resolved_work_id: |
| 4002 | raise ValueError("No active director work. Call start_director first.") |
| 4003 | if not _allow_workflow_operation("merge_shot"): |
| 4004 | raise ValueError(_workflow_gate_error("merge_shot")) |
| 4005 | state = self._load_state(resolved_work_id) |
| 4006 | state["stage"] = "merging" |
| 4007 | state.pop("generation_error", None) |
| 4008 | available = self._shot_entries(state) |
| 4009 | if not available: |
| 4010 | raise ValueError("No shots exist yet. Create shot prompts first.") |
| 4011 | selected_ids = shot_ids or [int(item["shot_id"]) for item in available] |
| 4012 | selected_shots = [] |
| 4013 | for shot_id in selected_ids: |
| 4014 | shot = self._load_shot(resolved_work_id, shot_id) |
| 4015 | if not shot: |
| 4016 | raise ValueError(f"Shot {shot_id} does not exist in work {resolved_work_id}.") |
| 4017 | selected_shots.append(shot) |
| 4018 | job_id = _job_id("merge", resolved_work_id, "final") |
| 4019 | payload = self._build_merge_payload( |
| 4020 | resolved_work_id, |
| 4021 | selected_ids, |
| 4022 | selected_shots, |
| 4023 | ) |
| 4024 | job = self._submit_remote_request( |
| 4025 | resolved_work_id, |
| 4026 | job_id, |
| 4027 | payload, |
| 4028 | target="final", |
| 4029 | operation="merge_shot", |
| 4030 | ) |
| 4031 | self._write_json(self._job_path(resolved_work_id, job_id), job) |
| 4032 | self._clear_pending_remote_jobs_for_target(state, "merge_shot", "final") |
| 4033 | self._register_pending_remote_job(state, job) |
| 4034 | state.pop("merge_confirmation_requested_at", None) |
| 4035 | state["latest_merge_job_id"] = job_id |
| 4036 | self._sync_stage_from_state(state) |
| 4037 | self._save_state(resolved_work_id, state) |
| 4038 | self._refresh_fact(resolved_work_id, state) |
| 4039 | return job |
| 4040 | |
| 4041 | |
| 4042 | @tool_parameters( |
| 4043 | tool_parameters_schema( |
| 4044 | job_id=StringSchema("Director job ID to inspect"), |
| 4045 | work_id=StringSchema( |
| 4046 | "Optional explicit work ID; defaults to the active work or a repo-wide search", |
| 4047 | nullable=True, |
| 4048 | ), |
| 4049 | required=["job_id"], |
| 4050 | ) |
| 4051 | ) |
| 4052 | class GetDirectorJobTool(DirectorTool): |
| 4053 | @property |
| 4054 | def name(self) -> str: |
| 4055 | return "get_director_job" |
| 4056 | |
| 4057 | async def execute(self, job_id: str, work_id: str | None = None, **kwargs: Any) -> str: |
| 4058 | candidates: list[Path] = [] |
| 4059 | resolved_work_id, _ = self._resolve_work_id(work_id) |
| 4060 | if resolved_work_id: |
| 4061 | candidates.append(self._job_path(resolved_work_id, job_id)) |
| 4062 | else: |
| 4063 | self._ensure_root() |
| 4064 | for jobs_dir in self.works_root.glob("*/jobs"): |
| 4065 | candidates.append(jobs_dir / f"{job_id}.json") |
| 4066 | for path in candidates: |
| 4067 | if path.exists(): |
| 4068 | data = self._read_json(path, {}) |
| 4069 | if isinstance(data, dict): |
| 4070 | return _json_dump(data) |
| 4071 | return f"Error: Director job '{job_id}' was not found." |
| 4072 | |
| 4073 | |
| 4074 | def apply_echo_generate_shot_callback( |
| 4075 | workspace: Path, |
| 4076 | callback_payload: dict[str, Any], |
| 4077 | *, |
| 4078 | tools_config: Any | None = None, |
| 4079 | ) -> dict[str, Any]: |
| 4080 | """Apply one generate_echo_shot remote callback to the director workspace. |
| 4081 | |
| 4082 | Expected callback payload shape: |
| 4083 | - required: `work_id`, `job_id` |
| 4084 | - optional: `status` (`completed` or `failed`, defaults to `completed`) |
| 4085 | - completed result: |
| 4086 | - either top-level `shot_id` + a public `asset_urls` entry (preferred) or `result_url` |
| 4087 | - or `result={"shot_id": 8, "asset_urls": {"primary": {"url": "https://..."}}}` / `result_url` |
| 4088 | - injection routing: |
| 4089 | - `session_key`, `channel`, `chat_id` |
| 4090 | """ |
| 4091 | tool = GenerateEchoShotTool(workspace=workspace, tools_config=tools_config) |
| 4092 | return tool.apply_echo_callback_payload(callback_payload) |
| 4093 | |
| 4094 | |
| 4095 | def apply_merge_shot_callback( |
| 4096 | workspace: Path, |
| 4097 | callback_payload: dict[str, Any], |
| 4098 | *, |
| 4099 | tools_config: Any | None = None, |
| 4100 | ) -> dict[str, Any]: |
| 4101 | """Apply one merge_shot remote callback to the director workspace. |
| 4102 | |
| 4103 | Expected callback payload shape: |
| 4104 | - required: `work_id`, `job_id` |
| 4105 | - optional: `status` (`completed` or `failed`, defaults to `completed`) |
| 4106 | - completed result: |
| 4107 | - preferred: the first public URL in top-level or `result.asset_urls` |
| 4108 | - fallback: `artifact_path`, `artifact_url`, `result_url`, or `result={...}` equivalents |
| 4109 | - injection routing: |
| 4110 | - `session_key`, `channel`, `chat_id` |
| 4111 | """ |
| 4112 | tool = MergeShotTool(workspace=workspace, tools_config=tools_config) |
| 4113 | return tool.apply_merge_callback_payload(callback_payload) |
| 4114 |