| 1 | //! Last-round coverage floor for compaction replacement history. |
| 2 | //! |
| 3 | //! Compaction may summarize older turns, but the latest user round (user |
| 4 | //! text plus following assistant/tool results) must survive verbatim, |
| 5 | //! bounded, or the pass is refused. See [`SURVIVAL_CONTRACT.md`]. |
| 6 | |
| 7 | use anyhow::Result; |
| 8 | use std::collections::HashSet; |
| 9 | |
| 10 | use codewhale_models::{ContentBlock, Message, SystemPrompt}; |
| 11 | |
| 12 | use super::{ |
| 13 | compaction_checkpoint_message, is_compaction_checkpoint_message, retained_user_messages, |
| 14 | truncate_retained_block, user_text_of, |
| 15 | }; |
| 16 | |
| 17 | const LAST_ROUND_TOOL_RESULT_MAX_CHARS: usize = 8 * 1024; |
| 18 | |
| 19 | #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)] |
| 20 | pub enum CompactionPath { |
| 21 | #[default] |
| 22 | Summary, |
| 23 | PruneOnly, |
| 24 | } |
| 25 | |
| 26 | #[derive(Debug, Clone, PartialEq, Eq, Default)] |
| 27 | pub struct CompactionCoverage { |
| 28 | pub path: CompactionPath, |
| 29 | pub last_round_messages: usize, |
| 30 | pub last_round_tool_results: usize, |
| 31 | pub last_round_assistant: bool, |
| 32 | pub dropped_messages: usize, |
| 33 | pub anchors_chars: usize, |
| 34 | /// Effective `[compaction] retained_user_message_tokens` budget this pass |
| 35 | /// spent on verbatim user messages (#5956). `0` on the prune-only path, |
| 36 | /// which never builds a replacement history. |
| 37 | pub retained_user_message_tokens: usize, |
| 38 | /// Whether `[compaction] summary_instructions` was appended to the |
| 39 | /// summarizer prompt on this pass (#5956). |
| 40 | pub operator_instructions_applied: bool, |
| 41 | } |
| 42 | |
| 43 | impl CompactionCoverage { |
| 44 | #[must_use] |
| 45 | pub fn receipt_clause(&self) -> String { |
| 46 | let path = match self.path { |
| 47 | CompactionPath::Summary => "summary", |
| 48 | CompactionPath::PruneOnly => "prune-only", |
| 49 | }; |
| 50 | let assistant = if self.last_round_assistant { |
| 51 | ", assistant" |
| 52 | } else { |
| 53 | "" |
| 54 | }; |
| 55 | let mut clause = format!( |
| 56 | "{path}; last round kept: {} messages ({} tool results{assistant})", |
| 57 | self.last_round_messages, self.last_round_tool_results |
| 58 | ); |
| 59 | if self.anchors_chars > 0 { |
| 60 | clause.push_str(&format!("; anchors {} chars", self.anchors_chars)); |
| 61 | } |
| 62 | // Name the tuning knobs so an operator who set them can tell they took |
| 63 | // effect without reading the log (#5956). The prune-only path builds no |
| 64 | // replacement history, so it reports no budget. |
| 65 | if self.retained_user_message_tokens > 0 { |
| 66 | clause.push_str(&format!( |
| 67 | "; verbatim user budget {} tokens", |
| 68 | self.retained_user_message_tokens |
| 69 | )); |
| 70 | } |
| 71 | if self.operator_instructions_applied { |
| 72 | clause.push_str("; operator instructions applied"); |
| 73 | } |
| 74 | clause |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 79 | pub struct LastCompactionSnapshot { |
| 80 | pub auto: bool, |
| 81 | pub coverage: CompactionCoverage, |
| 82 | pub messages_before: usize, |
| 83 | pub messages_after: usize, |
| 84 | } |
| 85 | |
| 86 | #[derive(Debug, Clone, PartialEq, Eq, Default)] |
| 87 | pub struct CompactionKeep { |
| 88 | pub has_checkpoint: bool, |
| 89 | pub last_round_messages: usize, |
| 90 | pub last_round_tool_results: usize, |
| 91 | pub last_round_assistant: bool, |
| 92 | } |
| 93 | |
| 94 | #[must_use] |
| 95 | pub fn inspect_compaction_keep(messages: &[Message]) -> CompactionKeep { |
| 96 | let last_round = last_round_slice(messages); |
| 97 | CompactionKeep { |
| 98 | has_checkpoint: messages.iter().any(is_compaction_checkpoint_message), |
| 99 | last_round_messages: last_round.len(), |
| 100 | last_round_tool_results: last_round.iter().flat_map(tool_result_ids).count(), |
| 101 | last_round_assistant: last_round |
| 102 | .iter() |
| 103 | .any(|message| message.role.is_assistant_like()), |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | #[must_use] |
| 108 | pub fn pinned_anchors_text(workspace: Option<&std::path::Path>) -> Option<String> { |
| 109 | let workspace = workspace?; |
| 110 | let primary = workspace.join(".codewhale").join("anchors.md"); |
| 111 | let path = if primary.exists() { |
| 112 | primary |
| 113 | } else { |
| 114 | workspace.join(".deepseek").join("anchors.md") |
| 115 | }; |
| 116 | std::fs::read_to_string(path) |
| 117 | .ok() |
| 118 | .map(|contents| contents.trim().to_string()) |
| 119 | .filter(|contents| !contents.is_empty()) |
| 120 | } |
| 121 | |
| 122 | fn is_plain_user_text(message: &Message) -> bool { |
| 123 | !is_compaction_checkpoint_message(message) |
| 124 | && !crate::runtime_handoff::is_runtime_owned_user_message(message) |
| 125 | && user_text_of(message).is_some() |
| 126 | } |
| 127 | |
| 128 | fn user_prompt_text_of(message: &Message) -> Option<String> { |
| 129 | is_plain_user_text(message) |
| 130 | .then(|| user_text_of(message)) |
| 131 | .flatten() |
| 132 | } |
| 133 | |
| 134 | fn last_plain_user_index(messages: &[Message], end: usize) -> Option<usize> { |
| 135 | messages[..end] |
| 136 | .iter() |
| 137 | .enumerate() |
| 138 | .rev() |
| 139 | .find_map(|(idx, message)| is_plain_user_text(message).then_some(idx)) |
| 140 | } |
| 141 | |
| 142 | fn slice_has_tool_result(messages: &[Message], start: usize) -> bool { |
| 143 | messages[start..].iter().any(|message| { |
| 144 | message |
| 145 | .content |
| 146 | .iter() |
| 147 | .any(|block| matches!(block, ContentBlock::ToolResult { .. })) |
| 148 | }) |
| 149 | } |
| 150 | |
| 151 | #[must_use] |
| 152 | pub(crate) fn last_round_start(messages: &[Message]) -> usize { |
| 153 | let Some(last_user) = last_plain_user_index(messages, messages.len()) else { |
| 154 | return 0; |
| 155 | }; |
| 156 | if slice_has_tool_result(messages, last_user) { |
| 157 | return last_user; |
| 158 | } |
| 159 | // Trailing toolless user/assistant turns still need the previous |
| 160 | // tool-bearing round; otherwise those results vanish behind the summary. |
| 161 | // If no tool round exists, keep only the latest user turn so chat-only |
| 162 | // sessions can still summarize older text. |
| 163 | let mut candidate = last_user; |
| 164 | loop { |
| 165 | let Some(prev) = last_plain_user_index(messages, candidate) else { |
| 166 | return last_user; |
| 167 | }; |
| 168 | if slice_has_tool_result(messages, prev) { |
| 169 | return prev; |
| 170 | } |
| 171 | candidate = prev; |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | #[must_use] |
| 176 | pub(crate) fn last_round_range(messages: &[Message]) -> (usize, usize) { |
| 177 | let start = last_round_start(messages).min(messages.len()); |
| 178 | // A previous checkpoint can sit in the middle of an uninterrupted task. |
| 179 | // Stopping at that marker hid every tool step after the first compact |
| 180 | // from the next pass's survival checks. |
| 181 | let end = messages.len().saturating_sub(usize::from( |
| 182 | messages |
| 183 | .last() |
| 184 | .is_some_and(is_compaction_checkpoint_message), |
| 185 | )); |
| 186 | (start, end) |
| 187 | } |
| 188 | |
| 189 | /// How many messages of the open round sit in `messages` before a checkpoint. |
| 190 | #[must_use] |
| 191 | pub fn last_round_kept_count(messages: &[Message]) -> Option<usize> { |
| 192 | let checkpoint = messages |
| 193 | .iter() |
| 194 | .rposition(is_compaction_checkpoint_message)?; |
| 195 | if checkpoint == 0 { |
| 196 | return None; |
| 197 | } |
| 198 | let start = last_round_start(&messages[..checkpoint]); |
| 199 | Some(checkpoint.saturating_sub(start)) |
| 200 | } |
| 201 | |
| 202 | fn last_round_slice(messages: &[Message]) -> &[Message] { |
| 203 | let (start, end) = last_round_range(messages); |
| 204 | &messages[start..end] |
| 205 | } |
| 206 | |
| 207 | /// Retain the current user instructions and the two most recent tool |
| 208 | /// exchanges. A user round can contain thousands of steps: retaining that |
| 209 | /// entire round forever makes a continuous task impossible to compact. |
| 210 | /// Older completed exchanges are covered by the summary and durable history. |
| 211 | /// A split is legal only when all preceding tool calls have their results. |
| 212 | fn protected_last_round(messages: &[Message]) -> Vec<&Message> { |
| 213 | let round = last_round_slice(messages); |
| 214 | let mut pending = HashSet::new(); |
| 215 | let mut boundaries = Vec::new(); |
| 216 | for (idx, message) in round.iter().enumerate() { |
| 217 | let calls = tool_use_ids(message); |
| 218 | if !calls.is_empty() && pending.is_empty() { |
| 219 | boundaries.push(idx); |
| 220 | } |
| 221 | pending.extend(calls); |
| 222 | for id in tool_result_ids(message) { |
| 223 | pending.remove(&id); |
| 224 | } |
| 225 | } |
| 226 | let start = if boundaries.len() > 2 { |
| 227 | boundaries[boundaries.len() - 2] |
| 228 | } else { |
| 229 | 0 |
| 230 | }; |
| 231 | round |
| 232 | .iter() |
| 233 | .enumerate() |
| 234 | .filter_map(|(idx, message)| { |
| 235 | (!is_compaction_checkpoint_message(message) |
| 236 | && (idx >= start || is_plain_user_text(message))) |
| 237 | .then_some(message) |
| 238 | }) |
| 239 | .collect() |
| 240 | } |
| 241 | |
| 242 | pub(super) fn replacement_messages( |
| 243 | messages: &[Message], |
| 244 | retained_user_message_tokens: usize, |
| 245 | ) -> Vec<Message> { |
| 246 | let (start, _) = last_round_range(messages); |
| 247 | let mut retained = retained_user_messages(&messages[..start], retained_user_message_tokens); |
| 248 | let round = protected_last_round(messages) |
| 249 | .into_iter() |
| 250 | .cloned() |
| 251 | .collect::<Vec<_>>(); |
| 252 | retained.extend(bound_last_round(&round)); |
| 253 | // The Operate contract applies to the current tool loop as well as later |
| 254 | // turns. It must survive compaction even when old-user retention is full. |
| 255 | let current_contract = messages |
| 256 | .iter() |
| 257 | .rev() |
| 258 | .find(|message| crate::runtime_handoff::is_current_operate_contract_message(message)); |
| 259 | let contract = current_contract.or_else(|| { |
| 260 | messages |
| 261 | .iter() |
| 262 | .rev() |
| 263 | .find(|message| crate::runtime_handoff::is_operate_contract_message(message)) |
| 264 | }); |
| 265 | if let Some(contract) = contract { |
| 266 | retained.retain(|message| !crate::runtime_handoff::is_operate_contract_message(message)); |
| 267 | retained.insert(0, contract.clone()); |
| 268 | } |
| 269 | retained |
| 270 | } |
| 271 | |
| 272 | pub(super) fn bound_last_round(messages: &[Message]) -> Vec<Message> { |
| 273 | let mut round = messages.to_vec(); |
| 274 | for message in &mut round { |
| 275 | for block in &mut message.content { |
| 276 | if let ContentBlock::ToolResult { |
| 277 | content, |
| 278 | content_blocks, |
| 279 | .. |
| 280 | } = block |
| 281 | && truncate_retained_block("tool result", content, LAST_ROUND_TOOL_RESULT_MAX_CHARS) |
| 282 | { |
| 283 | *content_blocks = None; |
| 284 | } |
| 285 | } |
| 286 | } |
| 287 | round |
| 288 | } |
| 289 | |
| 290 | fn tool_result_ids(message: &Message) -> Vec<String> { |
| 291 | message |
| 292 | .content |
| 293 | .iter() |
| 294 | .filter_map(|block| match block { |
| 295 | ContentBlock::ToolResult { tool_use_id, .. } => Some(tool_use_id.clone()), |
| 296 | _ => None, |
| 297 | }) |
| 298 | .collect() |
| 299 | } |
| 300 | |
| 301 | fn has_tool_result_id(message: &Message, id: &str) -> bool { |
| 302 | message.content.iter().any(|block| { |
| 303 | matches!( |
| 304 | block, |
| 305 | ContentBlock::ToolResult { tool_use_id, .. } if tool_use_id == id |
| 306 | ) |
| 307 | }) |
| 308 | } |
| 309 | |
| 310 | fn tool_use_ids(message: &Message) -> Vec<String> { |
| 311 | message |
| 312 | .content |
| 313 | .iter() |
| 314 | .filter_map(|block| match block { |
| 315 | ContentBlock::ToolUse { id, .. } => Some(id.clone()), |
| 316 | _ => None, |
| 317 | }) |
| 318 | .collect() |
| 319 | } |
| 320 | |
| 321 | fn has_tool_use_id(message: &Message, id: &str) -> bool { |
| 322 | message.content.iter().any(|block| { |
| 323 | matches!( |
| 324 | block, |
| 325 | ContentBlock::ToolUse { id: seen, .. } if seen == id |
| 326 | ) |
| 327 | }) |
| 328 | } |
| 329 | |
| 330 | fn assistant_text_of(message: &Message) -> Option<String> { |
| 331 | if !message.role.is_assistant_like() { |
| 332 | return None; |
| 333 | } |
| 334 | let text = message |
| 335 | .content |
| 336 | .iter() |
| 337 | .filter_map(|block| match block { |
| 338 | ContentBlock::Text { text, .. } => Some(text.as_str()), |
| 339 | _ => None, |
| 340 | }) |
| 341 | .collect::<Vec<_>>() |
| 342 | .join("\n"); |
| 343 | let text = text.trim(); |
| 344 | (!text.is_empty()).then(|| text.to_string()) |
| 345 | } |
| 346 | |
| 347 | /// A retained copy may be truncated (`bound_last_round` caps oversized blocks), |
| 348 | /// so a prefix either way counts as survival -- but nothing weaker does. |
| 349 | fn survives(text: &str, replacement: &[Message], of: fn(&Message) -> Option<String>) -> bool { |
| 350 | replacement.iter().any(|message| { |
| 351 | of(message) |
| 352 | .is_some_and(|kept| kept == text || text.starts_with(&kept) || kept.starts_with(text)) |
| 353 | }) |
| 354 | } |
| 355 | |
| 356 | pub(crate) fn validate_last_round_coverage( |
| 357 | original: &[Message], |
| 358 | replacement: &[Message], |
| 359 | ) -> Result<()> { |
| 360 | let last_round = protected_last_round(original); |
| 361 | if last_round.is_empty() { |
| 362 | return Ok(()); |
| 363 | } |
| 364 | // Every user turn in the round, not the first one `find_map` happens to |
| 365 | // reach. `last_round_start` walks back past a toolless tail to the previous |
| 366 | // tool-bearing turn, so the round routinely spans two user messages -- and |
| 367 | // checking only the earliest let a rewrite drop the *latest* one, which is |
| 368 | // the turn this whole contract exists to keep. |
| 369 | for text in last_round.iter().copied().filter_map(user_prompt_text_of) { |
| 370 | if !survives(&text, replacement, user_prompt_text_of) { |
| 371 | anyhow::bail!( |
| 372 | "Compaction coverage floor: a last-round user message was dropped; history was not replaced." |
| 373 | ); |
| 374 | } |
| 375 | } |
| 376 | for id in last_round.iter().copied().flat_map(tool_result_ids) { |
| 377 | if !replacement |
| 378 | .iter() |
| 379 | .any(|message| has_tool_result_id(message, &id)) |
| 380 | { |
| 381 | anyhow::bail!( |
| 382 | "Compaction coverage floor: last-round tool result {id} was dropped; history was not replaced." |
| 383 | ); |
| 384 | } |
| 385 | } |
| 386 | // The call, not just its result. Keeping a tool_result whose tool_use was |
| 387 | // summarized away leaves an orphaned result that providers reject outright. |
| 388 | for id in last_round.iter().copied().flat_map(tool_use_ids) { |
| 389 | if !replacement |
| 390 | .iter() |
| 391 | .any(|message| has_tool_use_id(message, &id)) |
| 392 | { |
| 393 | anyhow::bail!( |
| 394 | "Compaction coverage floor: last-round tool call {id} was dropped; history was not replaced." |
| 395 | ); |
| 396 | } |
| 397 | } |
| 398 | // Match the assistant's actual output. An existential "some assistant |
| 399 | // message survived" check passed on a replacement whose only assistant |
| 400 | // message was the summary the rewrite had just written. |
| 401 | for text in last_round.iter().copied().filter_map(assistant_text_of) { |
| 402 | if !survives(&text, replacement, assistant_text_of) { |
| 403 | anyhow::bail!( |
| 404 | "Compaction coverage floor: last-round assistant output was dropped; history was not replaced." |
| 405 | ); |
| 406 | } |
| 407 | } |
| 408 | if last_round |
| 409 | .iter() |
| 410 | .any(|message| message.role.is_assistant_like()) |
| 411 | && !replacement |
| 412 | .iter() |
| 413 | .any(|message| message.role.is_assistant_like()) |
| 414 | { |
| 415 | anyhow::bail!( |
| 416 | "Compaction coverage floor: last-round assistant output was dropped; history was not replaced." |
| 417 | ); |
| 418 | } |
| 419 | Ok(()) |
| 420 | } |
| 421 | |
| 422 | pub(crate) fn require_text_survives( |
| 423 | replacement: &[Message], |
| 424 | needle: &str, |
| 425 | label: &str, |
| 426 | ) -> Result<()> { |
| 427 | let needle = needle.trim(); |
| 428 | if needle.is_empty() { |
| 429 | return Ok(()); |
| 430 | } |
| 431 | let kept = replacement.iter().any(|message| { |
| 432 | message.content.iter().any(|block| match block { |
| 433 | ContentBlock::Text { text, .. } => text.contains(needle), |
| 434 | ContentBlock::ToolResult { content, .. } => content.contains(needle), |
| 435 | _ => false, |
| 436 | }) |
| 437 | }); |
| 438 | if !kept { |
| 439 | anyhow::bail!("Compaction coverage floor: {label} was dropped; history was not replaced."); |
| 440 | } |
| 441 | Ok(()) |
| 442 | } |
| 443 | |
| 444 | pub(crate) fn validate_survival_contract( |
| 445 | original: &[Message], |
| 446 | replacement: &[Message], |
| 447 | anchors: Option<&str>, |
| 448 | ) -> Result<()> { |
| 449 | validate_last_round_coverage(original, replacement)?; |
| 450 | let checkpoints = replacement |
| 451 | .iter() |
| 452 | .filter(|message| is_compaction_checkpoint_message(message)) |
| 453 | .count(); |
| 454 | if checkpoints == 0 { |
| 455 | anyhow::bail!( |
| 456 | "Compaction coverage floor: checkpoint receipt was dropped; history was not replaced." |
| 457 | ); |
| 458 | } |
| 459 | if checkpoints > 1 { |
| 460 | anyhow::bail!( |
| 461 | "Compaction coverage floor: prior summaries were duplicated; history was not replaced." |
| 462 | ); |
| 463 | } |
| 464 | if let Some(anchors) = anchors { |
| 465 | require_text_survives(replacement, anchors, "pinned /anchor text")?; |
| 466 | } |
| 467 | Ok(()) |
| 468 | } |
| 469 | |
| 470 | pub(super) fn measure_coverage( |
| 471 | original: &[Message], |
| 472 | replacement: &[Message], |
| 473 | path: CompactionPath, |
| 474 | anchors_chars: usize, |
| 475 | ) -> CompactionCoverage { |
| 476 | let last_round = last_round_slice(replacement); |
| 477 | CompactionCoverage { |
| 478 | path, |
| 479 | last_round_messages: last_round.len(), |
| 480 | last_round_tool_results: last_round.iter().flat_map(tool_result_ids).count(), |
| 481 | last_round_assistant: last_round |
| 482 | .iter() |
| 483 | .any(|message| message.role.is_assistant_like()), |
| 484 | dropped_messages: original.len().saturating_sub(replacement.len()), |
| 485 | anchors_chars, |
| 486 | // Tuning provenance is owned by the caller that holds the |
| 487 | // `CompactionConfig`; measurement over two message lists cannot know it. |
| 488 | retained_user_message_tokens: 0, |
| 489 | operator_instructions_applied: false, |
| 490 | } |
| 491 | } |
| 492 | |
| 493 | /// Build the post-compaction history: recent plain user messages kept |
| 494 | /// verbatim within `retained_user_message_tokens`, the bounded last round, and |
| 495 | /// the checkpoint. The budget is `[compaction] retained_user_message_tokens` |
| 496 | /// (#5956); it was a hard-coded 20 000 before that key existed. |
| 497 | pub(super) fn build_replacement_history( |
| 498 | messages: &[Message], |
| 499 | checkpoint_text: &str, |
| 500 | anchors: Option<&str>, |
| 501 | retained_user_message_tokens: usize, |
| 502 | ) -> Result<Vec<Message>> { |
| 503 | let mut retained = replacement_messages(messages, retained_user_message_tokens); |
| 504 | retained.push(compaction_checkpoint_message(&SystemPrompt::Text( |
| 505 | checkpoint_text.to_string(), |
| 506 | ))); |
| 507 | validate_survival_contract(messages, &retained, anchors)?; |
| 508 | Ok(retained) |
| 509 | } |
| 510 | |
| 511 | #[cfg(test)] |
| 512 | mod tests { |
| 513 | use super::*; |
| 514 | use crate::compaction::{COMPACTION_SUMMARY_MARKER, compaction_checkpoint_message}; |
| 515 | use codewhale_models::{ContentBlock, Role}; |
| 516 | use serde_json::json; |
| 517 | |
| 518 | fn msg(role: &str, text: &str) -> Message { |
| 519 | Message { |
| 520 | role: Role::from(role), |
| 521 | content: vec![ContentBlock::Text { |
| 522 | text: text.to_string(), |
| 523 | cache_control: None, |
| 524 | }], |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | fn tool_use(id: &str, name: &str, input: serde_json::Value) -> Message { |
| 529 | Message { |
| 530 | role: Role::Assistant, |
| 531 | content: vec![ContentBlock::ToolUse { |
| 532 | id: id.to_string(), |
| 533 | name: name.to_string(), |
| 534 | input, |
| 535 | caller: None, |
| 536 | thought_signature: None, |
| 537 | }], |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | fn tool_result(id: &str, content: &str) -> Message { |
| 542 | Message { |
| 543 | role: Role::User, |
| 544 | content: vec![ContentBlock::ToolResult { |
| 545 | tool_use_id: id.to_string(), |
| 546 | content: content.to_string(), |
| 547 | is_error: None, |
| 548 | content_blocks: None, |
| 549 | }], |
| 550 | } |
| 551 | } |
| 552 | |
| 553 | fn checkpoint(summary: &str) -> Message { |
| 554 | compaction_checkpoint_message(&SystemPrompt::Text(format!( |
| 555 | "{COMPACTION_SUMMARY_MARKER}: {summary}" |
| 556 | ))) |
| 557 | } |
| 558 | |
| 559 | #[test] |
| 560 | fn coverage_floor_rejects_a_replacement_that_drops_last_round_tools() { |
| 561 | let original = vec![ |
| 562 | msg("user", "Run the failing test."), |
| 563 | msg("assistant", "Running."), |
| 564 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 565 | tool_result("live", "test session_store::roundtrip ... FAILED"), |
| 566 | ]; |
| 567 | let gutting = vec![ |
| 568 | msg("user", "Run the failing test."), |
| 569 | checkpoint("and kept going"), |
| 570 | ]; |
| 571 | let error = validate_last_round_coverage(&original, &gutting) |
| 572 | .expect_err("dropping the last tool result must fail the coverage floor"); |
| 573 | assert!(error.to_string().contains("tool result live"), "{error}"); |
| 574 | assert!(validate_last_round_coverage(&original, &original).is_ok()); |
| 575 | } |
| 576 | |
| 577 | #[test] |
| 578 | fn coverage_floor_rejects_a_replacement_that_drops_last_round_assistant() { |
| 579 | let original = vec![ |
| 580 | msg("user", "What failed?"), |
| 581 | msg("assistant", "session_store::roundtrip panics on reload."), |
| 582 | ]; |
| 583 | let error = validate_last_round_coverage(&original, &[msg("user", "What failed?")]) |
| 584 | .expect_err("dropping last-round assistant text must fail closed"); |
| 585 | assert!(error.to_string().contains("assistant"), "{error}"); |
| 586 | } |
| 587 | |
| 588 | /// The round spans the tool-bearing turn *and* the toolless tail after it, |
| 589 | /// because `last_round_start` walks back for the tools. Checking only the |
| 590 | /// first user text it found meant a rewrite could keep the older question |
| 591 | /// and drop the one the person actually just asked. |
| 592 | #[test] |
| 593 | fn coverage_floor_rejects_a_replacement_that_drops_the_latest_user_turn() { |
| 594 | let original = vec![ |
| 595 | msg("user", "Run the suite."), |
| 596 | msg("assistant", "Running."), |
| 597 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 598 | tool_result("live", "ok"), |
| 599 | msg("user", "Now ship it."), |
| 600 | msg("assistant", "Shipping."), |
| 601 | ]; |
| 602 | assert_eq!(last_round_start(&original), 0, "round must span both turns"); |
| 603 | |
| 604 | let drops_latest = vec![ |
| 605 | msg("user", "Run the suite."), |
| 606 | msg("assistant", "Running."), |
| 607 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 608 | tool_result("live", "ok"), |
| 609 | checkpoint("then shipped"), |
| 610 | ]; |
| 611 | let error = validate_last_round_coverage(&original, &drops_latest) |
| 612 | .expect_err("dropping the latest user turn must fail the coverage floor"); |
| 613 | assert!(error.to_string().contains("user message"), "{error}"); |
| 614 | assert!(validate_last_round_coverage(&original, &original).is_ok()); |
| 615 | } |
| 616 | |
| 617 | /// A surviving `tool_result` whose `tool_use` was summarized away is an |
| 618 | /// orphan the provider rejects, so the floor must cover the call too. |
| 619 | #[test] |
| 620 | fn coverage_floor_rejects_a_replacement_that_drops_the_tool_call() { |
| 621 | let original = vec![ |
| 622 | msg("user", "Run the failing test."), |
| 623 | msg("assistant", "Running."), |
| 624 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 625 | tool_result("live", "FAILED"), |
| 626 | ]; |
| 627 | let orphaned = vec![ |
| 628 | msg("user", "Run the failing test."), |
| 629 | msg("assistant", "Running."), |
| 630 | tool_result("live", "FAILED"), |
| 631 | checkpoint("and it failed"), |
| 632 | ]; |
| 633 | let error = validate_last_round_coverage(&original, &orphaned) |
| 634 | .expect_err("dropping the tool call must fail the coverage floor"); |
| 635 | assert!(error.to_string().contains("tool call live"), "{error}"); |
| 636 | } |
| 637 | |
| 638 | /// "Some assistant message survived" was satisfied by the summary the |
| 639 | /// rewrite had just written, so the round's real output could vanish. |
| 640 | #[test] |
| 641 | fn coverage_floor_rejects_assistant_output_replaced_by_a_summary() { |
| 642 | let original = vec![ |
| 643 | msg("user", "What failed?"), |
| 644 | msg("assistant", "session_store::roundtrip panics on reload."), |
| 645 | ]; |
| 646 | let summarized = vec![ |
| 647 | msg("user", "What failed?"), |
| 648 | msg("assistant", "Earlier we discussed several test failures."), |
| 649 | ]; |
| 650 | let error = validate_last_round_coverage(&original, &summarized) |
| 651 | .expect_err("substituting a summary for the round's output must fail closed"); |
| 652 | assert!(error.to_string().contains("assistant"), "{error}"); |
| 653 | } |
| 654 | |
| 655 | #[test] |
| 656 | fn survival_contract_rejects_dropped_anchors_and_receipts() { |
| 657 | let original = vec![msg("user", "Keep the pin."), msg("assistant", "Anchored.")]; |
| 658 | let without_receipt = vec![msg("user", "Keep the pin."), msg("assistant", "Anchored.")]; |
| 659 | let error = validate_survival_contract(&original, &without_receipt, Some("ship 0.9.12")) |
| 660 | .expect_err("missing checkpoint receipt must fail closed"); |
| 661 | assert!(error.to_string().contains("receipt"), "{error}"); |
| 662 | |
| 663 | let without_anchor = vec![ |
| 664 | msg("user", "Keep the pin."), |
| 665 | msg("assistant", "Anchored."), |
| 666 | checkpoint("progress without the pin"), |
| 667 | ]; |
| 668 | let error = validate_survival_contract(&original, &without_anchor, Some("ship 0.9.12")) |
| 669 | .expect_err("dropped /anchor text must fail closed"); |
| 670 | assert!(error.to_string().contains("anchor"), "{error}"); |
| 671 | } |
| 672 | |
| 673 | #[test] |
| 674 | fn last_round_starts_at_the_latest_plain_user_message() { |
| 675 | let messages = vec![ |
| 676 | msg("user", "older"), |
| 677 | msg("assistant", "working"), |
| 678 | tool_result("old", "stale"), |
| 679 | msg("user", "Run the suite now."), |
| 680 | msg("assistant", "Rerunning."), |
| 681 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 682 | tool_result("live", "ok"), |
| 683 | ]; |
| 684 | assert_eq!(last_round_start(&messages), 3); // last user with tools |
| 685 | let (start, end) = last_round_range(&messages); |
| 686 | let kept = bound_last_round(&messages[start..end]); |
| 687 | assert!(kept.iter().any(|message| { |
| 688 | message.content.iter().any(|block| { |
| 689 | matches!( |
| 690 | block, |
| 691 | ContentBlock::ToolResult { tool_use_id, content, .. } |
| 692 | if tool_use_id == "live" && content == "ok" |
| 693 | ) |
| 694 | }) |
| 695 | })); |
| 696 | } |
| 697 | |
| 698 | #[test] |
| 699 | fn second_compaction_keeps_long_user_question_and_tool_pair_past_retention_budget() { |
| 700 | const PRODUCTION_MIN_RETAINED_TOKENS: usize = 2_000; |
| 701 | let long_question = format!( |
| 702 | "{}?", |
| 703 | "Analyze every step of this case carefully. ".repeat(400) |
| 704 | ); |
| 705 | assert!(long_question.len() > PRODUCTION_MIN_RETAINED_TOKENS * 3); |
| 706 | let original = vec![ |
| 707 | msg("user", &long_question), |
| 708 | tool_use("call_1", "Bash", json!({"command": "echo ready"})), |
| 709 | tool_result("call_1", "ready"), |
| 710 | ]; |
| 711 | let first_summary = |
| 712 | crate::compaction::build_compaction_summary_block_text("First pass complete", ""); |
| 713 | let mut first = build_replacement_history( |
| 714 | &original, |
| 715 | &first_summary, |
| 716 | None, |
| 717 | PRODUCTION_MIN_RETAINED_TOKENS, |
| 718 | ) |
| 719 | .expect("first compaction"); |
| 720 | crate::runtime_handoff::replace_agent_topology_checkpoint(&mut first, &[]); |
| 721 | assert_eq!(last_round_start(&first), 0); |
| 722 | let topology = first |
| 723 | .iter() |
| 724 | .find(|message| crate::runtime_handoff::is_agent_topology_checkpoint(message)) |
| 725 | .expect("first compaction topology checkpoint") |
| 726 | .clone(); |
| 727 | assert!( |
| 728 | crate::compaction::retained_user_messages( |
| 729 | std::slice::from_ref(&topology), |
| 730 | PRODUCTION_MIN_RETAINED_TOKENS, |
| 731 | ) |
| 732 | .is_empty(), |
| 733 | "runtime topology must not consume the older-user retention budget" |
| 734 | ); |
| 735 | |
| 736 | let second_summary = |
| 737 | crate::compaction::build_compaction_summary_block_text("Second pass complete", ""); |
| 738 | let second = build_replacement_history( |
| 739 | &first, |
| 740 | &second_summary, |
| 741 | None, |
| 742 | PRODUCTION_MIN_RETAINED_TOKENS, |
| 743 | ) |
| 744 | .expect("second compaction"); |
| 745 | assert!( |
| 746 | second.iter().any(|message| { |
| 747 | user_text_of(message).as_deref() == Some(long_question.as_str()) |
| 748 | }) |
| 749 | ); |
| 750 | assert!( |
| 751 | second |
| 752 | .iter() |
| 753 | .any(|message| has_tool_use_id(message, "call_1")) |
| 754 | ); |
| 755 | assert!( |
| 756 | second |
| 757 | .iter() |
| 758 | .any(|message| has_tool_result_id(message, "call_1")) |
| 759 | ); |
| 760 | let without_question = second |
| 761 | .iter() |
| 762 | .filter(|message| user_text_of(message).as_deref() != Some(long_question.as_str())) |
| 763 | .cloned() |
| 764 | .collect::<Vec<_>>(); |
| 765 | assert!(validate_last_round_coverage(&first, &without_question).is_err()); |
| 766 | } |
| 767 | |
| 768 | #[test] |
| 769 | fn runtime_text_cannot_satisfy_real_user_coverage() { |
| 770 | let runtime = crate::runtime_handoff::operate_contract_runtime_message(); |
| 771 | let copied_text = user_text_of(&runtime).expect("runtime text"); |
| 772 | let original = vec![msg("user", &copied_text), msg("assistant", "Acknowledged")]; |
| 773 | let replacement = vec![runtime, msg("assistant", "Acknowledged")]; |
| 774 | assert!( |
| 775 | validate_last_round_coverage(&original, &replacement).is_err(), |
| 776 | "runtime-owned text must not stand in for the user's actual prompt" |
| 777 | ); |
| 778 | } |
| 779 | |
| 780 | #[test] |
| 781 | fn operate_contract_survives_compaction_without_spending_user_budget() { |
| 782 | let contract = crate::runtime_handoff::operate_contract_runtime_message(); |
| 783 | let original = vec![ |
| 784 | contract.clone(), |
| 785 | msg("user", "First task"), |
| 786 | msg("assistant", "Working"), |
| 787 | msg("user", "Continue the same task"), |
| 788 | msg("assistant", "Continuing"), |
| 789 | ]; |
| 790 | let replaced = build_replacement_history( |
| 791 | &original, |
| 792 | &format!("{COMPACTION_SUMMARY_MARKER}: work continues"), |
| 793 | None, |
| 794 | 1, |
| 795 | ) |
| 796 | .expect("compaction must retain the active Operate contract"); |
| 797 | assert_eq!(replaced.first(), Some(&contract)); |
| 798 | assert_eq!( |
| 799 | replaced |
| 800 | .iter() |
| 801 | .filter(|message| **message == contract) |
| 802 | .count(), |
| 803 | 1 |
| 804 | ); |
| 805 | } |
| 806 | |
| 807 | #[test] |
| 808 | fn compaction_prefers_current_operate_contract_over_legacy() { |
| 809 | let legacy = crate::runtime_handoff::legacy_operate_contract_runtime_message(); |
| 810 | let current = crate::runtime_handoff::operate_contract_runtime_message(); |
| 811 | let original = vec![ |
| 812 | legacy.clone(), |
| 813 | current.clone(), |
| 814 | msg("user", "Continue"), |
| 815 | msg("assistant", "Working"), |
| 816 | ]; |
| 817 | let replaced = build_replacement_history( |
| 818 | &original, |
| 819 | &format!("{COMPACTION_SUMMARY_MARKER}: work continues"), |
| 820 | None, |
| 821 | 1, |
| 822 | ) |
| 823 | .expect("current contract must survive compaction"); |
| 824 | assert_eq!(replaced.first(), Some(¤t)); |
| 825 | assert!(!replaced.contains(&legacy)); |
| 826 | assert_eq!( |
| 827 | replaced |
| 828 | .iter() |
| 829 | .filter(|message| crate::runtime_handoff::is_operate_contract_message(message)) |
| 830 | .count(), |
| 831 | 1 |
| 832 | ); |
| 833 | } |
| 834 | |
| 835 | #[test] |
| 836 | fn last_round_walks_back_through_toolless_tails_to_the_tool_round() { |
| 837 | let original = vec![ |
| 838 | msg("user", "Run the failing test."), |
| 839 | msg("assistant", "Running."), |
| 840 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 841 | tool_result("live", "test session_store::roundtrip ... FAILED"), |
| 842 | msg("user", "ok thanks"), |
| 843 | msg("assistant", "you're welcome"), |
| 844 | msg("user", "one more thing"), |
| 845 | msg("assistant", "sure"), |
| 846 | ]; |
| 847 | assert_eq!(last_round_start(&original), 0); |
| 848 | let next = format!("{COMPACTION_SUMMARY_MARKER}: keep the failing test result"); |
| 849 | let replaced = build_replacement_history( |
| 850 | &original, |
| 851 | &next, |
| 852 | None, |
| 853 | crate::compaction::COMPACT_RETAINED_USER_MESSAGE_MAX_TOKENS, |
| 854 | ) |
| 855 | .expect("toolless tails must not drop the last tool result"); |
| 856 | assert!(replaced.iter().any(|message| { |
| 857 | message.content.iter().any(|block| { |
| 858 | matches!( |
| 859 | block, |
| 860 | ContentBlock::ToolResult { tool_use_id, content, .. } |
| 861 | if tool_use_id == "live" && content.contains("FAILED") |
| 862 | ) |
| 863 | }) |
| 864 | })); |
| 865 | } |
| 866 | |
| 867 | #[test] |
| 868 | fn chat_only_history_keeps_the_latest_user_round() { |
| 869 | let messages = vec![ |
| 870 | msg("user", "hello"), |
| 871 | msg("assistant", "hi"), |
| 872 | msg("user", "how are you"), |
| 873 | msg("assistant", "fine"), |
| 874 | ]; |
| 875 | assert_eq!(last_round_start(&messages), 2); |
| 876 | } |
| 877 | |
| 878 | #[derive(serde::Deserialize)] |
| 879 | struct FixtureMatrix { |
| 880 | schema_version: u32, |
| 881 | cases: Vec<FixtureCase>, |
| 882 | } |
| 883 | |
| 884 | #[derive(serde::Deserialize)] |
| 885 | struct FixtureCase { |
| 886 | id: String, |
| 887 | expect: String, |
| 888 | #[serde(default)] |
| 889 | anchors: Option<String>, |
| 890 | original: Vec<Message>, |
| 891 | replacement: Vec<Message>, |
| 892 | #[serde(default)] |
| 893 | last_round_start: Option<usize>, |
| 894 | } |
| 895 | |
| 896 | #[test] |
| 897 | fn fixture_matrix_enforces_survival_contract() { |
| 898 | let matrix: FixtureMatrix = |
| 899 | serde_json::from_str(include_str!("fixtures/matrix.json")).expect("matrix.json"); |
| 900 | assert_eq!(matrix.schema_version, 2); |
| 901 | assert!( |
| 902 | matrix.cases.len() >= 8, |
| 903 | "fixture matrix must cover last-round, toolless-tail, chat-only, anchor, and receipt cases" |
| 904 | ); |
| 905 | for case in &matrix.cases { |
| 906 | if let Some(start) = case.last_round_start { |
| 907 | assert_eq!( |
| 908 | last_round_start(&case.original), |
| 909 | start, |
| 910 | "{} last_round_start", |
| 911 | case.id |
| 912 | ); |
| 913 | } |
| 914 | let result = validate_survival_contract( |
| 915 | &case.original, |
| 916 | &case.replacement, |
| 917 | case.anchors.as_deref(), |
| 918 | ); |
| 919 | match case.expect.as_str() { |
| 920 | "pass" => { |
| 921 | result.unwrap_or_else(|error| panic!("{} should pass: {error}", case.id)); |
| 922 | } |
| 923 | "fail" => { |
| 924 | result.expect_err(&format!("{} should fail closed", case.id)); |
| 925 | } |
| 926 | other => panic!("{}: unknown expect {other}", case.id), |
| 927 | } |
| 928 | } |
| 929 | } |
| 930 | |
| 931 | #[test] |
| 932 | fn second_compact_does_not_duplicate_prior_summaries() { |
| 933 | let first = vec![ |
| 934 | msg("user", "older"), |
| 935 | msg("user", "Run the suite now."), |
| 936 | msg("assistant", "Rerunning."), |
| 937 | tool_use("live", "Bash", json!({"command": "cargo test"})), |
| 938 | tool_result("live", "ok"), |
| 939 | checkpoint("first handoff: suite still running"), |
| 940 | ]; |
| 941 | let next = format!( |
| 942 | "{COMPACTION_SUMMARY_MARKER}: second handoff with User-pinned anchors (verbatim):\nship 0.9.12" |
| 943 | ); |
| 944 | let replaced = build_replacement_history( |
| 945 | &first, |
| 946 | &next, |
| 947 | Some("ship 0.9.12"), |
| 948 | crate::compaction::COMPACT_RETAINED_USER_MESSAGE_MAX_TOKENS, |
| 949 | ) |
| 950 | .expect("second compact must keep last round and one receipt"); |
| 951 | let checkpoints = replaced |
| 952 | .iter() |
| 953 | .filter(|message| is_compaction_checkpoint_message(message)) |
| 954 | .count(); |
| 955 | assert_eq!(checkpoints, 1, "{replaced:?}"); |
| 956 | assert!(replaced.iter().any(|message| { |
| 957 | message.content.iter().any(|block| { |
| 958 | matches!( |
| 959 | block, |
| 960 | ContentBlock::ToolResult { tool_use_id, .. } if tool_use_id == "live" |
| 961 | ) |
| 962 | }) |
| 963 | })); |
| 964 | require_text_survives(&replaced, "ship 0.9.12", "pinned /anchor text").unwrap(); |
| 965 | } |
| 966 | } |
| 967 |