| 1 | // Package agent wires a Provider, a tool Registry, and a Session into the |
| 2 | // harness loop that drives a coding task to completion. |
| 3 | package agent |
| 4 | |
| 5 | import ( |
| 6 | "bytes" |
| 7 | "slices" |
| 8 | "strings" |
| 9 | "sync" |
| 10 | |
| 11 | "reasonix/internal/provider" |
| 12 | ) |
| 13 | |
| 14 | // Session holds the conversation history for one task. The run loop (one turn at |
| 15 | // a time) is the only writer, but a frontend can read History/Save from another |
| 16 | // goroutine while a turn appends, so mu guards Messages. Direct Messages reads on |
| 17 | // the run-loop goroutine stay lock-free (serial with its own writes); cross- |
| 18 | // goroutine access goes through Snapshot. |
| 19 | type Session struct { |
| 20 | cacheSessionID string // ephemeral transport identity; never model-visible or persisted |
| 21 | mu sync.RWMutex |
| 22 | Messages []provider.Message |
| 23 | version uint64 |
| 24 | recoveryMetadataVersion uint64 // local receipt edits require persistence, not a model-history rewrite |
| 25 | rewriteVersion int // bumped each time the log is rewritten (compact/fold) |
| 26 | // persistedRewriteVersion is the highest rewriteVersion whose transcript |
| 27 | // has fully reached disk. It lives on the Session — not on the controller |
| 28 | // — so swapping session objects can never orphan or misattribute the |
| 29 | // baseline: NeedsRewriteSave always compares a session against its own |
| 30 | // save history. Save paths advance it under s.mu with the rewriteVersion |
| 31 | // captured alongside the message snapshot, never a re-read one, so a |
| 32 | // compaction landing mid-save stays unpersisted. |
| 33 | persistedRewriteVersion int |
| 34 | persisted sessionPersistState |
| 35 | // normalizedDirty is set when LoadSession repaired the history on the way in |
| 36 | // (empty tool-call names, dangling calls, truncated args, …). The repair |
| 37 | // already lives in Messages, so the next Save persists it automatically as |
| 38 | // part of the usual full rewrite; the flag exists for observability and to |
| 39 | // let callers opt out of work that a dirty session would make redundant. |
| 40 | normalizedDirty bool |
| 41 | // eventLogDamaged is set when LoadSession found the on-disk event log torn |
| 42 | // or corrupt and returned the replayable prefix (or the .jsonl checkpoint). |
| 43 | // The next save heals the log with a rewrite-and-compact. |
| 44 | eventLogDamaged bool |
| 45 | // rawMessages preserves the pre-normalization transcript when the load-time |
| 46 | // repairs changed it (normalizedDirty). It is only meaningful on a freshly |
| 47 | // loaded Session: checkSnapshotWrite compares a pending snapshot against |
| 48 | // what is actually on disk, and the repaired view no longer represents |
| 49 | // those bytes — a session that kept running extends the raw transcript. |
| 50 | rawMessages []provider.Message |
| 51 | // pendingContentReasons accumulates a reason string each time Rewrite() |
| 52 | // actually replaces provider-visible message bytes (compact, prune/snip, |
| 53 | // summarize, rewind, guardian merge). ReplaceLocalMetadata bumps |
| 54 | // rewriteVersion for the same save-path (NeedsRewriteSave) purpose without |
| 55 | // appending here, because ModelMessages strips or never serializes the |
| 56 | // local-only metadata it changes — so that path must never report a |
| 57 | // cache-prefix change. DrainContentRewriteReasons (run_loop.go, once per |
| 58 | // provider request) is the sole consumer. |
| 59 | pendingContentReasons []string |
| 60 | // persistObserver receives non-blocking post-commit projection hints. It is |
| 61 | // deliberately session-local so multiple runtimes cannot steal each other's |
| 62 | // observer registration. |
| 63 | persistObserver SessionPersistObserver |
| 64 | // writeAuth is the generation-bound write permit for this session's path. |
| 65 | // Controllers bind it after acquiring a SessionLease; save/ownership paths |
| 66 | // consult it instead of a process-level "I hold a lease" boolean. |
| 67 | writeAuth *SessionWriteAuthority |
| 68 | // authRequired becomes true once any authority has been bound. From then |
| 69 | // on, saves fail closed without a live authority rather than forking |
| 70 | // recovery under a stale controller. |
| 71 | authRequired bool |
| 72 | // persistedMessages is the last paired on-disk view for persistedViewPath. |
| 73 | persistedMessages []provider.Message |
| 74 | // persistedViewPath is empty when the persist baseline has no paired view. |
| 75 | persistedViewPath string |
| 76 | // recoveryLane is a session-instance identity, allocated lazily on the |
| 77 | // first true conflict. It bounds repeated saves by this live controller to |
| 78 | // one recovery file without letting a replacement controller overwrite it. |
| 79 | recoveryLane string |
| 80 | head sessionHeadState |
| 81 | } |
| 82 | |
| 83 | // NewSession initializes a session with an optional system prompt. |
| 84 | func NewSession(system string) *Session { |
| 85 | s := &Session{} |
| 86 | if system != "" { |
| 87 | s.Messages = append(s.Messages, provider.Message{Role: provider.RoleSystem, Content: system, ID: NewMessageID()}) |
| 88 | } |
| 89 | return s |
| 90 | } |
| 91 | |
| 92 | // Add appends a message. |
| 93 | func (s *Session) Add(m provider.Message) { |
| 94 | s.mu.Lock() |
| 95 | defer s.mu.Unlock() |
| 96 | if m.ID == "" { |
| 97 | m.ID = NewMessageID() |
| 98 | } |
| 99 | s.expireProtocolRecoveryLocked([]provider.Message{m}) |
| 100 | s.Messages = append(s.Messages, m) |
| 101 | s.version++ |
| 102 | } |
| 103 | |
| 104 | // AddBatch appends one logical transcript batch under a single lock. Turn |
| 105 | // admission uses it for an optional host context revision plus the real user |
| 106 | // message so autosave can never observe only half of the admitted boundary. |
| 107 | func (s *Session) AddBatch(messages ...provider.Message) { |
| 108 | if s == nil || len(messages) == 0 { |
| 109 | return |
| 110 | } |
| 111 | s.mu.Lock() |
| 112 | mintMessageIDs(messages) |
| 113 | s.expireProtocolRecoveryLocked(messages) |
| 114 | s.Messages = append(s.Messages, messages...) |
| 115 | s.version++ |
| 116 | s.mu.Unlock() |
| 117 | } |
| 118 | |
| 119 | // SetLeadingSystemPrompt updates or sets the leading system prompt message. |
| 120 | func (s *Session) SetLeadingSystemPrompt(prompt string) { |
| 121 | s.SetLeadingSystemPromptWithReason(prompt, "system_prompt_refresh") |
| 122 | } |
| 123 | |
| 124 | // SetLeadingSystemPromptWithReason refreshes the authoritative system prompt |
| 125 | // and records the provider-visible rewrite boundary exactly once. It is used |
| 126 | // for low-frequency host prompt migrations, never ordinary pinned-context |
| 127 | // updates (which append user-role revisions instead). |
| 128 | func (s *Session) SetLeadingSystemPromptWithReason(prompt, reason string) bool { |
| 129 | if s == nil { |
| 130 | return false |
| 131 | } |
| 132 | s.mu.Lock() |
| 133 | defer s.mu.Unlock() |
| 134 | if len(s.Messages) > 0 && s.Messages[0].Role == provider.RoleSystem { |
| 135 | if s.Messages[0].Content == prompt { |
| 136 | return false |
| 137 | } |
| 138 | s.Messages[0].Content = prompt |
| 139 | } else if prompt != "" { |
| 140 | s.Messages = append([]provider.Message{{Role: provider.RoleSystem, Content: prompt, ID: NewMessageID()}}, s.Messages...) |
| 141 | } else { |
| 142 | return false |
| 143 | } |
| 144 | s.rewriteVersion++ |
| 145 | if reason = strings.TrimSpace(reason); reason != "" { |
| 146 | s.pendingContentReasons = append(s.pendingContentReasons, reason) |
| 147 | } |
| 148 | s.version++ |
| 149 | return true |
| 150 | } |
| 151 | |
| 152 | // ConsumeFinalReadinessRecovery marks the newest pending readiness checkpoint |
| 153 | // consumed before any next user turn (explicit recovery or ordinary follow-up). |
| 154 | // This is local metadata only, so the rewrite does not alter provider bytes or |
| 155 | // prompt-cache identity. |
| 156 | func (s *Session) ConsumeFinalReadinessRecovery() bool { |
| 157 | if s == nil { |
| 158 | return false |
| 159 | } |
| 160 | s.mu.Lock() |
| 161 | defer s.mu.Unlock() |
| 162 | for i := range slices.Backward(s.Messages) { |
| 163 | message := &s.Messages[i] |
| 164 | if message.LocalOnly && message.FinalReadinessRecovery != nil && message.FinalReadinessRecovery.Pending { |
| 165 | consumed := *message.FinalReadinessRecovery |
| 166 | consumed.Pending = false |
| 167 | consumed.Missing = append([]string(nil), consumed.Missing...) |
| 168 | consumed.Checkpoint = append([]byte(nil), consumed.Checkpoint...) |
| 169 | message.FinalReadinessRecovery = &consumed |
| 170 | s.rewriteVersion++ |
| 171 | s.version++ |
| 172 | return true |
| 173 | } |
| 174 | if IsUserAuthoredTurnMessage(*message) { |
| 175 | return false |
| 176 | } |
| 177 | } |
| 178 | return false |
| 179 | } |
| 180 | |
| 181 | // AddDecisionReceipt persists local decision metadata without inserting a |
| 182 | // standalone message into the current tool turn. Tool results must remain |
| 183 | // directly adjacent to the assistant message that requested them; otherwise |
| 184 | // session normalization fabricates interrupted placeholders and older readers |
| 185 | // can lose the real result. Attaching to the newest assistant message keeps the |
| 186 | // provider-visible transcript byte-for-byte equivalent after ModelMessages. |
| 187 | // |
| 188 | // The fallback sentinel covers host decisions made before any assistant message |
| 189 | // exists. Older readers already discard this unmatched tool record safely. |
| 190 | func (s *Session) AddDecisionReceipt(receipt *provider.DecisionReceipt) { |
| 191 | if s == nil || receipt == nil { |
| 192 | return |
| 193 | } |
| 194 | s.mu.Lock() |
| 195 | defer s.mu.Unlock() |
| 196 | //nolint:modernize // slices.Backward yields element copies; this body writes through the index. |
| 197 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 198 | if IsUserAuthoredTurnMessage(s.Messages[i]) { |
| 199 | break |
| 200 | } |
| 201 | if s.Messages[i].Role != provider.RoleAssistant || s.Messages[i].LocalOnly { |
| 202 | continue |
| 203 | } |
| 204 | receipts := append([]*provider.DecisionReceipt(nil), s.Messages[i].DecisionReceipts...) |
| 205 | s.Messages[i].DecisionReceipts = append(receipts, receipt) |
| 206 | // A mid-turn snapshot may already contain this assistant message. Force |
| 207 | // the next save to replace it instead of treating the later tool result |
| 208 | // as the only append-only change. |
| 209 | s.rewriteVersion++ |
| 210 | s.version++ |
| 211 | return |
| 212 | } |
| 213 | s.Messages = append(s.Messages, provider.Message{ |
| 214 | Role: provider.RoleTool, |
| 215 | ToolCallID: provider.LocalOnlyToolID, |
| 216 | Name: provider.LocalOnlyToolName, |
| 217 | LocalOnly: true, |
| 218 | DecisionReceipt: receipt, |
| 219 | }) |
| 220 | s.version++ |
| 221 | } |
| 222 | |
| 223 | // UpdateToolCallPreview replaces the preview fields of the newest matching |
| 224 | // assistant tool call. A dependent writer can only be previewed after an |
| 225 | // earlier writer in the same model batch succeeds; updating under the session |
| 226 | // lock keeps live History/Snapshot readers race-free and ensures the refreshed |
| 227 | // preview is what a resumed session archives. |
| 228 | func (s *Session) UpdateToolCallPreview(call provider.ToolCall) bool { |
| 229 | if call.ID == "" { |
| 230 | return false |
| 231 | } |
| 232 | s.mu.Lock() |
| 233 | defer s.mu.Unlock() |
| 234 | //nolint:modernize // slices.Backward yields element copies; this body writes through the index. |
| 235 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 236 | if s.Messages[i].Role != provider.RoleAssistant { |
| 237 | continue |
| 238 | } |
| 239 | calls := s.Messages[i].ToolCalls |
| 240 | for j := range calls { |
| 241 | if calls[j].ID != call.ID { |
| 242 | continue |
| 243 | } |
| 244 | cloned := append([]provider.ToolCall(nil), calls...) |
| 245 | cloned[j].Diff = call.Diff |
| 246 | cloned[j].Added = call.Added |
| 247 | cloned[j].Removed = call.Removed |
| 248 | s.Messages[i].ToolCalls = cloned |
| 249 | // A snapshot may have persisted the original assistant message while |
| 250 | // its tools were still running. Mark this as a rewrite so a later |
| 251 | // autosave replaces that message instead of misclassifying the tool |
| 252 | // results as an append-only suffix. |
| 253 | s.rewriteVersion++ |
| 254 | s.version++ |
| 255 | return true |
| 256 | } |
| 257 | } |
| 258 | return false |
| 259 | } |
| 260 | |
| 261 | // UpdateToolCallResolution persists the host-resolved target metadata for the |
| 262 | // newest matching stable proxy call. The model-visible Name/Arguments remain |
| 263 | // unchanged; this metadata exists only so live and reloaded frontends classify |
| 264 | // MCP readers and writers accurately. |
| 265 | func (s *Session) UpdateToolCallResolution(call provider.ToolCall) bool { |
| 266 | if call.ID == "" || call.ResolvedReadOnly == nil { |
| 267 | return false |
| 268 | } |
| 269 | s.mu.Lock() |
| 270 | defer s.mu.Unlock() |
| 271 | //nolint:modernize // slices.Backward yields element copies; this body writes through the index. |
| 272 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 273 | if s.Messages[i].Role != provider.RoleAssistant { |
| 274 | continue |
| 275 | } |
| 276 | calls := s.Messages[i].ToolCalls |
| 277 | for j := range calls { |
| 278 | if calls[j].ID != call.ID { |
| 279 | continue |
| 280 | } |
| 281 | cloned := append([]provider.ToolCall(nil), calls...) |
| 282 | readOnly := *call.ResolvedReadOnly |
| 283 | cloned[j].ResolvedName = call.ResolvedName |
| 284 | cloned[j].CapabilityID = call.CapabilityID |
| 285 | cloned[j].ResolvedReadOnly = &readOnly |
| 286 | s.Messages[i].ToolCalls = cloned |
| 287 | // A mid-turn snapshot may already contain the unresolved proxy call. |
| 288 | // Force the next save to rewrite that assistant message with its |
| 289 | // resolved local metadata. |
| 290 | s.rewriteVersion++ |
| 291 | s.version++ |
| 292 | return true |
| 293 | } |
| 294 | } |
| 295 | return false |
| 296 | } |
| 297 | |
| 298 | // Replace swaps the whole message log without classifying the change as a |
| 299 | // persisted-history rewrite. Call Rewrite when a live session changes messages |
| 300 | // that a mid-turn snapshot may already have written. |
| 301 | func (s *Session) Replace(msgs []provider.Message) { |
| 302 | s.mu.Lock() |
| 303 | defer s.mu.Unlock() |
| 304 | msgs = retainUnresolvedToolRecords(s.Messages, msgs) |
| 305 | mintMessageIDs(msgs) |
| 306 | s.Messages = msgs |
| 307 | s.version++ |
| 308 | } |
| 309 | |
| 310 | // Rewrite atomically replaces the message log and marks it as a rewrite. The |
| 311 | // atomic classification matters when a periodic snapshot races compaction, |
| 312 | // pruning, or local metadata edits: a later autosave must use owned-rewrite |
| 313 | // conflict checks instead of mistaking the modified prefix for another writer. |
| 314 | // |
| 315 | // reason names the provider-visible change (e.g. "rewind_truncate", |
| 316 | // "guardian_merge") and is queued for the next DrainContentRewriteReasons |
| 317 | // call, which feeds cache-diagnostics attribution. Callers whose msgs only |
| 318 | // change local-only display metadata (never serialized to the provider) must |
| 319 | // use ReplaceLocalMetadata instead, so they don't misreport a cache-prefix |
| 320 | // change that never happened. |
| 321 | func (s *Session) Rewrite(msgs []provider.Message, reason string) { |
| 322 | s.mu.Lock() |
| 323 | defer s.mu.Unlock() |
| 324 | msgs = retainUnresolvedToolRecords(s.Messages, msgs) |
| 325 | mintMessageIDs(msgs) |
| 326 | s.Messages = msgs |
| 327 | s.rewriteVersion++ |
| 328 | s.version++ |
| 329 | if reason != "" { |
| 330 | s.pendingContentReasons = append(s.pendingContentReasons, reason) |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | // ReplaceLocalMetadata atomically replaces the message log exactly like |
| 335 | // Rewrite (including the rewriteVersion bump that forces the next save to use |
| 336 | // owned-rewrite conflict checks), for callers that only changed local-only |
| 337 | // display metadata (e.g. marking a resubmitted message Edited) rather than any |
| 338 | // provider-visible byte. Unlike Rewrite, it never queues a cache-prefix-change |
| 339 | // reason, since ModelMessages strips or never serializes what changed. |
| 340 | func (s *Session) ReplaceLocalMetadata(msgs []provider.Message) { |
| 341 | s.mu.Lock() |
| 342 | defer s.mu.Unlock() |
| 343 | msgs = retainUnresolvedToolRecords(s.Messages, msgs) |
| 344 | mintMessageIDs(msgs) |
| 345 | s.Messages = msgs |
| 346 | s.rewriteVersion++ |
| 347 | s.version++ |
| 348 | } |
| 349 | |
| 350 | // DrainContentRewriteReasons returns and clears the reasons queued by Rewrite |
| 351 | // since the last drain. Called once per provider request (run_loop.go) so |
| 352 | // CompareShape can attribute a cache-prefix change to the operation that |
| 353 | // actually caused it. |
| 354 | func (s *Session) DrainContentRewriteReasons() []string { |
| 355 | s.mu.Lock() |
| 356 | defer s.mu.Unlock() |
| 357 | reasons := s.pendingContentReasons |
| 358 | s.pendingContentReasons = nil |
| 359 | return reasons |
| 360 | } |
| 361 | |
| 362 | // NoteContentRewrite queues a provider-visible prefix-change reason without |
| 363 | // mutating Messages. Projection installs and resume-time system migrations use |
| 364 | // this so cache diagnostics attribute the next request's miss while the |
| 365 | // canonical transcript and its persistence baseline stay intact. |
| 366 | func (s *Session) NoteContentRewrite(reason string) { |
| 367 | if s == nil || reason == "" { |
| 368 | return |
| 369 | } |
| 370 | s.mu.Lock() |
| 371 | defer s.mu.Unlock() |
| 372 | s.pendingContentReasons = append(s.pendingContentReasons, reason) |
| 373 | } |
| 374 | |
| 375 | // Snapshot returns a copy of the messages, safe to read from another goroutine |
| 376 | // while a turn appends. Frontends (History, Save) use it instead of touching the |
| 377 | // live slice. |
| 378 | func (s *Session) Snapshot() []provider.Message { |
| 379 | msgs, _, _ := s.snapshotWithVersion() |
| 380 | return msgs |
| 381 | } |
| 382 | |
| 383 | // DisplayBaseline captures messages and their rewrite/head identity together. |
| 384 | // The controller calls this at an idle/admission boundary, before any new |
| 385 | // streaming event can commit to its display projection. |
| 386 | func (s *Session) DisplayBaseline() (messages []provider.Message, headID string, rewriteEpoch uint64) { |
| 387 | s.mu.RLock() |
| 388 | defer s.mu.RUnlock() |
| 389 | return append([]provider.Message(nil), s.Messages...), s.head.ref.HeadID, uint64(s.rewriteVersion) |
| 390 | } |
| 391 | |
| 392 | // Len returns the number of messages, safe to call from any goroutine. |
| 393 | func (s *Session) Len() int { |
| 394 | s.mu.RLock() |
| 395 | defer s.mu.RUnlock() |
| 396 | return len(s.Messages) |
| 397 | } |
| 398 | |
| 399 | // MessageRange returns a copy of the messages in [start, end), clamped to the |
| 400 | // current log bounds, safe to read from another goroutine while a turn |
| 401 | // appends. Paging frontends use it to fetch a display window without paying |
| 402 | // for a Snapshot of the whole history. |
| 403 | func (s *Session) MessageRange(start, end int) []provider.Message { |
| 404 | s.mu.RLock() |
| 405 | defer s.mu.RUnlock() |
| 406 | if start < 0 { |
| 407 | start = 0 |
| 408 | } |
| 409 | if end > len(s.Messages) { |
| 410 | end = len(s.Messages) |
| 411 | } |
| 412 | if start >= end { |
| 413 | return []provider.Message{} |
| 414 | } |
| 415 | return append([]provider.Message(nil), s.Messages[start:end]...) |
| 416 | } |
| 417 | |
| 418 | // CloneWithMessages returns a fresh Session carrying msgs while preserving the |
| 419 | // persistence baseline of the source session. Resume paths use this when they |
| 420 | // need to adjust loaded history before a rewrite; dropping persisted would make |
| 421 | // CAS treat the first legitimate rewrite as a stale-runtime conflict. |
| 422 | // |
| 423 | // Callers that are handed history from outside this Session should prefer |
| 424 | // CloneWithMessagesIfCompatible, so stale carried history cannot borrow a newer |
| 425 | // on-disk baseline. |
| 426 | func (s *Session) CloneWithMessages(msgs []provider.Message) *Session { |
| 427 | if s == nil { |
| 428 | return nil |
| 429 | } |
| 430 | s.mu.RLock() |
| 431 | defer s.mu.RUnlock() |
| 432 | version := s.version |
| 433 | if !messagesEqualForStorageList(s.Messages, msgs) { |
| 434 | version++ |
| 435 | } |
| 436 | return &Session{ |
| 437 | Messages: append([]provider.Message(nil), msgs...), |
| 438 | version: version, |
| 439 | recoveryMetadataVersion: s.recoveryMetadataVersion, |
| 440 | rewriteVersion: s.rewriteVersion, |
| 441 | persistedRewriteVersion: s.persistedRewriteVersion, |
| 442 | persisted: s.persisted, |
| 443 | normalizedDirty: s.normalizedDirty, |
| 444 | eventLogDamaged: s.eventLogDamaged, |
| 445 | rawMessages: append([]provider.Message(nil), s.rawMessages...), |
| 446 | pendingContentReasons: append([]string(nil), s.pendingContentReasons...), |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | // CloneWithMessagesIfCompatible preserves the persistence baseline only when |
| 451 | // msgs is the same persisted history, optionally with a refreshed leading system |
| 452 | // prompt. Other history changes must happen after Resume so SaveRewrite can |
| 453 | // still detect genuine stale-controller conflicts. |
| 454 | func (s *Session) CloneWithMessagesIfCompatible(msgs []provider.Message) (*Session, bool) { |
| 455 | if s == nil { |
| 456 | return nil, false |
| 457 | } |
| 458 | s.mu.RLock() |
| 459 | defer s.mu.RUnlock() |
| 460 | if !messagesCompatibleForStorageBaseline(s.Messages, msgs) { |
| 461 | return nil, false |
| 462 | } |
| 463 | version := s.version |
| 464 | if !messagesEqualForStorageList(s.Messages, msgs) { |
| 465 | version++ |
| 466 | } |
| 467 | return &Session{ |
| 468 | Messages: append([]provider.Message(nil), msgs...), |
| 469 | version: version, |
| 470 | recoveryMetadataVersion: s.recoveryMetadataVersion, |
| 471 | rewriteVersion: s.rewriteVersion, |
| 472 | persistedRewriteVersion: s.persistedRewriteVersion, |
| 473 | persisted: s.persisted, |
| 474 | normalizedDirty: s.normalizedDirty, |
| 475 | eventLogDamaged: s.eventLogDamaged, |
| 476 | rawMessages: append([]provider.Message(nil), s.rawMessages...), |
| 477 | pendingContentReasons: append([]string(nil), s.pendingContentReasons...), |
| 478 | }, true |
| 479 | } |
| 480 | |
| 481 | // projectionValidationMessages returns the current canonical transcript and, |
| 482 | // when LoadSession repaired it, the exact pre-repair disk view. Resume wrappers |
| 483 | // preserve both so projection sidecars can be migrated without weakening the |
| 484 | // covered-prefix check. |
| 485 | func (s *Session) projectionValidationMessages() (current, preRepair []provider.Message) { |
| 486 | if s == nil { |
| 487 | return nil, nil |
| 488 | } |
| 489 | s.mu.RLock() |
| 490 | defer s.mu.RUnlock() |
| 491 | current = append([]provider.Message(nil), s.Messages...) |
| 492 | if s.normalizedDirty && len(s.rawMessages) > 0 { |
| 493 | preRepair = append([]provider.Message(nil), s.rawMessages...) |
| 494 | } |
| 495 | return current, preRepair |
| 496 | } |
| 497 | |
| 498 | // snapshotWithVersion returns the messages together with the version and |
| 499 | // rewriteVersion they were captured under, in one lock window: save paths |
| 500 | // persist exactly this rewriteVersion as the new baseline, so a rewrite that |
| 501 | // lands after the capture cannot be misrecorded as saved. |
| 502 | func (s *Session) snapshotWithVersion() ([]provider.Message, uint64, int) { |
| 503 | s.mu.RLock() |
| 504 | defer s.mu.RUnlock() |
| 505 | return append([]provider.Message(nil), s.Messages...), s.version, s.rewriteVersion |
| 506 | } |
| 507 | |
| 508 | // snapshotMessagesVersion returns a copy of the messages with the transcript |
| 509 | // version, for projection validity checks that do not need rewriteVersion. |
| 510 | func (s *Session) snapshotMessagesVersion() ([]provider.Message, uint64) { |
| 511 | msgs, version, _ := s.snapshotWithVersion() |
| 512 | return msgs, version |
| 513 | } |
| 514 | |
| 515 | // TranscriptVersion returns the current append/rewrite counter used by |
| 516 | // context-projection validity checks. |
| 517 | func (s *Session) TranscriptVersion() uint64 { |
| 518 | s.mu.RLock() |
| 519 | defer s.mu.RUnlock() |
| 520 | return s.version |
| 521 | } |
| 522 | |
| 523 | // RewriteVersion returns the current rewrite version. |
| 524 | func (s *Session) RewriteVersion() int { |
| 525 | s.mu.RLock() |
| 526 | defer s.mu.RUnlock() |
| 527 | return s.rewriteVersion |
| 528 | } |
| 529 | |
| 530 | // NeedsRewriteSave reports whether the message log was rewritten in place — |
| 531 | // rather than appended to — since the last successful full save of this |
| 532 | // session. Snapshot paths use it to decide that the next write must be an |
| 533 | // owned rewrite instead of an append. |
| 534 | func (s *Session) NeedsRewriteSave() bool { |
| 535 | s.mu.RLock() |
| 536 | defer s.mu.RUnlock() |
| 537 | return s.rewriteVersion > s.persistedRewriteVersion || s.recoveryMetadataVersion > s.persisted.version |
| 538 | } |
| 539 | |
| 540 | // HasUnsavedChanges reports whether the in-memory transcript contains storage |
| 541 | // changes that have not been durably recorded at path. It is intentionally |
| 542 | // conservative when no verified baseline exists: an idle controller must not |
| 543 | // replace an in-memory conversation with a possibly older disk copy after a |
| 544 | // bounded lock failure or an interrupted save. |
| 545 | func (s *Session) HasUnsavedChanges(path string) bool { |
| 546 | if s == nil || strings.TrimSpace(path) == "" { |
| 547 | return false |
| 548 | } |
| 549 | msgs, _, rewriteVersion := s.snapshotWithVersion() |
| 550 | digest, err := digestSessionMessages(msgs) |
| 551 | if err != nil { |
| 552 | return true |
| 553 | } |
| 554 | key := canonicalSessionSavePath(path) |
| 555 | s.mu.RLock() |
| 556 | defer s.mu.RUnlock() |
| 557 | if !s.persisted.ok || s.persisted.path != key { |
| 558 | return true |
| 559 | } |
| 560 | if s.normalizedDirty || s.eventLogDamaged || rewriteVersion > s.persistedRewriteVersion { |
| 561 | return true |
| 562 | } |
| 563 | return !bytes.Equal(digest[:], s.persisted.digest[:]) |
| 564 | } |
| 565 | |
| 566 | // IncrementRewrite bumps the rewrite version by 1. |
| 567 | func (s *Session) IncrementRewrite() { |
| 568 | s.mu.Lock() |
| 569 | defer s.mu.Unlock() |
| 570 | s.rewriteVersion++ |
| 571 | s.version++ |
| 572 | } |
| 573 | |
| 574 | // HasContent returns true when the session carries at least one user, |
| 575 | // assistant, or tool message — i.e. more than just a system prompt. An |
| 576 | // "empty" conversation that has never been used should not be persisted. |
| 577 | func (s *Session) HasContent() bool { |
| 578 | s.mu.RLock() |
| 579 | defer s.mu.RUnlock() |
| 580 | for _, m := range s.Messages { |
| 581 | if m.Role != provider.RoleSystem { |
| 582 | return true |
| 583 | } |
| 584 | } |
| 585 | return false |
| 586 | } |
| 587 | |
| 588 | // HasSystemMessage reports whether the session starts with a system message, |
| 589 | // which carries the agent's stable identity and behavioural contract. Sessions |
| 590 | // without one are not safe to persist: when reloaded the model has no identity |
| 591 | // context and falls back to its training-data defaults. |
| 592 | func (s *Session) HasSystemMessage() bool { |
| 593 | s.mu.RLock() |
| 594 | defer s.mu.RUnlock() |
| 595 | return len(s.Messages) > 0 && s.Messages[0].Role == provider.RoleSystem |
| 596 | } |
| 597 |