| 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 | "sync" |
| 7 | |
| 8 | "reasonix/internal/provider" |
| 9 | ) |
| 10 | |
| 11 | // Session holds the conversation history for one task. The run loop (one turn at |
| 12 | // a time) is the only writer, but a frontend can read History/Save from another |
| 13 | // goroutine while a turn appends, so mu guards Messages. Direct Messages reads on |
| 14 | // the run-loop goroutine stay lock-free (serial with its own writes); cross- |
| 15 | // goroutine access goes through Snapshot. |
| 16 | type Session struct { |
| 17 | mu sync.RWMutex |
| 18 | Messages []provider.Message |
| 19 | version uint64 |
| 20 | rewriteVersion int // bumped each time the log is rewritten (compact/fold) |
| 21 | // persistedRewriteVersion is the highest rewriteVersion whose transcript |
| 22 | // has fully reached disk. It lives on the Session — not on the controller |
| 23 | // — so swapping session objects can never orphan or misattribute the |
| 24 | // baseline: NeedsRewriteSave always compares a session against its own |
| 25 | // save history. Save paths advance it under s.mu with the rewriteVersion |
| 26 | // captured alongside the message snapshot, never a re-read one, so a |
| 27 | // compaction landing mid-save stays unpersisted. |
| 28 | persistedRewriteVersion int |
| 29 | persisted sessionPersistState |
| 30 | // normalizedDirty is set when LoadSession repaired the history on the way in |
| 31 | // (empty tool-call names, dangling calls, truncated args, …). The repair |
| 32 | // already lives in Messages, so the next Save persists it automatically as |
| 33 | // part of the usual full rewrite; the flag exists for observability and to |
| 34 | // let callers opt out of work that a dirty session would make redundant. |
| 35 | normalizedDirty bool |
| 36 | // eventLogDamaged is set when LoadSession found the on-disk event log torn |
| 37 | // or corrupt and returned the replayable prefix (or the .jsonl checkpoint). |
| 38 | // The next save heals the log with a rewrite-and-compact. |
| 39 | eventLogDamaged bool |
| 40 | // rawMessages preserves the pre-normalization transcript when the load-time |
| 41 | // repairs changed it (normalizedDirty). It is only meaningful on a freshly |
| 42 | // loaded Session: checkSnapshotWrite compares a pending snapshot against |
| 43 | // what is actually on disk, and the repaired view no longer represents |
| 44 | // those bytes — a session that kept running extends the raw transcript. |
| 45 | rawMessages []provider.Message |
| 46 | // pendingContentReasons accumulates a reason string each time Rewrite() |
| 47 | // actually replaces provider-visible message bytes (compact, prune/snip, |
| 48 | // summarize, rewind, guardian merge). ReplaceLocalMetadata bumps |
| 49 | // rewriteVersion for the same save-path (NeedsRewriteSave) purpose without |
| 50 | // appending here, because ModelMessages strips or never serializes the |
| 51 | // local-only metadata it changes — so that path must never report a |
| 52 | // cache-prefix change. DrainContentRewriteReasons (run_loop.go, once per |
| 53 | // provider request) is the sole consumer. |
| 54 | pendingContentReasons []string |
| 55 | } |
| 56 | |
| 57 | // NewSession initializes a session with an optional system prompt. |
| 58 | func NewSession(system string) *Session { |
| 59 | s := &Session{} |
| 60 | if system != "" { |
| 61 | s.Messages = append(s.Messages, provider.Message{Role: provider.RoleSystem, Content: system}) |
| 62 | } |
| 63 | return s |
| 64 | } |
| 65 | |
| 66 | // Add appends a message. |
| 67 | func (s *Session) Add(m provider.Message) { |
| 68 | s.mu.Lock() |
| 69 | defer s.mu.Unlock() |
| 70 | s.Messages = append(s.Messages, m) |
| 71 | s.version++ |
| 72 | } |
| 73 | |
| 74 | // AddDecisionReceipt persists local decision metadata without inserting a |
| 75 | // standalone message into the current tool turn. Tool results must remain |
| 76 | // directly adjacent to the assistant message that requested them; otherwise |
| 77 | // session normalization fabricates interrupted placeholders and older readers |
| 78 | // can lose the real result. Attaching to the newest assistant message keeps the |
| 79 | // provider-visible transcript byte-for-byte equivalent after ModelMessages. |
| 80 | // |
| 81 | // The fallback sentinel covers host decisions made before any assistant message |
| 82 | // exists. Older readers already discard this unmatched tool record safely. |
| 83 | func (s *Session) AddDecisionReceipt(receipt *provider.DecisionReceipt) { |
| 84 | if s == nil || receipt == nil { |
| 85 | return |
| 86 | } |
| 87 | s.mu.Lock() |
| 88 | defer s.mu.Unlock() |
| 89 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 90 | if s.Messages[i].Role == provider.RoleUser && !s.Messages[i].LocalOnly { |
| 91 | break |
| 92 | } |
| 93 | if s.Messages[i].Role != provider.RoleAssistant || s.Messages[i].LocalOnly { |
| 94 | continue |
| 95 | } |
| 96 | receipts := append([]*provider.DecisionReceipt(nil), s.Messages[i].DecisionReceipts...) |
| 97 | s.Messages[i].DecisionReceipts = append(receipts, receipt) |
| 98 | // A mid-turn snapshot may already contain this assistant message. Force |
| 99 | // the next save to replace it instead of treating the later tool result |
| 100 | // as the only append-only change. |
| 101 | s.rewriteVersion++ |
| 102 | s.version++ |
| 103 | return |
| 104 | } |
| 105 | s.Messages = append(s.Messages, provider.Message{ |
| 106 | Role: provider.RoleTool, |
| 107 | ToolCallID: provider.LocalOnlyToolID, |
| 108 | Name: provider.LocalOnlyToolName, |
| 109 | LocalOnly: true, |
| 110 | DecisionReceipt: receipt, |
| 111 | }) |
| 112 | s.version++ |
| 113 | } |
| 114 | |
| 115 | // UpdateToolCallPreview replaces the preview fields of the newest matching |
| 116 | // assistant tool call. A dependent writer can only be previewed after an |
| 117 | // earlier writer in the same model batch succeeds; updating under the session |
| 118 | // lock keeps live History/Snapshot readers race-free and ensures the refreshed |
| 119 | // preview is what a resumed session archives. |
| 120 | func (s *Session) UpdateToolCallPreview(call provider.ToolCall) bool { |
| 121 | if call.ID == "" { |
| 122 | return false |
| 123 | } |
| 124 | s.mu.Lock() |
| 125 | defer s.mu.Unlock() |
| 126 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 127 | if s.Messages[i].Role != provider.RoleAssistant { |
| 128 | continue |
| 129 | } |
| 130 | calls := s.Messages[i].ToolCalls |
| 131 | for j := range calls { |
| 132 | if calls[j].ID != call.ID { |
| 133 | continue |
| 134 | } |
| 135 | cloned := append([]provider.ToolCall(nil), calls...) |
| 136 | cloned[j].Diff = call.Diff |
| 137 | cloned[j].Added = call.Added |
| 138 | cloned[j].Removed = call.Removed |
| 139 | s.Messages[i].ToolCalls = cloned |
| 140 | // A snapshot may have persisted the original assistant message while |
| 141 | // its tools were still running. Mark this as a rewrite so a later |
| 142 | // autosave replaces that message instead of misclassifying the tool |
| 143 | // results as an append-only suffix. |
| 144 | s.rewriteVersion++ |
| 145 | s.version++ |
| 146 | return true |
| 147 | } |
| 148 | } |
| 149 | return false |
| 150 | } |
| 151 | |
| 152 | // UpdateToolCallResolution persists the host-resolved target metadata for the |
| 153 | // newest matching stable proxy call. The model-visible Name/Arguments remain |
| 154 | // unchanged; this metadata exists only so live and reloaded frontends classify |
| 155 | // MCP readers and writers accurately. |
| 156 | func (s *Session) UpdateToolCallResolution(call provider.ToolCall) bool { |
| 157 | if call.ID == "" || call.ResolvedReadOnly == nil { |
| 158 | return false |
| 159 | } |
| 160 | s.mu.Lock() |
| 161 | defer s.mu.Unlock() |
| 162 | for i := len(s.Messages) - 1; i >= 0; i-- { |
| 163 | if s.Messages[i].Role != provider.RoleAssistant { |
| 164 | continue |
| 165 | } |
| 166 | calls := s.Messages[i].ToolCalls |
| 167 | for j := range calls { |
| 168 | if calls[j].ID != call.ID { |
| 169 | continue |
| 170 | } |
| 171 | cloned := append([]provider.ToolCall(nil), calls...) |
| 172 | readOnly := *call.ResolvedReadOnly |
| 173 | cloned[j].ResolvedName = call.ResolvedName |
| 174 | cloned[j].CapabilityID = call.CapabilityID |
| 175 | cloned[j].ResolvedReadOnly = &readOnly |
| 176 | s.Messages[i].ToolCalls = cloned |
| 177 | // A mid-turn snapshot may already contain the unresolved proxy call. |
| 178 | // Force the next save to rewrite that assistant message with its |
| 179 | // resolved local metadata. |
| 180 | s.rewriteVersion++ |
| 181 | s.version++ |
| 182 | return true |
| 183 | } |
| 184 | } |
| 185 | return false |
| 186 | } |
| 187 | |
| 188 | // Replace swaps the whole message log without classifying the change as a |
| 189 | // persisted-history rewrite. Call Rewrite when a live session changes messages |
| 190 | // that a mid-turn snapshot may already have written. |
| 191 | func (s *Session) Replace(msgs []provider.Message) { |
| 192 | s.mu.Lock() |
| 193 | defer s.mu.Unlock() |
| 194 | s.Messages = msgs |
| 195 | s.version++ |
| 196 | } |
| 197 | |
| 198 | // Rewrite atomically replaces the message log and marks it as a rewrite. The |
| 199 | // atomic classification matters when a periodic snapshot races compaction, |
| 200 | // pruning, or local metadata edits: a later autosave must use owned-rewrite |
| 201 | // conflict checks instead of mistaking the modified prefix for another writer. |
| 202 | // |
| 203 | // reason names the provider-visible change (e.g. "compact_auto", "snip", |
| 204 | // "rewind_truncate") and is queued for the next DrainContentRewriteReasons |
| 205 | // call, which feeds cache-diagnostics attribution. Callers whose msgs only |
| 206 | // change local-only display metadata (never serialized to the provider) must |
| 207 | // use ReplaceLocalMetadata instead, so they don't misreport a cache-prefix |
| 208 | // change that never happened. |
| 209 | func (s *Session) Rewrite(msgs []provider.Message, reason string) { |
| 210 | s.mu.Lock() |
| 211 | defer s.mu.Unlock() |
| 212 | s.Messages = msgs |
| 213 | s.rewriteVersion++ |
| 214 | s.version++ |
| 215 | if reason != "" { |
| 216 | s.pendingContentReasons = append(s.pendingContentReasons, reason) |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | // ReplaceLocalMetadata atomically replaces the message log exactly like |
| 221 | // Rewrite (including the rewriteVersion bump that forces the next save to use |
| 222 | // owned-rewrite conflict checks), for callers that only changed local-only |
| 223 | // display metadata (e.g. marking a resubmitted message Edited) rather than any |
| 224 | // provider-visible byte. Unlike Rewrite, it never queues a cache-prefix-change |
| 225 | // reason, since ModelMessages strips or never serializes what changed. |
| 226 | func (s *Session) ReplaceLocalMetadata(msgs []provider.Message) { |
| 227 | s.mu.Lock() |
| 228 | defer s.mu.Unlock() |
| 229 | s.Messages = msgs |
| 230 | s.rewriteVersion++ |
| 231 | s.version++ |
| 232 | } |
| 233 | |
| 234 | // DrainContentRewriteReasons returns and clears the reasons queued by Rewrite |
| 235 | // since the last drain. Called once per provider request (run_loop.go) so |
| 236 | // CompareShape can attribute a cache-prefix change to the operation that |
| 237 | // actually caused it. |
| 238 | func (s *Session) DrainContentRewriteReasons() []string { |
| 239 | s.mu.Lock() |
| 240 | defer s.mu.Unlock() |
| 241 | reasons := s.pendingContentReasons |
| 242 | s.pendingContentReasons = nil |
| 243 | return reasons |
| 244 | } |
| 245 | |
| 246 | // Snapshot returns a copy of the messages, safe to read from another goroutine |
| 247 | // while a turn appends. Frontends (History, Save) use it instead of touching the |
| 248 | // live slice. |
| 249 | func (s *Session) Snapshot() []provider.Message { |
| 250 | msgs, _, _ := s.snapshotWithVersion() |
| 251 | return msgs |
| 252 | } |
| 253 | |
| 254 | // Len returns the number of messages, safe to call from any goroutine. |
| 255 | func (s *Session) Len() int { |
| 256 | s.mu.RLock() |
| 257 | defer s.mu.RUnlock() |
| 258 | return len(s.Messages) |
| 259 | } |
| 260 | |
| 261 | // CloneWithMessages returns a fresh Session carrying msgs while preserving the |
| 262 | // persistence baseline of the source session. Resume paths use this when they |
| 263 | // need to adjust loaded history before a rewrite; dropping persisted would make |
| 264 | // CAS treat the first legitimate rewrite as a stale-runtime conflict. |
| 265 | // |
| 266 | // Callers that are handed history from outside this Session should prefer |
| 267 | // CloneWithMessagesIfCompatible, so stale carried history cannot borrow a newer |
| 268 | // on-disk baseline. |
| 269 | func (s *Session) CloneWithMessages(msgs []provider.Message) *Session { |
| 270 | if s == nil { |
| 271 | return nil |
| 272 | } |
| 273 | s.mu.RLock() |
| 274 | defer s.mu.RUnlock() |
| 275 | version := s.version |
| 276 | if !messagesEqualForStorageList(s.Messages, msgs) { |
| 277 | version++ |
| 278 | } |
| 279 | return &Session{ |
| 280 | Messages: append([]provider.Message(nil), msgs...), |
| 281 | version: version, |
| 282 | rewriteVersion: s.rewriteVersion, |
| 283 | persistedRewriteVersion: s.persistedRewriteVersion, |
| 284 | persisted: s.persisted, |
| 285 | normalizedDirty: s.normalizedDirty, |
| 286 | eventLogDamaged: s.eventLogDamaged, |
| 287 | pendingContentReasons: append([]string(nil), s.pendingContentReasons...), |
| 288 | } |
| 289 | } |
| 290 | |
| 291 | // CloneWithMessagesIfCompatible preserves the persistence baseline only when |
| 292 | // msgs is the same persisted history, optionally with a refreshed leading system |
| 293 | // prompt. Other history changes must happen after Resume so SaveRewrite can |
| 294 | // still detect genuine stale-controller conflicts. |
| 295 | func (s *Session) CloneWithMessagesIfCompatible(msgs []provider.Message) (*Session, bool) { |
| 296 | if s == nil { |
| 297 | return nil, false |
| 298 | } |
| 299 | s.mu.RLock() |
| 300 | defer s.mu.RUnlock() |
| 301 | if !messagesCompatibleForStorageBaseline(s.Messages, msgs) { |
| 302 | return nil, false |
| 303 | } |
| 304 | version := s.version |
| 305 | if !messagesEqualForStorageList(s.Messages, msgs) { |
| 306 | version++ |
| 307 | } |
| 308 | return &Session{ |
| 309 | Messages: append([]provider.Message(nil), msgs...), |
| 310 | version: version, |
| 311 | rewriteVersion: s.rewriteVersion, |
| 312 | persistedRewriteVersion: s.persistedRewriteVersion, |
| 313 | persisted: s.persisted, |
| 314 | normalizedDirty: s.normalizedDirty, |
| 315 | eventLogDamaged: s.eventLogDamaged, |
| 316 | pendingContentReasons: append([]string(nil), s.pendingContentReasons...), |
| 317 | }, true |
| 318 | } |
| 319 | |
| 320 | // snapshotWithVersion returns the messages together with the version and |
| 321 | // rewriteVersion they were captured under, in one lock window: save paths |
| 322 | // persist exactly this rewriteVersion as the new baseline, so a rewrite that |
| 323 | // lands after the capture cannot be misrecorded as saved. |
| 324 | func (s *Session) snapshotWithVersion() ([]provider.Message, uint64, int) { |
| 325 | s.mu.RLock() |
| 326 | defer s.mu.RUnlock() |
| 327 | return append([]provider.Message(nil), s.Messages...), s.version, s.rewriteVersion |
| 328 | } |
| 329 | |
| 330 | // RewriteVersion returns the current rewrite version. |
| 331 | func (s *Session) RewriteVersion() int { |
| 332 | s.mu.RLock() |
| 333 | defer s.mu.RUnlock() |
| 334 | return s.rewriteVersion |
| 335 | } |
| 336 | |
| 337 | // NeedsRewriteSave reports whether the history has been rewritten in memory |
| 338 | // (compaction, prune) since the last successful full save of this session. |
| 339 | // Snapshot paths use it to decide that the next write must be an owned |
| 340 | // rewrite instead of an append. |
| 341 | func (s *Session) NeedsRewriteSave() bool { |
| 342 | s.mu.RLock() |
| 343 | defer s.mu.RUnlock() |
| 344 | return s.rewriteVersion > s.persistedRewriteVersion |
| 345 | } |
| 346 | |
| 347 | // IncrementRewrite bumps the rewrite version by 1. |
| 348 | func (s *Session) IncrementRewrite() { |
| 349 | s.mu.Lock() |
| 350 | defer s.mu.Unlock() |
| 351 | s.rewriteVersion++ |
| 352 | s.version++ |
| 353 | } |
| 354 | |
| 355 | // HasContent returns true when the session carries at least one user, |
| 356 | // assistant, or tool message — i.e. more than just a system prompt. An |
| 357 | // "empty" conversation that has never been used should not be persisted. |
| 358 | func (s *Session) HasContent() bool { |
| 359 | s.mu.RLock() |
| 360 | defer s.mu.RUnlock() |
| 361 | for _, m := range s.Messages { |
| 362 | if m.Role != provider.RoleSystem { |
| 363 | return true |
| 364 | } |
| 365 | } |
| 366 | return false |
| 367 | } |
| 368 | |
| 369 | // HasSystemMessage reports whether the session starts with a system message, |
| 370 | // which carries the agent's stable identity and behavioural contract. Sessions |
| 371 | // without one are not safe to persist: when reloaded the model has no identity |
| 372 | // context and falls back to its training-data defaults. |
| 373 | func (s *Session) HasSystemMessage() bool { |
| 374 | s.mu.RLock() |
| 375 | defer s.mu.RUnlock() |
| 376 | return len(s.Messages) > 0 && s.Messages[0].Role == provider.RoleSystem |
| 377 | } |
| 378 |