| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "crypto/rand" |
| 5 | "crypto/sha256" |
| 6 | "encoding/hex" |
| 7 | "encoding/json" |
| 8 | "errors" |
| 9 | "fmt" |
| 10 | "os" |
| 11 | "path/filepath" |
| 12 | "sort" |
| 13 | "strings" |
| 14 | "sync" |
| 15 | "time" |
| 16 | |
| 17 | "reasonix/internal/fileutil" |
| 18 | fileencoding "reasonix/internal/fileutil/encoding" |
| 19 | "reasonix/internal/store" |
| 20 | "reasonix/internal/tool" |
| 21 | ) |
| 22 | |
| 23 | type SubagentStatus string |
| 24 | |
| 25 | const ( |
| 26 | SubagentRunning SubagentStatus = "running" |
| 27 | SubagentCompleted SubagentStatus = "completed" |
| 28 | SubagentFailed SubagentStatus = "failed" |
| 29 | SubagentInterrupted SubagentStatus = "interrupted" |
| 30 | ) |
| 31 | |
| 32 | // SubagentMeta is the sidecar for a persisted sub-agent transcript. It captures |
| 33 | // the execution identity that must stay stable for continuation/fork. |
| 34 | type SubagentMeta struct { |
| 35 | Ref string `json:"ref"` |
| 36 | CreatedAt time.Time `json:"createdAt"` |
| 37 | UpdatedAt time.Time `json:"updatedAt"` |
| 38 | Status SubagentStatus `json:"status"` |
| 39 | Kind string `json:"kind"` // task | skill |
| 40 | Name string `json:"name"` |
| 41 | WorkspaceRoot string `json:"workspaceRoot"` |
| 42 | ParentSession string `json:"parentSession,omitempty"` |
| 43 | ParentToolCallID string `json:"parentToolCallId,omitempty"` |
| 44 | ForkedFrom string `json:"forkedFrom,omitempty"` |
| 45 | SystemPromptHash string `json:"systemPromptHash"` |
| 46 | ToolScope []string `json:"toolScope"` |
| 47 | ToolSchemaHash string `json:"toolSchemaHash"` |
| 48 | Model string `json:"model"` |
| 49 | Effort string `json:"effort"` |
| 50 | } |
| 51 | |
| 52 | // subagentMetaDecodeError distinguishes malformed metadata content from file |
| 53 | // I/O failures. Cleanup may safely skip one undecodable record, but storage |
| 54 | // errors must remain visible because they can affect every subagent record. |
| 55 | type subagentMetaDecodeError struct { |
| 56 | ref string |
| 57 | err error |
| 58 | } |
| 59 | |
| 60 | func (e *subagentMetaDecodeError) Error() string { |
| 61 | return fmt.Sprintf("decode subagent metadata %q: %v", e.ref, e.err) |
| 62 | } |
| 63 | |
| 64 | func (e *subagentMetaDecodeError) Unwrap() error { return e.err } |
| 65 | |
| 66 | func isSubagentMetaDecodeError(err error) bool { |
| 67 | var decodeErr *subagentMetaDecodeError |
| 68 | return errors.As(err, &decodeErr) |
| 69 | } |
| 70 | |
| 71 | // SubagentSpec describes the current invocation identity. |
| 72 | type SubagentSpec struct { |
| 73 | Kind string |
| 74 | Name string |
| 75 | WorkspaceRoot string |
| 76 | ParentSession string |
| 77 | ParentToolCallID string |
| 78 | SystemPrompt string |
| 79 | Registry *tool.Registry |
| 80 | Model string |
| 81 | Effort string |
| 82 | } |
| 83 | |
| 84 | // SubagentRun is a prepared transcript run. Call Release exactly once. |
| 85 | type SubagentRun struct { |
| 86 | Ref string |
| 87 | Session *Session |
| 88 | Meta SubagentMeta |
| 89 | ForkedFrom string |
| 90 | |
| 91 | store *SubagentStore |
| 92 | release func() |
| 93 | } |
| 94 | |
| 95 | // SubagentArtifact is a persisted sub-agent transcript and metadata pair owned |
| 96 | // by a parent session. One file may be missing after a crash; lifecycle cleanup |
| 97 | // should operate on the paths that exist. |
| 98 | type SubagentArtifact struct { |
| 99 | Ref string |
| 100 | SessionPath string |
| 101 | MetaPath string |
| 102 | Meta SubagentMeta |
| 103 | } |
| 104 | |
| 105 | func (r *SubagentRun) Release() { |
| 106 | if r != nil && r.release != nil { |
| 107 | r.release() |
| 108 | r.release = nil |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | // EphemeralSubagentRun is a non-persisted run for callers without an owning |
| 113 | // parent session — e.g. headless `reasonix run`, which never mints a session |
| 114 | // path. Its empty Ref makes the store's MarkRunning/SaveCompleted/SaveFailed |
| 115 | // methods no-op and keeps FormatSubagentResult from emitting a transcript |
| 116 | // reference, so the sub-agent behaves exactly as it did before persisted |
| 117 | // transcripts existed. It holds no lock, so Release is a no-op. |
| 118 | func EphemeralSubagentRun(systemPrompt string) *SubagentRun { |
| 119 | return &SubagentRun{Session: NewSession(systemPrompt)} |
| 120 | } |
| 121 | |
| 122 | // SubagentStore persists sub-agent transcripts under config.SessionDir()/subagents. |
| 123 | // Its locks are process-local; cross-process mutation is intentionally out of v1. |
| 124 | type SubagentStore struct { |
| 125 | dir string |
| 126 | destroyed func(parentSession string) bool |
| 127 | parentSessionProbe func(sessionPath string) bool |
| 128 | |
| 129 | // cleanupBeforeReread is a test seam for deterministic lease interleavings. |
| 130 | cleanupBeforeReread func(parentSession, ref string) |
| 131 | |
| 132 | mu sync.Mutex |
| 133 | locked map[string]bool |
| 134 | } |
| 135 | |
| 136 | func NewSubagentStore(dir string) *SubagentStore { |
| 137 | if strings.TrimSpace(dir) == "" { |
| 138 | return nil |
| 139 | } |
| 140 | return &SubagentStore{dir: dir, locked: map[string]bool{}} |
| 141 | } |
| 142 | |
| 143 | // WithDestroyedChecker makes saves for destroyed parent sessions no-op. This is |
| 144 | // used when a background sub-agent is cancelled because its parent session was |
| 145 | // cleared or moved out of active history. |
| 146 | func (s *SubagentStore) WithDestroyedChecker(fn func(parentSession string) bool) *SubagentStore { |
| 147 | if s != nil { |
| 148 | s.destroyed = fn |
| 149 | } |
| 150 | return s |
| 151 | } |
| 152 | |
| 153 | // WithParentSessionProbe installs a process-local liveness check used before |
| 154 | // stale cleanup probes a parent transcript lease. Desktop supplies this for |
| 155 | // tabs and builds that are live before their durable lease is bound. A nil |
| 156 | // probe preserves the lease-only behavior used by CLI and server frontends. |
| 157 | func (s *SubagentStore) WithParentSessionProbe(fn func(sessionPath string) bool) *SubagentStore { |
| 158 | if s != nil { |
| 159 | s.parentSessionProbe = fn |
| 160 | } |
| 161 | return s |
| 162 | } |
| 163 | |
| 164 | // ListSubagentsByParent returns persisted sub-agent artifacts whose metadata |
| 165 | // declares the given parent session owner. |
| 166 | func ListSubagentsByParent(sessionDir, parentSession string) ([]SubagentArtifact, error) { |
| 167 | parentSession = strings.TrimSpace(parentSession) |
| 168 | if strings.TrimSpace(sessionDir) == "" || parentSession == "" { |
| 169 | return nil, nil |
| 170 | } |
| 171 | dir := filepath.Join(sessionDir, "subagents") |
| 172 | entries, err := os.ReadDir(dir) |
| 173 | if err != nil { |
| 174 | if os.IsNotExist(err) { |
| 175 | return nil, nil |
| 176 | } |
| 177 | return nil, err |
| 178 | } |
| 179 | out := []SubagentArtifact{} |
| 180 | for _, entry := range entries { |
| 181 | if entry.IsDir() || !strings.HasSuffix(entry.Name(), ".meta.json") { |
| 182 | continue |
| 183 | } |
| 184 | ref := strings.TrimSuffix(entry.Name(), ".meta.json") |
| 185 | if !validSubagentRef(ref) { |
| 186 | continue |
| 187 | } |
| 188 | metaPath := filepath.Join(dir, entry.Name()) |
| 189 | data, err := fileencoding.ReadFileUTF8(metaPath) |
| 190 | if err != nil { |
| 191 | return nil, err |
| 192 | } |
| 193 | var meta SubagentMeta |
| 194 | if err := json.Unmarshal(data, &meta); err != nil { |
| 195 | continue |
| 196 | } |
| 197 | if strings.TrimSpace(meta.ParentSession) != parentSession { |
| 198 | continue |
| 199 | } |
| 200 | out = append(out, SubagentArtifact{ |
| 201 | Ref: ref, |
| 202 | SessionPath: filepath.Join(dir, ref+".jsonl"), |
| 203 | MetaPath: metaPath, |
| 204 | Meta: meta, |
| 205 | }) |
| 206 | } |
| 207 | return out, nil |
| 208 | } |
| 209 | |
| 210 | // DeleteSubagentsByParent permanently removes sub-agent artifacts owned by a |
| 211 | // parent session. Missing counterpart files are ignored. |
| 212 | func DeleteSubagentsByParent(sessionDir, parentSession string) error { |
| 213 | artifacts, err := ListSubagentsByParent(sessionDir, parentSession) |
| 214 | if err != nil { |
| 215 | return err |
| 216 | } |
| 217 | for _, artifact := range artifacts { |
| 218 | paths := []string{artifact.SessionPath, artifact.MetaPath} |
| 219 | // Sub-agent saves are single-file today, but sweep transcript sidecars |
| 220 | // (event log, event index, …) so no earlier build's artifacts survive |
| 221 | // the delete. |
| 222 | paths = append(paths, store.SessionSidecarFiles(artifact.SessionPath)...) |
| 223 | for _, path := range paths { |
| 224 | if path == "" { |
| 225 | continue |
| 226 | } |
| 227 | if err := os.Remove(path); err != nil && !os.IsNotExist(err) { |
| 228 | return err |
| 229 | } |
| 230 | } |
| 231 | } |
| 232 | return nil |
| 233 | } |
| 234 | |
| 235 | // CleanupStaleRunning marks persisted running sub-agents as interrupted while |
| 236 | // holding each parent transcript's session lease. A live parent in this or |
| 237 | // another process therefore keeps its children untouched, while crash leftovers |
| 238 | // remain repairable before this process accepts new background work. |
| 239 | func (s *SubagentStore) CleanupStaleRunning() (int, error) { |
| 240 | if s == nil { |
| 241 | return 0, nil |
| 242 | } |
| 243 | // On Windows, os.ReadDir can report ERROR_DIRECTORY as an IsNotExist |
| 244 | // error when the store path exists but is a regular file. Check the leaf |
| 245 | // first so a malformed store remains a startup error instead of being |
| 246 | // mistaken for an absent store. |
| 247 | info, err := os.Stat(s.dir) |
| 248 | if err != nil { |
| 249 | if os.IsNotExist(err) { |
| 250 | return 0, nil |
| 251 | } |
| 252 | return 0, err |
| 253 | } |
| 254 | if !info.IsDir() { |
| 255 | return 0, fmt.Errorf("subagent store path %q is not a directory", s.dir) |
| 256 | } |
| 257 | entries, err := os.ReadDir(s.dir) |
| 258 | if err != nil { |
| 259 | // Windows reports a non-directory at this path as ENOENT, so a plain |
| 260 | // IsNotExist check would silently accept a corrupt store instead of |
| 261 | // surfacing it. Only treat it as "no store yet" when nothing is there. |
| 262 | if os.IsNotExist(err) { |
| 263 | if _, statErr := os.Lstat(s.dir); statErr != nil { |
| 264 | return 0, nil |
| 265 | } |
| 266 | } |
| 267 | return 0, err |
| 268 | } |
| 269 | type staleParent struct { |
| 270 | sessionPath string |
| 271 | refs []string |
| 272 | } |
| 273 | parents := map[string]*staleParent{} |
| 274 | for _, entry := range entries { |
| 275 | if entry.IsDir() || !strings.HasSuffix(entry.Name(), ".meta.json") { |
| 276 | continue |
| 277 | } |
| 278 | ref := strings.TrimSuffix(entry.Name(), ".meta.json") |
| 279 | if !validSubagentRef(ref) { |
| 280 | continue |
| 281 | } |
| 282 | meta, err := s.LoadMeta(ref) |
| 283 | if err != nil { |
| 284 | // A corrupt metadata file (truncated write, killed process) must |
| 285 | // not abort startup. Skip all content decode failures, including |
| 286 | // errors from custom field decoders such as time.Time, while genuine |
| 287 | // file I/O errors remain fatal so storage problems stay visible. |
| 288 | if isSubagentMetaDecodeError(err) { |
| 289 | continue |
| 290 | } |
| 291 | return 0, err |
| 292 | } |
| 293 | if meta.Status != SubagentRunning { |
| 294 | continue |
| 295 | } |
| 296 | parentSession := strings.TrimSpace(meta.ParentSession) |
| 297 | sessionPath, ok := s.parentSessionPath(parentSession) |
| 298 | if !ok { |
| 299 | // Old or malformed metadata without a provable parent cannot |
| 300 | // authorize a destructive lifecycle rewrite. |
| 301 | continue |
| 302 | } |
| 303 | parent := parents[parentSession] |
| 304 | if parent == nil { |
| 305 | parent = &staleParent{sessionPath: sessionPath} |
| 306 | parents[parentSession] = parent |
| 307 | } |
| 308 | parent.refs = append(parent.refs, ref) |
| 309 | } |
| 310 | |
| 311 | parentIDs := make([]string, 0, len(parents)) |
| 312 | for parentID := range parents { |
| 313 | parentIDs = append(parentIDs, parentID) |
| 314 | } |
| 315 | sort.Strings(parentIDs) |
| 316 | |
| 317 | now := time.Now().UTC() |
| 318 | cleaned := 0 |
| 319 | for _, parentID := range parentIDs { |
| 320 | parent := parents[parentID] |
| 321 | if s.parentSessionProbe != nil && s.parentSessionProbe(parent.sessionPath) { |
| 322 | continue |
| 323 | } |
| 324 | lease, err := TryAcquireSessionLease(parent.sessionPath) |
| 325 | if errors.Is(err, ErrSessionLeaseHeld) { |
| 326 | continue |
| 327 | } |
| 328 | if err != nil { |
| 329 | return cleaned, fmt.Errorf("acquire parent session lease %q: %w", parentID, err) |
| 330 | } |
| 331 | for _, ref := range parent.refs { |
| 332 | if s.cleanupBeforeReread != nil { |
| 333 | s.cleanupBeforeReread(parentID, ref) |
| 334 | } |
| 335 | // Re-read after acquiring the parent lease: the former owner may |
| 336 | // have completed the child between the initial scan and handoff. |
| 337 | meta, err := s.LoadMeta(ref) |
| 338 | if err != nil { |
| 339 | if isSubagentMetaDecodeError(err) { |
| 340 | continue |
| 341 | } |
| 342 | lease.Release() |
| 343 | return cleaned, err |
| 344 | } |
| 345 | if meta.Status != SubagentRunning || strings.TrimSpace(meta.ParentSession) != parentID { |
| 346 | continue |
| 347 | } |
| 348 | meta.Status = SubagentInterrupted |
| 349 | meta.UpdatedAt = now |
| 350 | if err := s.saveMeta(meta); err != nil { |
| 351 | lease.Release() |
| 352 | return cleaned, err |
| 353 | } |
| 354 | cleaned++ |
| 355 | } |
| 356 | lease.Release() |
| 357 | } |
| 358 | return cleaned, nil |
| 359 | } |
| 360 | |
| 361 | func (s *SubagentStore) parentSessionPath(parentSession string) (string, bool) { |
| 362 | parentSession = strings.TrimSpace(parentSession) |
| 363 | if parentSession == "" || parentSession == "." || parentSession == ".." || filepath.Base(parentSession) != parentSession { |
| 364 | return "", false |
| 365 | } |
| 366 | return filepath.Join(filepath.Dir(s.dir), parentSession+".jsonl"), true |
| 367 | } |
| 368 | |
| 369 | func (s *SubagentStore) PrepareFresh(spec SubagentSpec) (*SubagentRun, error) { |
| 370 | if s == nil { |
| 371 | return nil, fmt.Errorf("subagent transcript store is required") |
| 372 | } |
| 373 | if err := requireParentSession(spec); err != nil { |
| 374 | return nil, err |
| 375 | } |
| 376 | ref, err := s.newRef() |
| 377 | if err != nil { |
| 378 | return nil, err |
| 379 | } |
| 380 | release, err := s.lock(ref) |
| 381 | if err != nil { |
| 382 | return nil, err |
| 383 | } |
| 384 | now := time.Now().UTC() |
| 385 | meta := metaFromSpec(ref, SubagentRunning, now, now, spec) |
| 386 | return &SubagentRun{Ref: ref, Session: NewSession(spec.SystemPrompt), Meta: meta, store: s, release: release}, nil |
| 387 | } |
| 388 | |
| 389 | func (s *SubagentStore) PrepareContinue(ref string, spec SubagentSpec) (*SubagentRun, error) { |
| 390 | if s == nil { |
| 391 | return nil, fmt.Errorf("subagent continuation is not available in this session") |
| 392 | } |
| 393 | if err := requireParentSession(spec); err != nil { |
| 394 | return nil, err |
| 395 | } |
| 396 | ref = strings.TrimSpace(ref) |
| 397 | if ref == "" { |
| 398 | return nil, fmt.Errorf("continue_from requires a subagent reference") |
| 399 | } |
| 400 | release, err := s.lock(ref) |
| 401 | if err != nil { |
| 402 | return nil, err |
| 403 | } |
| 404 | meta, err := s.LoadMeta(ref) |
| 405 | if err != nil { |
| 406 | release() |
| 407 | return nil, err |
| 408 | } |
| 409 | if strings.TrimSpace(meta.ParentSession) != strings.TrimSpace(spec.ParentSession) { |
| 410 | release() |
| 411 | return s.prepareContinueFromAncestor(ref, spec) |
| 412 | } |
| 413 | if err := validateContinueOwner(meta, spec); err != nil { |
| 414 | release() |
| 415 | return nil, err |
| 416 | } |
| 417 | if err := validateMeta(meta, spec); err != nil { |
| 418 | release() |
| 419 | return nil, err |
| 420 | } |
| 421 | sess, err := LoadSession(s.sessionPath(ref)) |
| 422 | if err != nil { |
| 423 | release() |
| 424 | return nil, fmt.Errorf("load subagent transcript %q: %w", ref, err) |
| 425 | } |
| 426 | meta.ParentSession = spec.ParentSession |
| 427 | meta.ParentToolCallID = spec.ParentToolCallID |
| 428 | return &SubagentRun{Ref: ref, Session: sess, Meta: meta, store: s, release: release}, nil |
| 429 | } |
| 430 | |
| 431 | func (s *SubagentStore) PrepareLegacyForkFrom(ref string, spec SubagentSpec) (*SubagentRun, error) { |
| 432 | if s == nil { |
| 433 | return nil, fmt.Errorf("subagent continuation is not available in this session") |
| 434 | } |
| 435 | if err := requireParentSession(spec); err != nil { |
| 436 | return nil, err |
| 437 | } |
| 438 | ref = strings.TrimSpace(ref) |
| 439 | if ref == "" { |
| 440 | return nil, fmt.Errorf("fork_from requires a subagent reference") |
| 441 | } |
| 442 | release, err := s.lock(ref) |
| 443 | if err != nil { |
| 444 | return nil, err |
| 445 | } |
| 446 | meta, err := s.LoadMeta(ref) |
| 447 | if err != nil { |
| 448 | release() |
| 449 | return nil, err |
| 450 | } |
| 451 | owner := strings.TrimSpace(meta.ParentSession) |
| 452 | current := strings.TrimSpace(spec.ParentSession) |
| 453 | if owner == "" { |
| 454 | release() |
| 455 | return nil, fmt.Errorf("subagent reference %q has no parent session; run a fresh subagent instead", ref) |
| 456 | } |
| 457 | if owner == current { |
| 458 | release() |
| 459 | if err := validateMeta(meta, spec); err != nil { |
| 460 | return nil, err |
| 461 | } |
| 462 | return nil, fmt.Errorf("fork_from cannot be safely converted for subagent reference %q in the current conversation; use continue_from to continue it in place or start a fresh subagent for independent work", ref) |
| 463 | } |
| 464 | release() |
| 465 | return s.PrepareContinue(ref, spec) |
| 466 | } |
| 467 | |
| 468 | func (s *SubagentStore) prepareContinueFromAncestor(sourceRef string, spec SubagentSpec) (*SubagentRun, error) { |
| 469 | sourceRef, err := s.nearestLineageSource(sourceRef, spec) |
| 470 | if err != nil { |
| 471 | return nil, err |
| 472 | } |
| 473 | copies, err := s.compatibleCopiesFromSource(sourceRef, spec) |
| 474 | if err != nil { |
| 475 | return nil, err |
| 476 | } |
| 477 | switch len(copies) { |
| 478 | case 0: |
| 479 | return s.prepareFork(sourceRef, spec) |
| 480 | case 1: |
| 481 | run, err := s.PrepareContinue(copies[0].Ref, spec) |
| 482 | if err != nil { |
| 483 | return nil, err |
| 484 | } |
| 485 | run.ForkedFrom = sourceRef |
| 486 | return run, nil |
| 487 | default: |
| 488 | return nil, fmt.Errorf("subagent reference %q has multiple copied transcripts in current parent session %q", sourceRef, spec.ParentSession) |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | func (s *SubagentStore) nearestLineageSource(requestedRef string, spec SubagentSpec) (string, error) { |
| 493 | ancestors, err := s.sessionAncestors(spec.ParentSession) |
| 494 | if err != nil { |
| 495 | return "", err |
| 496 | } |
| 497 | for _, ancestor := range ancestors { |
| 498 | artifacts, err := ListSubagentsByParent(filepath.Dir(s.dir), ancestor) |
| 499 | if err != nil { |
| 500 | return "", err |
| 501 | } |
| 502 | var candidates []SubagentArtifact |
| 503 | for _, artifact := range artifacts { |
| 504 | if artifact.Ref == requestedRef || s.derivesFrom(artifact.Meta, requestedRef) { |
| 505 | candidates = append(candidates, artifact) |
| 506 | } |
| 507 | } |
| 508 | if len(candidates) > 1 { |
| 509 | return "", fmt.Errorf("subagent reference %q has multiple candidate transcripts in ancestor parent session %q", requestedRef, ancestor) |
| 510 | } |
| 511 | if len(candidates) == 1 { |
| 512 | if err := validateMeta(candidates[0].Meta, spec); err != nil { |
| 513 | return "", err |
| 514 | } |
| 515 | return candidates[0].Ref, nil |
| 516 | } |
| 517 | } |
| 518 | return "", fmt.Errorf("subagent reference %q is not in current parent session %q lineage", requestedRef, spec.ParentSession) |
| 519 | } |
| 520 | |
| 521 | func (s *SubagentStore) sessionAncestors(current string) ([]string, error) { |
| 522 | current = strings.TrimSpace(current) |
| 523 | if current == "" { |
| 524 | return nil, nil |
| 525 | } |
| 526 | var ancestors []string |
| 527 | seen := map[string]bool{} |
| 528 | for cursor := current; cursor != ""; { |
| 529 | if seen[cursor] { |
| 530 | return nil, fmt.Errorf("cycle at session %q", cursor) |
| 531 | } |
| 532 | seen[cursor] = true |
| 533 | // Route through parentSessionPath so both the caller-provided id and |
| 534 | // every parent id read from disk metadata get the same bare-filename |
| 535 | // validation — a raw Join would let "../"-shaped ids escape the |
| 536 | // session directory. |
| 537 | metaPath, valid := s.parentSessionPath(cursor) |
| 538 | if !valid { |
| 539 | return nil, fmt.Errorf("invalid session identifier %q", cursor) |
| 540 | } |
| 541 | meta, ok, err := LoadBranchMeta(metaPath) |
| 542 | if err != nil { |
| 543 | return nil, err |
| 544 | } |
| 545 | if !ok { |
| 546 | return nil, fmt.Errorf("missing branch metadata for session %q", cursor) |
| 547 | } |
| 548 | if strings.TrimSpace(meta.ID) != cursor { |
| 549 | return nil, fmt.Errorf("branch metadata for session %q declares id %q", cursor, meta.ID) |
| 550 | } |
| 551 | parent := strings.TrimSpace(meta.ParentID) |
| 552 | if parent == "" { |
| 553 | break |
| 554 | } |
| 555 | ancestors = append(ancestors, parent) |
| 556 | cursor = parent |
| 557 | } |
| 558 | return ancestors, nil |
| 559 | } |
| 560 | |
| 561 | func (s *SubagentStore) derivesFrom(meta SubagentMeta, sourceRef string) bool { |
| 562 | sourceRef = strings.TrimSpace(sourceRef) |
| 563 | seen := map[string]bool{} |
| 564 | for cursor := strings.TrimSpace(meta.ForkedFrom); cursor != ""; { |
| 565 | if cursor == sourceRef { |
| 566 | return true |
| 567 | } |
| 568 | if seen[cursor] { |
| 569 | return false |
| 570 | } |
| 571 | seen[cursor] = true |
| 572 | parent, err := s.LoadMeta(cursor) |
| 573 | if err != nil { |
| 574 | return false |
| 575 | } |
| 576 | cursor = strings.TrimSpace(parent.ForkedFrom) |
| 577 | } |
| 578 | return false |
| 579 | } |
| 580 | |
| 581 | func (s *SubagentStore) compatibleCopiesFromSource(sourceRef string, spec SubagentSpec) ([]SubagentArtifact, error) { |
| 582 | artifacts, err := ListSubagentsByParent(filepath.Dir(s.dir), spec.ParentSession) |
| 583 | if err != nil { |
| 584 | return nil, err |
| 585 | } |
| 586 | var copies []SubagentArtifact |
| 587 | for _, artifact := range artifacts { |
| 588 | if strings.TrimSpace(artifact.Meta.ForkedFrom) != sourceRef { |
| 589 | continue |
| 590 | } |
| 591 | if err := validateMeta(artifact.Meta, spec); err != nil { |
| 592 | return nil, err |
| 593 | } |
| 594 | copies = append(copies, artifact) |
| 595 | } |
| 596 | return copies, nil |
| 597 | } |
| 598 | |
| 599 | func (s *SubagentStore) prepareFork(ref string, spec SubagentSpec) (*SubagentRun, error) { |
| 600 | if s == nil { |
| 601 | return nil, fmt.Errorf("subagent continuation is not available in this session") |
| 602 | } |
| 603 | if err := requireParentSession(spec); err != nil { |
| 604 | return nil, err |
| 605 | } |
| 606 | sourceRef := strings.TrimSpace(ref) |
| 607 | if sourceRef == "" { |
| 608 | return nil, fmt.Errorf("subagent copy requires a source reference") |
| 609 | } |
| 610 | sourceRelease, err := s.lock(sourceRef) |
| 611 | if err != nil { |
| 612 | return nil, err |
| 613 | } |
| 614 | meta, err := s.LoadMeta(sourceRef) |
| 615 | if err != nil { |
| 616 | sourceRelease() |
| 617 | return nil, err |
| 618 | } |
| 619 | if strings.TrimSpace(meta.ParentSession) == "" { |
| 620 | sourceRelease() |
| 621 | return nil, fmt.Errorf("subagent reference %q has no parent session; run a fresh subagent instead", sourceRef) |
| 622 | } |
| 623 | if err := validateMeta(meta, spec); err != nil { |
| 624 | sourceRelease() |
| 625 | return nil, err |
| 626 | } |
| 627 | if err := s.validateForkOwner(meta, spec); err != nil { |
| 628 | sourceRelease() |
| 629 | return nil, err |
| 630 | } |
| 631 | sess, err := LoadSession(s.sessionPath(sourceRef)) |
| 632 | if err != nil { |
| 633 | sourceRelease() |
| 634 | return nil, fmt.Errorf("load subagent transcript %q: %w", sourceRef, err) |
| 635 | } |
| 636 | sourceRelease() |
| 637 | newRef, err := s.newRef() |
| 638 | if err != nil { |
| 639 | return nil, err |
| 640 | } |
| 641 | newRelease, err := s.lock(newRef) |
| 642 | if err != nil { |
| 643 | return nil, err |
| 644 | } |
| 645 | now := time.Now().UTC() |
| 646 | newMeta := metaFromSpec(newRef, SubagentRunning, now, now, spec) |
| 647 | newMeta.ForkedFrom = sourceRef |
| 648 | return &SubagentRun{Ref: newRef, Session: sess, Meta: newMeta, ForkedFrom: sourceRef, store: s, release: newRelease}, nil |
| 649 | } |
| 650 | |
| 651 | func (s *SubagentStore) MarkRunning(run *SubagentRun) error { |
| 652 | if s == nil || run == nil || run.Ref == "" { |
| 653 | return nil |
| 654 | } |
| 655 | if s.parentDestroyed(run) { |
| 656 | return nil |
| 657 | } |
| 658 | meta := run.Meta |
| 659 | meta.Status = SubagentRunning |
| 660 | meta.UpdatedAt = time.Now().UTC() |
| 661 | return s.saveMeta(meta) |
| 662 | } |
| 663 | |
| 664 | func (s *SubagentStore) SaveCompleted(run *SubagentRun) error { |
| 665 | if s == nil || run == nil || run.Ref == "" { |
| 666 | return nil |
| 667 | } |
| 668 | if s.parentDestroyed(run) { |
| 669 | return nil |
| 670 | } |
| 671 | if err := s.ensureBranchCreatedAt(run); err != nil { |
| 672 | return err |
| 673 | } |
| 674 | if err := run.Session.Save(s.sessionPath(run.Ref)); err != nil { |
| 675 | return err |
| 676 | } |
| 677 | meta := run.Meta |
| 678 | meta.Status = SubagentCompleted |
| 679 | meta.UpdatedAt = time.Now().UTC() |
| 680 | run.Meta = meta |
| 681 | return s.saveMeta(meta) |
| 682 | } |
| 683 | |
| 684 | func (s *SubagentStore) SaveFailed(run *SubagentRun) error { |
| 685 | if s == nil || run == nil || run.Ref == "" { |
| 686 | return nil |
| 687 | } |
| 688 | if s.parentDestroyed(run) { |
| 689 | return nil |
| 690 | } |
| 691 | // Terminal status is independent from transcript persistence. Keep going so |
| 692 | // a sidecar failure cannot leave a failed run marked as running on disk. |
| 693 | branchErr := s.ensureBranchCreatedAt(run) |
| 694 | var sessionErr error |
| 695 | if run.Session != nil { |
| 696 | sessionErr = run.Session.Save(s.sessionPath(run.Ref)) |
| 697 | } |
| 698 | meta := run.Meta |
| 699 | meta.Status = SubagentFailed |
| 700 | meta.UpdatedAt = time.Now().UTC() |
| 701 | run.Meta = meta |
| 702 | return errors.Join(branchErr, sessionErr, s.saveMeta(meta)) |
| 703 | } |
| 704 | |
| 705 | // ensureBranchCreatedAt seeds the session list sidecar before the first |
| 706 | // transcript save. Subagent transcripts are written only on completion, so |
| 707 | // Session.Save would otherwise backfill BranchMeta.CreatedAt with the save |
| 708 | // moment (completion time). The real start time already lives on run.Meta. |
| 709 | func (s *SubagentStore) ensureBranchCreatedAt(run *SubagentRun) error { |
| 710 | if s == nil || run == nil || run.Ref == "" { |
| 711 | return nil |
| 712 | } |
| 713 | path := s.sessionPath(run.Ref) |
| 714 | if _, ok, err := LoadBranchMeta(path); err != nil { |
| 715 | return err |
| 716 | } else if ok { |
| 717 | return nil |
| 718 | } |
| 719 | created := run.Meta.CreatedAt.UTC() |
| 720 | if created.IsZero() { |
| 721 | created = time.Now().UTC() |
| 722 | } |
| 723 | return SaveBranchMetaPreserveUpdated(path, BranchMeta{ |
| 724 | ID: BranchID(path), |
| 725 | CreatedAt: created, |
| 726 | }) |
| 727 | } |
| 728 | |
| 729 | func (s *SubagentStore) LoadMeta(ref string) (SubagentMeta, error) { |
| 730 | var meta SubagentMeta |
| 731 | if !validSubagentRef(ref) { |
| 732 | return meta, fmt.Errorf("invalid subagent reference %q", ref) |
| 733 | } |
| 734 | data, err := fileencoding.ReadFileUTF8(s.metaPath(ref)) |
| 735 | if err != nil { |
| 736 | return meta, fmt.Errorf("load subagent metadata %q: %w", ref, err) |
| 737 | } |
| 738 | if err := json.Unmarshal(data, &meta); err != nil { |
| 739 | return meta, &subagentMetaDecodeError{ref: ref, err: err} |
| 740 | } |
| 741 | return meta, nil |
| 742 | } |
| 743 | |
| 744 | func metaFromSpec(ref string, status SubagentStatus, created, updated time.Time, spec SubagentSpec) SubagentMeta { |
| 745 | scope, schemaHash := toolIdentity(spec.Registry) |
| 746 | return SubagentMeta{ |
| 747 | Ref: ref, |
| 748 | CreatedAt: created, |
| 749 | UpdatedAt: updated, |
| 750 | Status: status, |
| 751 | Kind: strings.TrimSpace(spec.Kind), |
| 752 | Name: strings.TrimSpace(spec.Name), |
| 753 | WorkspaceRoot: strings.TrimSpace(spec.WorkspaceRoot), |
| 754 | ParentSession: strings.TrimSpace(spec.ParentSession), |
| 755 | ParentToolCallID: strings.TrimSpace(spec.ParentToolCallID), |
| 756 | SystemPromptHash: bytesHash([]byte(spec.SystemPrompt)), |
| 757 | ToolScope: scope, |
| 758 | ToolSchemaHash: schemaHash, |
| 759 | Model: strings.TrimSpace(spec.Model), |
| 760 | Effort: strings.TrimSpace(spec.Effort), |
| 761 | } |
| 762 | } |
| 763 | |
| 764 | func validateMeta(meta SubagentMeta, spec SubagentSpec) error { |
| 765 | if meta.Status == SubagentRunning { |
| 766 | return fmt.Errorf("subagent reference %q is still in progress", meta.Ref) |
| 767 | } |
| 768 | if meta.Status == SubagentFailed { |
| 769 | return fmt.Errorf("subagent reference %q failed and cannot be continued", meta.Ref) |
| 770 | } |
| 771 | if meta.Status == SubagentInterrupted { |
| 772 | return fmt.Errorf("subagent reference %q was interrupted by a previous shutdown or crash and cannot be continued or forked; run a fresh subagent instead", meta.Ref) |
| 773 | } |
| 774 | want := metaFromSpec(meta.Ref, meta.Status, meta.CreatedAt, meta.UpdatedAt, spec) |
| 775 | switch { |
| 776 | case meta.Kind != want.Kind: |
| 777 | return fmt.Errorf("subagent reference %q has kind %q, want %q", meta.Ref, meta.Kind, want.Kind) |
| 778 | case meta.Name != want.Name: |
| 779 | return fmt.Errorf("subagent reference %q has name %q, want %q", meta.Ref, meta.Name, want.Name) |
| 780 | case meta.WorkspaceRoot != want.WorkspaceRoot: |
| 781 | return fmt.Errorf("subagent reference %q belongs to workspace %q, current workspace is %q", meta.Ref, meta.WorkspaceRoot, want.WorkspaceRoot) |
| 782 | case meta.SystemPromptHash != want.SystemPromptHash: |
| 783 | return fmt.Errorf("subagent reference %q uses a different subagent persona; run a fresh subagent to use the current persona", meta.Ref) |
| 784 | case !sameStrings(meta.ToolScope, want.ToolScope): |
| 785 | return fmt.Errorf("subagent reference %q uses a different tool scope", meta.Ref) |
| 786 | case meta.ToolSchemaHash != want.ToolSchemaHash: |
| 787 | return fmt.Errorf("subagent reference %q uses different tool schemas", meta.Ref) |
| 788 | case meta.Model != want.Model || meta.Effort != want.Effort: |
| 789 | return fmt.Errorf("subagent reference %q uses model/effort %q/%q, current run would use %q/%q", meta.Ref, meta.Model, meta.Effort, want.Model, want.Effort) |
| 790 | } |
| 791 | return nil |
| 792 | } |
| 793 | |
| 794 | func requireParentSession(spec SubagentSpec) error { |
| 795 | if strings.TrimSpace(spec.ParentSession) == "" { |
| 796 | return fmt.Errorf("subagent transcript parent session is required") |
| 797 | } |
| 798 | return nil |
| 799 | } |
| 800 | |
| 801 | func validateContinueOwner(meta SubagentMeta, spec SubagentSpec) error { |
| 802 | current := strings.TrimSpace(spec.ParentSession) |
| 803 | owner := strings.TrimSpace(meta.ParentSession) |
| 804 | if owner == current { |
| 805 | return nil |
| 806 | } |
| 807 | if owner == "" { |
| 808 | return fmt.Errorf("subagent reference %q has no parent session; run a fresh subagent instead", meta.Ref) |
| 809 | } |
| 810 | return fmt.Errorf("subagent reference %q belongs to parent session %q, current parent session is %q", meta.Ref, owner, current) |
| 811 | } |
| 812 | |
| 813 | func (s *SubagentStore) validateForkOwner(meta SubagentMeta, spec SubagentSpec) error { |
| 814 | current := strings.TrimSpace(spec.ParentSession) |
| 815 | owner := strings.TrimSpace(meta.ParentSession) |
| 816 | if owner == current { |
| 817 | return nil |
| 818 | } |
| 819 | if owner == "" { |
| 820 | return fmt.Errorf("subagent reference %q has no parent session; run a fresh subagent instead", meta.Ref) |
| 821 | } |
| 822 | ok, err := s.isAncestorSession(owner, current) |
| 823 | if err != nil { |
| 824 | return fmt.Errorf("subagent reference %q belongs to parent session %q, but current parent session %q lineage could not be verified: %w", meta.Ref, owner, current, err) |
| 825 | } |
| 826 | if ok { |
| 827 | return nil |
| 828 | } |
| 829 | return fmt.Errorf("subagent reference %q belongs to parent session %q, which is not in current parent session %q lineage", meta.Ref, owner, current) |
| 830 | } |
| 831 | |
| 832 | func (s *SubagentStore) isAncestorSession(ancestor, current string) (bool, error) { |
| 833 | ancestor = strings.TrimSpace(ancestor) |
| 834 | current = strings.TrimSpace(current) |
| 835 | if ancestor == "" || current == "" { |
| 836 | return false, nil |
| 837 | } |
| 838 | seen := map[string]bool{} |
| 839 | for cursor := current; cursor != ""; { |
| 840 | if seen[cursor] { |
| 841 | return false, fmt.Errorf("cycle at session %q", cursor) |
| 842 | } |
| 843 | seen[cursor] = true |
| 844 | // Route through parentSessionPath so both the caller-provided id and |
| 845 | // every parent id read from disk metadata get the same bare-filename |
| 846 | // validation — a raw Join would let "../"-shaped ids escape the |
| 847 | // session directory. |
| 848 | metaPath, valid := s.parentSessionPath(cursor) |
| 849 | if !valid { |
| 850 | return false, fmt.Errorf("invalid session identifier %q", cursor) |
| 851 | } |
| 852 | meta, ok, err := LoadBranchMeta(metaPath) |
| 853 | if err != nil { |
| 854 | return false, err |
| 855 | } |
| 856 | if !ok { |
| 857 | return false, fmt.Errorf("missing branch metadata for session %q", cursor) |
| 858 | } |
| 859 | if strings.TrimSpace(meta.ID) != cursor { |
| 860 | return false, fmt.Errorf("branch metadata for session %q declares id %q", cursor, meta.ID) |
| 861 | } |
| 862 | if cursor == ancestor { |
| 863 | return true, nil |
| 864 | } |
| 865 | parent := strings.TrimSpace(meta.ParentID) |
| 866 | cursor = parent |
| 867 | } |
| 868 | return false, nil |
| 869 | } |
| 870 | |
| 871 | func (s *SubagentStore) lock(ref string) (func(), error) { |
| 872 | if !validSubagentRef(ref) { |
| 873 | return nil, fmt.Errorf("invalid subagent reference %q", ref) |
| 874 | } |
| 875 | s.mu.Lock() |
| 876 | defer s.mu.Unlock() |
| 877 | if s.locked[ref] { |
| 878 | return nil, fmt.Errorf("subagent reference %q is already running; retry after it finishes", ref) |
| 879 | } |
| 880 | s.locked[ref] = true |
| 881 | return func() { |
| 882 | s.mu.Lock() |
| 883 | delete(s.locked, ref) |
| 884 | s.mu.Unlock() |
| 885 | }, nil |
| 886 | } |
| 887 | |
| 888 | func (s *SubagentStore) newRef() (string, error) { |
| 889 | var b [6]byte |
| 890 | if _, err := rand.Read(b[:]); err != nil { |
| 891 | return "", err |
| 892 | } |
| 893 | return "sa_" + time.Now().UTC().Format("20060102_150405_000000000") + "_" + hex.EncodeToString(b[:]), nil |
| 894 | } |
| 895 | |
| 896 | func (s *SubagentStore) sessionPath(ref string) string { return filepath.Join(s.dir, ref+".jsonl") } |
| 897 | func (s *SubagentStore) metaPath(ref string) string { return filepath.Join(s.dir, ref+".meta.json") } |
| 898 | |
| 899 | func (s *SubagentStore) saveMeta(meta SubagentMeta) error { |
| 900 | if err := os.MkdirAll(s.dir, 0o755); err != nil { |
| 901 | return err |
| 902 | } |
| 903 | data, err := json.MarshalIndent(meta, "", " ") |
| 904 | if err != nil { |
| 905 | return err |
| 906 | } |
| 907 | data = append(data, '\n') |
| 908 | tmp, err := os.CreateTemp(s.dir, ".subagent-meta.*.tmp") |
| 909 | if err != nil { |
| 910 | return err |
| 911 | } |
| 912 | tmpPath := tmp.Name() |
| 913 | if _, err := tmp.Write(data); err != nil { |
| 914 | tmp.Close() |
| 915 | os.Remove(tmpPath) |
| 916 | return err |
| 917 | } |
| 918 | if err := tmp.Close(); err != nil { |
| 919 | os.Remove(tmpPath) |
| 920 | return err |
| 921 | } |
| 922 | return fileutil.ReplaceFile(tmpPath, s.metaPath(meta.Ref)) |
| 923 | } |
| 924 | |
| 925 | func (s *SubagentStore) parentDestroyed(run *SubagentRun) bool { |
| 926 | if s == nil || s.destroyed == nil || run == nil { |
| 927 | return false |
| 928 | } |
| 929 | return s.destroyed(run.Meta.ParentSession) |
| 930 | } |
| 931 | |
| 932 | func validSubagentRef(ref string) bool { |
| 933 | if !strings.HasPrefix(ref, "sa_") { |
| 934 | return false |
| 935 | } |
| 936 | for _, r := range ref { |
| 937 | if (r >= 'a' && r <= 'z') || (r >= 'A' && r <= 'Z') || (r >= '0' && r <= '9') || r == '_' || r == '-' { |
| 938 | continue |
| 939 | } |
| 940 | return false |
| 941 | } |
| 942 | return true |
| 943 | } |
| 944 | |
| 945 | func toolIdentity(reg *tool.Registry) ([]string, string) { |
| 946 | if reg == nil { |
| 947 | return nil, bytesHash(nil) |
| 948 | } |
| 949 | names := reg.Names() |
| 950 | sort.Strings(names) |
| 951 | schemas := normalizeToolSchemas(reg.Schemas()) |
| 952 | data, _ := json.Marshal(schemas) |
| 953 | return names, bytesHash(data) |
| 954 | } |
| 955 | |
| 956 | func bytesHash(data []byte) string { |
| 957 | h := sha256.Sum256(data) |
| 958 | return hex.EncodeToString(h[:]) |
| 959 | } |
| 960 | |
| 961 | func sameStrings(a, b []string) bool { |
| 962 | if len(a) != len(b) { |
| 963 | return false |
| 964 | } |
| 965 | for i := range a { |
| 966 | if a[i] != b[i] { |
| 967 | return false |
| 968 | } |
| 969 | } |
| 970 | return true |
| 971 | } |
| 972 |