@sema-agent/core 2.13.0 → 4.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (184) hide show
  1. package/CHANGELOG.md +2345 -0
  2. package/dist/agents/cascade.d.ts +6 -1
  3. package/dist/agents/cascade.js +76 -32
  4. package/dist/agents/cumulative-stats.d.ts +29 -0
  5. package/dist/agents/cumulative-stats.js +29 -0
  6. package/dist/agents/observer.d.ts +4 -0
  7. package/dist/agents/observer.js +11 -10
  8. package/dist/agents/repair-loop.d.ts +2 -0
  9. package/dist/agents/repair-loop.js +110 -3
  10. package/dist/agents/retain-ledger.d.ts +20 -2
  11. package/dist/agents/retain-ledger.js +38 -8
  12. package/dist/agents/roster-store.d.ts +4 -0
  13. package/dist/agents/roster-store.js +20 -1
  14. package/dist/agents/subagent.js +66 -64
  15. package/dist/agents/teacher.d.ts +1 -9
  16. package/dist/agents/teacher.js +23 -24
  17. package/dist/agents/team.js +7 -4
  18. package/dist/agents/verify.d.ts +15 -2
  19. package/dist/agents/verify.js +39 -3
  20. package/dist/bin/sema-tb.js +2 -2
  21. package/dist/brain/anthropic.js +24 -21
  22. package/dist/brain/circuit-breaker.d.ts +1 -0
  23. package/dist/brain/circuit-breaker.js +26 -9
  24. package/dist/brain/degrading.js +13 -1
  25. package/dist/brain/failover.js +2 -0
  26. package/dist/brain/openai.js +24 -21
  27. package/dist/brain/repetition.js +10 -1
  28. package/dist/brain/routing.js +5 -1
  29. package/dist/brain/status-sink.js +5 -1
  30. package/dist/brain/stream-engine.js +5 -2
  31. package/dist/brain/stream-shared.d.ts +1 -0
  32. package/dist/brain/stream-shared.js +14 -0
  33. package/dist/brain/terminal-cause.d.ts +4 -0
  34. package/dist/brain/terminal-cause.js +12 -0
  35. package/dist/brain/tool-call-id.d.ts +1 -0
  36. package/dist/brain/tool-call-id.js +3 -0
  37. package/dist/brain/tool-call-repair.js +34 -15
  38. package/dist/core/auto-compaction.js +33 -32
  39. package/dist/core/background-agent-store.d.ts +22 -22
  40. package/dist/core/background-agent-store.js +19 -23
  41. package/dist/core/checkpoint-store.d.ts +7 -13
  42. package/dist/core/checkpoint-store.js +39 -15
  43. package/dist/core/consolidate-scope.js +4 -3
  44. package/dist/core/context-edit.js +3 -0
  45. package/dist/core/context-guard.js +3 -0
  46. package/dist/core/file-snapshot-store.d.ts +8 -0
  47. package/dist/core/file-snapshot-store.js +4 -3
  48. package/dist/core/git-worktree-env.d.ts +6 -1
  49. package/dist/core/git-worktree-env.js +12 -1
  50. package/dist/core/hooks.d.ts +4 -1
  51. package/dist/core/hooks.js +35 -7
  52. package/dist/core/lsp-diagnostics.js +13 -7
  53. package/dist/core/lsp-protocol.d.ts +1 -1
  54. package/dist/core/lsp-protocol.js +20 -8
  55. package/dist/core/lsp-session.d.ts +12 -2
  56. package/dist/core/lsp-session.js +114 -49
  57. package/dist/core/lsp.d.ts +13 -1
  58. package/dist/core/lsp.js +65 -9
  59. package/dist/core/mailbox-store.d.ts +7 -10
  60. package/dist/core/mcp.d.ts +7 -10
  61. package/dist/core/mcp.js +7 -3
  62. package/dist/core/memory-engine/data-plane.js +4 -0
  63. package/dist/core/memory-engine/engine.d.ts +2 -2
  64. package/dist/core/memory-engine/engine.js +96 -31
  65. package/dist/core/memory-engine/file-backend.d.ts +2 -2
  66. package/dist/core/memory-engine/file-backend.js +38 -39
  67. package/dist/core/memory-engine/layout.d.ts +8 -0
  68. package/dist/core/memory-engine/layout.js +71 -4
  69. package/dist/core/memory-engine/scope-contract.js +9 -3
  70. package/dist/core/memory-engine/types.d.ts +1 -1
  71. package/dist/core/memory.js +3 -0
  72. package/dist/core/permission-rules.d.ts +2 -1
  73. package/dist/core/permission-rules.js +24 -3
  74. package/dist/core/runner/active-skill-scope.d.ts +0 -1
  75. package/dist/core/runner/active-skill-scope.js +1 -15
  76. package/dist/core/runner/assemble-result.d.ts +3 -8
  77. package/dist/core/runner/assemble-result.js +3 -5
  78. package/dist/core/runner/compaction-call-options.d.ts +7 -0
  79. package/dist/core/runner/compaction-call-options.js +30 -0
  80. package/dist/core/runner/prepare-task.d.ts +6 -23
  81. package/dist/core/runner/prepare-task.js +31 -53
  82. package/dist/core/runner/runtask.d.ts +1 -0
  83. package/dist/core/runner/runtask.js +86 -178
  84. package/dist/core/runner/session-rule-policy.js +4 -1
  85. package/dist/core/runner/teardown-bounded.d.ts +7 -0
  86. package/dist/core/runner/teardown-bounded.js +36 -0
  87. package/dist/core/runner/tool-disclosure.d.ts +2 -5
  88. package/dist/core/runner/turn-attachments.d.ts +5 -15
  89. package/dist/core/runner/turn-attachments.js +1 -1
  90. package/dist/core/runner/usage-accounting.d.ts +6 -0
  91. package/dist/core/runner/usage-accounting.js +9 -8
  92. package/dist/core/safe-notify.d.ts +17 -0
  93. package/dist/core/safe-notify.js +59 -0
  94. package/dist/core/secret-env.js +5 -1
  95. package/dist/core/sensitive-path-policy.js +11 -4
  96. package/dist/core/store-contracts/checkpoint-store-contract.js +1 -1
  97. package/dist/core/store-contracts/session-repo-contract.js +40 -0
  98. package/dist/core/stub-env.d.ts +2 -6
  99. package/dist/core/surrogate-safe-slice.d.ts +4 -0
  100. package/dist/core/surrogate-safe-slice.js +18 -0
  101. package/dist/core/task-registry-monitor.js +15 -14
  102. package/dist/core/task-registry-shared.d.ts +22 -0
  103. package/dist/core/task-registry-shared.js +16 -0
  104. package/dist/core/task-registry.d.ts +1 -0
  105. package/dist/core/task-registry.js +40 -57
  106. package/dist/core/tool-errors.js +5 -4
  107. package/dist/core/tool-name-aliases.js +2 -1
  108. package/dist/core/tool-policy.js +72 -19
  109. package/dist/core/tool-result-store.d.ts +1 -0
  110. package/dist/core/tool-result-store.js +17 -1
  111. package/dist/core/trace.d.ts +3 -9
  112. package/dist/core/types.d.ts +16 -12
  113. package/dist/core/untrusted-egress.js +7 -5
  114. package/dist/core/workflow-journal-store.d.ts +9 -20
  115. package/dist/core/workflow-run-store.d.ts +11 -10
  116. package/dist/core/workflow-run-store.js +22 -16
  117. package/dist/engine/compaction/compaction.d.ts +1 -0
  118. package/dist/engine/compaction/compaction.js +14 -5
  119. package/dist/engine/compaction/utils.js +4 -2
  120. package/dist/engine/execution-env/node-execution-env.d.ts +2 -24
  121. package/dist/engine/harness/agent-harness.d.ts +1 -0
  122. package/dist/engine/harness/agent-harness.js +13 -33
  123. package/dist/engine/harness/types.d.ts +29 -67
  124. package/dist/engine/llm/types.d.ts +1 -1
  125. package/dist/engine/loop/agent-loop.js +11 -4
  126. package/dist/engine/loop/types.d.ts +11 -9
  127. package/dist/engine/lsp/frame-decoder.js +6 -3
  128. package/dist/engine/lsp/node-lsp-manager.d.ts +2 -0
  129. package/dist/engine/lsp/node-lsp-manager.js +103 -26
  130. package/dist/engine/lsp/stdio-lsp-transport.js +16 -7
  131. package/dist/engine/session/memory-repo.d.ts +2 -6
  132. package/dist/engine/session/session.d.ts +5 -30
  133. package/dist/index.d.ts +3 -1
  134. package/dist/index.js +2 -1
  135. package/dist/internal/harness-types.d.ts +3 -1
  136. package/dist/internal/harness.d.ts +1 -1
  137. package/dist/internal/harness.js +1 -1
  138. package/dist/orchestration/run-spec.js +7 -4
  139. package/dist/orchestration/run-workflow-tool.js +3 -35
  140. package/dist/orchestration/workflow-governance.js +43 -9
  141. package/dist/orchestration/workflow-sandbox-conformance.js +85 -51
  142. package/dist/orchestration/workflow-script-store.js +34 -4
  143. package/dist/orchestration/workflow.d.ts +3 -11
  144. package/dist/orchestration/workflow.js +164 -204
  145. package/dist/prompt-assembly/artifact-store.d.ts +2 -0
  146. package/dist/prompt-assembly/artifact-store.js +39 -24
  147. package/dist/prompt-assembly/assemble.js +38 -7
  148. package/dist/prompt-assembly/epoch.js +19 -2
  149. package/dist/prompt-assembly/event-registry.js +3 -3
  150. package/dist/prompt-assembly/explain.js +1 -1
  151. package/dist/prompt-assembly/tool-catalog.d.ts +10 -9
  152. package/dist/prompt-assembly/tool-catalog.js +16 -1
  153. package/dist/prompt-assembly/turn-snapshot.d.ts +2 -5
  154. package/dist/prompt-assembly/turn-snapshot.js +4 -1
  155. package/dist/stores/file/background-agent-store.d.ts +3 -12
  156. package/dist/stores/file/background-agent-store.js +3 -41
  157. package/dist/stores/file/checkpoint-store.js +6 -6
  158. package/dist/stores/file/fs-atomic.d.ts +3 -0
  159. package/dist/stores/file/fs-atomic.js +53 -8
  160. package/dist/stores/file/index.d.ts +1 -0
  161. package/dist/stores/file/index.js +7 -0
  162. package/dist/stores/file/mailbox-store.d.ts +2 -6
  163. package/dist/stores/file/session-policy-store.js +15 -1
  164. package/dist/stores/file/session-store.d.ts +4 -6
  165. package/dist/stores/file/session-store.js +28 -1
  166. package/dist/stores/file/workflow-journal-store.d.ts +3 -11
  167. package/dist/stores/file/workflow-run-store.d.ts +3 -7
  168. package/dist/stores/file/workflow-run-store.js +4 -17
  169. package/dist/tools/fs/bash-readonly-classifier.d.ts +2 -0
  170. package/dist/tools/fs/bash-readonly-classifier.js +73 -21
  171. package/dist/tools/fs/fs-bash.d.ts +1 -1
  172. package/dist/tools/fs/fs-bash.js +407 -91
  173. package/dist/tools/fs/fs-pdf.d.ts +2 -12
  174. package/dist/tools/fs/fs-shared.js +2 -2
  175. package/dist/tools/fs/notebook.d.ts +2 -5
  176. package/dist/tools/fs/safety.d.ts +3 -0
  177. package/dist/tools/fs/safety.js +43 -14
  178. package/dist/tools/fs/search.js +5 -5
  179. package/dist/tools/loop-tick.js +1 -1
  180. package/dist/tools/monitor.d.ts +1 -0
  181. package/dist/tools/monitor.js +57 -9
  182. package/dist/tools/worktree.d.ts +4 -7
  183. package/dist/tools/worktree.js +235 -143
  184. package/package.json +9 -2
@@ -1,9 +1,10 @@
1
+ import { mintFallbackToolCallId } from "./tool-call-id.js";
1
2
  const TAG_RE = /<(tool_call|function_call|tool_calls|function_calls)>\s*([\s\S]*?)\s*<\/\1>/gi;
2
3
  const FENCE_RE = /```(?:json|tool_call|tool|function)?\s*([\s\S]*?)```/gi;
3
4
  function asRecord(v) {
4
5
  return v && typeof v === "object" && !Array.isArray(v) ? v : undefined;
5
6
  }
6
- function extractCall(obj, toolNames) {
7
+ function extractCall(obj, toolNames, framed) {
7
8
  const rec = asRecord(obj);
8
9
  if (!rec) {
9
10
  return undefined;
@@ -13,45 +14,63 @@ function extractCall(obj, toolNames) {
13
14
  if (typeof name !== "string" || !toolNames.has(name)) {
14
15
  return undefined;
15
16
  }
16
- let rawArgs = rec.arguments ?? rec.parameters ?? rec.input ?? rec.args ?? fn?.arguments;
17
- if (typeof rawArgs === "string") {
17
+ const rawArgs = rec.arguments ?? rec.parameters ?? rec.input ?? rec.args ?? fn?.arguments;
18
+ let args;
19
+ if (rawArgs === undefined) {
20
+ if (!framed)
21
+ return undefined;
22
+ args = {};
23
+ }
24
+ else if (typeof rawArgs === "string") {
25
+ let parsed;
18
26
  try {
19
- rawArgs = JSON.parse(rawArgs);
27
+ parsed = JSON.parse(rawArgs);
20
28
  }
21
29
  catch {
22
- rawArgs = {};
30
+ return undefined;
23
31
  }
32
+ const rec2 = asRecord(parsed);
33
+ if (!rec2)
34
+ return undefined;
35
+ args = rec2;
36
+ }
37
+ else {
38
+ const rec2 = asRecord(rawArgs);
39
+ if (!rec2)
40
+ return undefined;
41
+ args = rec2;
24
42
  }
25
- const args = asRecord(rawArgs) ?? {};
26
43
  return {
27
44
  type: "toolCall",
28
- id: `repair_${name}_${Math.random().toString(36).slice(2, 8)}`,
45
+ id: mintFallbackToolCallId(`repair_${name}_`),
29
46
  name,
30
47
  arguments: args,
31
48
  };
32
49
  }
33
- function parseBlock(raw, toolNames, out) {
50
+ function parseBlock(raw, toolNames, out, framed) {
34
51
  let parsed;
35
52
  try {
36
53
  parsed = JSON.parse(raw.trim());
37
54
  }
38
55
  catch {
39
- return false;
56
+ return "none";
40
57
  }
41
58
  const items = Array.isArray(parsed) ? parsed : [parsed];
42
- let any = false;
59
+ let promoted = 0;
43
60
  for (const item of items) {
44
- const call = extractCall(item, toolNames);
61
+ const call = extractCall(item, toolNames, framed);
45
62
  if (call) {
46
63
  out.push(call);
47
- any = true;
64
+ promoted++;
48
65
  }
49
66
  }
50
- return any;
67
+ if (promoted === 0)
68
+ return "none";
69
+ return promoted === items.length ? "all" : "some";
51
70
  }
52
71
  export function repairTextToolCalls(text, toolNames) {
53
72
  const toolCalls = [];
54
- let cleaned = text.replace(TAG_RE, (m, _tag, inner) => parseBlock(inner, toolNames, toolCalls) ? "" : m);
55
- cleaned = cleaned.replace(FENCE_RE, (m, inner) => parseBlock(inner, toolNames, toolCalls) ? "" : m);
73
+ let cleaned = text.replace(TAG_RE, (m, _tag, inner) => parseBlock(inner, toolNames, toolCalls, true) === "all" ? "" : m);
74
+ cleaned = cleaned.replace(FENCE_RE, (m, inner) => parseBlock(inner, toolNames, toolCalls, false) === "all" ? "" : m);
56
75
  return { toolCalls, cleanedText: cleaned.trim() };
57
76
  }
@@ -1,10 +1,11 @@
1
- import { DEFAULT_CHARS_PER_TOKEN, DEFAULT_CLAMP_TOLERANCE, DEFAULT_COMPACTION_SETTINGS, CompactionError, compact, computeFileLists, dryRunSummarizationClamp, estimateContextTokens, estimateTokens, formatPersistedOutputRefs, prepareCompaction, shouldCompact, } from "../internal/harness.js";
1
+ import { DEFAULT_CHARS_PER_TOKEN, DEFAULT_CLAMP_TOLERANCE, DEFAULT_COMPACTION_SETTINGS, CompactionError, compact, computeFileLists, dryRunSummarizationClamp, estimateContextTokens, estimateTokens, formatFileOperations, formatPersistedOutputRefs, prepareCompaction, shouldCompact, } from "../internal/harness.js";
2
2
  import { fileArgPath } from "../tools/fs/safety.js";
3
3
  import { contextEditFrontier } from "./context-edit.js";
4
4
  import { selectCompactionEpoch } from "../prompt-assembly/epoch.js";
5
5
  import { SEMA_DEFAULT_PACK } from "../prompt-assembly/packs/sema-default.js";
6
6
  import { brainToRuntime } from "./runtime.js";
7
7
  import { defuseFenceMarkers, sanitizeUntrustedText } from "./untrusted-text.js";
8
+ import { createSafeNotifier } from "./safe-notify.js";
8
9
  export const DEFAULT_COMPACTION_INSTRUCTIONS = "Preserve verbatim every code, identifier, name, number, file path, error message, credential " +
9
10
  "reference, and explicit user instruction or decision mentioned anywhere in the conversation. " +
10
11
  "When in doubt, keep the specific detail rather than generalizing it.";
@@ -17,6 +18,10 @@ export function sanitizeCompactionInstructions(text) {
17
18
  ? cps.slice(0, COMPACTION_INSTRUCTIONS_MAX_CHARS).join("")
18
19
  : defused;
19
20
  }
21
+ const WORKING_FILE_BREAKOUT_TAGS = ["working-file", "summary"];
22
+ function escapeWorkingFilePathAttr(path) {
23
+ return path.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/[\r\n]+/g, " ");
24
+ }
20
25
  export const COMPACTION_THROUGHPUT_PRIOR_TOKENS_PER_MS = 3;
21
26
  const COMPACTION_THROUGHPUT_EWMA_ALPHA = 0.3;
22
27
  const COMPACTION_WALLTIME_ABORT_FLAG = "semaCompactionWalltimeAbort";
@@ -69,8 +74,11 @@ export async function maybeCompact(opts) {
69
74
  const anchorStale = est.usageTokens > 0 && est.tokens > structuralTokens * STALE_ANCHOR_STRUCTURAL_MARGIN + overhead;
70
75
  const tokens = anchorStale || est.usageTokens === 0 ? structuralTokens + overhead : est.tokens;
71
76
  const window = opts.model.autoCompactTokens ?? opts.model.contextTokens ?? opts.model.contextWindow;
77
+ const windowKnown = Number.isFinite(window) && window > 0;
72
78
  const settings = sanitizeCompactionSettings(rawSettings, window);
73
- const contextUsage = { usedTokens: tokens, windowTokens: window, compactAtTokens: window - settings.reserveTokens };
79
+ const contextUsage = windowKnown
80
+ ? { usedTokens: tokens, windowTokens: window, compactAtTokens: window - settings.reserveTokens }
81
+ : undefined;
74
82
  const wantsCompact = shouldCompact(tokens, window, settings);
75
83
  const naturalTrigger = wantsCompact && tokens >= (opts.minTokens ?? 0);
76
84
  const force = opts.force || (opts.forceUnderThreshold === true && !naturalTrigger);
@@ -89,6 +97,7 @@ export async function maybeCompact(opts) {
89
97
  return { contextUsage, compacted: false, noop: true };
90
98
  }
91
99
  const trigger = opts.trigger ?? "auto";
100
+ const notifier = createSafeNotifier();
92
101
  let hookInstructions;
93
102
  if (opts.preCompact) {
94
103
  let pre;
@@ -118,7 +127,7 @@ export async function maybeCompact(opts) {
118
127
  : hookSanitized
119
128
  : specInstructions;
120
129
  const maxReuse = opts.maxConsecutiveProviderReuse ?? 3;
121
- const forceRealSummary = (opts.consecutiveProviderReuse ?? 0) >= maxReuse;
130
+ const forceRealSummary = maxReuse > 0 && (opts.consecutiveProviderReuse ?? 0) >= maxReuse;
122
131
  let providerSummary;
123
132
  if (opts.summaryProvider && !forceRealSummary && !prep.value.isSplitTurn) {
124
133
  const providerModifiedFiles = computeFileLists(prep.value.fileOps).modifiedFilesByRecency;
@@ -151,7 +160,9 @@ export async function maybeCompact(opts) {
151
160
  summary = providerSummary;
152
161
  firstKeptEntryId = prep.value.firstKeptEntryId;
153
162
  summaryTokensBefore = prep.value.tokensBefore;
154
- details = computeFileLists(prep.value.fileOps);
163
+ const providerFileLists = computeFileLists(prep.value.fileOps);
164
+ details = providerFileLists;
165
+ summary += formatFileOperations(providerFileLists.readFiles, providerFileLists.modifiedFiles);
155
166
  if (prep.value.invokedSkills.length > 0) {
156
167
  details.invokedSkills = prep.value.invokedSkills;
157
168
  }
@@ -174,11 +185,7 @@ export async function maybeCompact(opts) {
174
185
  }
175
186
  const contentTokens = dry !== undefined ? Math.ceil((dry.conversationChars + (prep.value.previousSummary?.length ?? 0)) / cpt) : 0;
176
187
  const observeWindowSafety = (info) => {
177
- try {
178
- opts.onWindowSafety?.(info);
179
- }
180
- catch {
181
- }
188
+ notifier.notify(() => opts.onWindowSafety?.(info), "compaction.onWindowSafety");
182
189
  };
183
190
  if (wantsWindowSafety && dry !== undefined) {
184
191
  const tolerance = settings.clampTolerance ?? DEFAULT_CLAMP_TOLERANCE;
@@ -330,7 +337,7 @@ export async function maybeCompact(opts) {
330
337
  const rawCandidateFiles = recentlyRead.length > 0 ? recentlyRead : (details?.modifiedFilesByRecency ?? details?.modifiedFiles);
331
338
  const maxFilesForSelection = Math.max(1, att?.maxFiles ?? 3);
332
339
  const excludedReadStatePreserveKeys = [];
333
- const candidateFiles = Array.isArray(rawCandidateFiles) && rawCandidateFiles.length > 0
340
+ const candidateFiles = att !== undefined && Array.isArray(rawCandidateFiles) && rawCandidateFiles.length > 0
334
341
  ? await (async () => {
335
342
  const keptTailReadPathsRaw = new Set();
336
343
  const firstKeptIdx = branch.findIndex((e) => e.id === firstKeptEntryId);
@@ -417,7 +424,8 @@ export async function maybeCompact(opts) {
417
424
  if (att && Array.isArray(candidateFiles) && candidateFiles.length > 0) {
418
425
  const maxFiles = maxFilesForSelection;
419
426
  const perFileCap = Math.max(200, att.maxCharsPerFile ?? 16_000);
420
- let remaining = Math.min(maxFiles * perFileCap, Math.floor(window * 0.15) * cpt);
427
+ const windowShareChars = windowKnown ? Math.floor(window * 0.15) * cpt : Number.POSITIVE_INFINITY;
428
+ let remaining = Math.min(maxFiles * perFileCap, windowShareChars);
421
429
  const blocks = [];
422
430
  for (const path of candidateFiles.slice(0, maxFiles)) {
423
431
  if (remaining <= 0)
@@ -437,7 +445,7 @@ export async function maybeCompact(opts) {
437
445
  ? `${content.slice(0, cap)}\n[... ${content.length - cap} more characters truncated — read the file for the rest]`
438
446
  : content;
439
447
  remaining -= clipped.length;
440
- blocks.push(`<working-file path="${path}">\n${clipped}\n</working-file>`);
448
+ blocks.push(`<working-file path="${escapeWorkingFilePathAttr(path)}">\n${sanitizeUntrustedText(clipped, WORKING_FILE_BREAKOUT_TAGS)}\n</working-file>`);
441
449
  attachedFiles.push({ path, chars: Math.min(content.length, cap), truncated: wasClipped });
442
450
  if (!wasClipped)
443
451
  attachedComplete.push({ path, content });
@@ -488,27 +496,20 @@ export async function maybeCompact(opts) {
488
496
  }
489
497
  catch {
490
498
  }
491
- try {
492
- opts.onCompaction?.({
493
- messagesCompacted: prep.value.messagesToSummarize.length,
494
- freedTokens,
495
- tokensBefore: summaryTokensBefore,
496
- reused,
497
- });
498
- }
499
- catch {
500
- }
499
+ const messagesCompacted = prep.value.messagesToSummarize.length;
500
+ notifier.notify(() => opts.onCompaction?.({
501
+ messagesCompacted,
502
+ freedTokens,
503
+ tokensBefore: summaryTokensBefore,
504
+ reused,
505
+ }), "compaction.onCompaction");
501
506
  if (opts.postCompact) {
502
- try {
503
- await opts.postCompact({
504
- trigger,
505
- summary,
506
- tokensBefore: summaryTokensBefore,
507
- ...(postTriggerTokens !== undefined ? { tokensAfter: postTriggerTokens } : {}),
508
- });
509
- }
510
- catch {
511
- }
507
+ await notifier.notifyAsync(() => opts.postCompact?.({
508
+ trigger,
509
+ summary,
510
+ tokensBefore: summaryTokensBefore,
511
+ ...(postTriggerTokens !== undefined ? { tokensAfter: postTriggerTokens } : {}),
512
+ }), "compaction.postCompact");
512
513
  }
513
514
  const endAt = Date.now();
514
515
  return {
@@ -73,18 +73,24 @@ export declare function canAccessAgentRecord(record: Pick<BackgroundAgentRecord,
73
73
  sessionId?: string;
74
74
  }): boolean;
75
75
  export declare const STALE_RUNNING_REAP_ATTRIBUTION = "the host process was interrupted while this agent was running \u2014 its in-process state was lost. Check its worktree / output file for partial work before assuming the task landed (stale running row reaped).";
76
+ export interface BackgroundAgentUpdateIfExpect {
77
+ rev: number;
78
+ status?: BackgroundAgentRecord["status"];
79
+ parkedCheckpointToken?: string | null;
80
+ parkClaimId?: string | null;
81
+ }
82
+ export interface BackgroundAgentReapOptions {
83
+ maxAgeMs?: number;
84
+ keep?: number;
85
+ staleRunningMaxAgeMs?: number;
86
+ }
76
87
  export interface BackgroundAgentStore {
77
88
  put(record: BackgroundAgentRecord): Promise<void>;
78
89
  get(handle: string, scope: string): Promise<BackgroundAgentRecord | null>;
79
90
  update(handle: string, scope: string, record: BackgroundAgentRecord, expect: {
80
91
  rev: number;
81
92
  }): Promise<boolean>;
82
- updateIf(handle: string, scope: string, record: BackgroundAgentRecord, expect: {
83
- rev: number;
84
- status?: BackgroundAgentRecord["status"];
85
- parkedCheckpointToken?: string | null;
86
- parkClaimId?: string | null;
87
- }): Promise<boolean>;
93
+ updateIf(handle: string, scope: string, record: BackgroundAgentRecord, expect: BackgroundAgentUpdateIfExpect): Promise<boolean>;
88
94
  listBySession(scope: string, sessionId: string, opts?: {
89
95
  status?: BackgroundAgentRecord["status"];
90
96
  limit?: number;
@@ -97,11 +103,7 @@ export interface BackgroundAgentStore {
97
103
  rev: number;
98
104
  }): Promise<boolean>;
99
105
  listScopes?(): Promise<string[]>;
100
- reap(scope: string, now: number, opts?: {
101
- maxAgeMs?: number;
102
- keep?: number;
103
- staleRunningMaxAgeMs?: number;
104
- }): Promise<number>;
106
+ reap(scope: string, now: number, opts?: BackgroundAgentReapOptions): Promise<number>;
105
107
  }
106
108
  export declare function reconcileParkedAgents(stores: {
107
109
  agentStore: BackgroundAgentStore;
@@ -114,6 +116,13 @@ export declare function reconcileParkedAgents(stores: {
114
116
  failed: number;
115
117
  rolledBack: number;
116
118
  }>;
119
+ export interface BackgroundAgentQuery {
120
+ scope: string;
121
+ sessionId?: string;
122
+ status?: BackgroundAgentRecord["status"];
123
+ limit?: number;
124
+ }
125
+ export declare function queryBackgroundAgents(rows: Iterable<BackgroundAgentRecord>, query: BackgroundAgentQuery): BackgroundAgentRowSummary[];
117
126
  export declare class InMemoryBackgroundAgentStore implements BackgroundAgentStore {
118
127
  private rows;
119
128
  private key;
@@ -122,12 +131,7 @@ export declare class InMemoryBackgroundAgentStore implements BackgroundAgentStor
122
131
  update(handle: string, scope: string, record: BackgroundAgentRecord, expect: {
123
132
  rev: number;
124
133
  }): Promise<boolean>;
125
- updateIf(handle: string, scope: string, record: BackgroundAgentRecord, expect: {
126
- rev: number;
127
- status?: BackgroundAgentRecord["status"];
128
- parkedCheckpointToken?: string | null;
129
- parkClaimId?: string | null;
130
- }): Promise<boolean>;
134
+ updateIf(handle: string, scope: string, record: BackgroundAgentRecord, expect: BackgroundAgentUpdateIfExpect): Promise<boolean>;
131
135
  listBySession(scope: string, sessionId: string, opts?: {
132
136
  status?: BackgroundAgentRecord["status"];
133
137
  limit?: number;
@@ -140,9 +144,5 @@ export declare class InMemoryBackgroundAgentStore implements BackgroundAgentStor
140
144
  rev: number;
141
145
  }): Promise<boolean>;
142
146
  listScopes(): Promise<string[]>;
143
- reap(scope: string, now: number, opts?: {
144
- maxAgeMs?: number;
145
- keep?: number;
146
- staleRunningMaxAgeMs?: number;
147
- }): Promise<number>;
147
+ reap(scope: string, now: number, opts?: BackgroundAgentReapOptions): Promise<number>;
148
148
  }
@@ -173,6 +173,23 @@ function toSummary(r) {
173
173
  ...(r.usage !== undefined ? { usage: structuredClone(r.usage) } : {}),
174
174
  };
175
175
  }
176
+ export function queryBackgroundAgents(rows, query) {
177
+ const out = [];
178
+ for (const r of rows) {
179
+ if (r.scope !== query.scope)
180
+ continue;
181
+ if (query.sessionId !== undefined) {
182
+ const anchor = r.sessionScoped ? r.owner : r.parentSessionId;
183
+ if (anchor !== query.sessionId && r.rootSessionId !== query.sessionId)
184
+ continue;
185
+ }
186
+ if (query.status !== undefined && r.status !== query.status)
187
+ continue;
188
+ out.push(toSummary(r));
189
+ }
190
+ out.sort((a, b) => b.spawnedAt - a.spawnedAt);
191
+ return query.limit !== undefined ? out.slice(0, Math.max(0, query.limit)) : out;
192
+ }
176
193
  export class InMemoryBackgroundAgentStore {
177
194
  rows = new Map();
178
195
  key(handle, scope) {
@@ -220,31 +237,10 @@ export class InMemoryBackgroundAgentStore {
220
237
  return this.update(handle, scope, record, { rev: expect.rev });
221
238
  }
222
239
  async listBySession(scope, sessionId, opts) {
223
- const out = [];
224
- for (const r of this.rows.values()) {
225
- if (r.scope !== scope)
226
- continue;
227
- const anchor = r.sessionScoped ? r.owner : r.parentSessionId;
228
- if (anchor !== sessionId && r.rootSessionId !== sessionId)
229
- continue;
230
- if (opts?.status !== undefined && r.status !== opts.status)
231
- continue;
232
- out.push(toSummary(r));
233
- }
234
- out.sort((a, b) => b.spawnedAt - a.spawnedAt);
235
- return opts?.limit !== undefined ? out.slice(0, Math.max(0, opts.limit)) : out;
240
+ return queryBackgroundAgents(this.rows.values(), { scope, sessionId, status: opts?.status, limit: opts?.limit });
236
241
  }
237
242
  async listByScope(scope, opts) {
238
- const out = [];
239
- for (const r of this.rows.values()) {
240
- if (r.scope !== scope)
241
- continue;
242
- if (opts?.status !== undefined && r.status !== opts.status)
243
- continue;
244
- out.push(toSummary(r));
245
- }
246
- out.sort((a, b) => b.spawnedAt - a.spawnedAt);
247
- return opts?.limit !== undefined ? out.slice(0, Math.max(0, opts.limit)) : out;
243
+ return queryBackgroundAgents(this.rows.values(), { scope, status: opts?.status, limit: opts?.limit });
248
244
  }
249
245
  async delete(handle, scope, expect) {
250
246
  const k = this.key(handle, scope);
@@ -3,7 +3,8 @@ import type { RepairBundle } from "../agents/repair-loop.js";
3
3
  import type { WorkspaceHandle } from "./remote-env.js";
4
4
  import type { ConsolidationNote } from "./runner/memory-consolidation.js";
5
5
  import type { SessionPermissionRules } from "./session-policy-store.js";
6
- import type { TaskResult } from "./types.js";
6
+ import type { NestedUsageAccum, TaskResult } from "./types.js";
7
+ import type { ActiveWorktreeSession } from "../internal/harness-types.js";
7
8
  export type CheckpointToken = string & {
8
9
  readonly __brand: "CheckpointToken";
9
10
  };
@@ -30,6 +31,7 @@ export declare function riskSeverity(axes: {
30
31
  shell?: boolean;
31
32
  }): 1 | 2 | 3 | 4 | 5;
32
33
  export declare const MAX_TOOL_INPUT_PREVIEW_CHARS = 512;
34
+ export declare const MAX_PENDING_STEER_CHARS = 16000;
33
35
  export declare function buildRiskDescriptor(input: {
34
36
  toolName: string;
35
37
  args: unknown;
@@ -111,12 +113,7 @@ export interface CheckpointState {
111
113
  value?: unknown;
112
114
  set?: boolean;
113
115
  };
114
- nestedStats: {
115
- tokens: number;
116
- turns: number;
117
- tasks: number;
118
- costMicroUsd: number;
119
- };
116
+ nestedStats: NestedUsageAccum;
120
117
  consolidationNotes?: ConsolidationNote[];
121
118
  readFileState?: Array<[string, ReadEntry]>;
122
119
  repairBundle?: RepairBundle;
@@ -130,12 +127,7 @@ export interface CheckpointState {
130
127
  description?: string;
131
128
  }>;
132
129
  handsCwd?: string;
133
- activeWorktree?: {
134
- worktreeDir: string;
135
- originalCwd: string;
136
- baseSha: string;
137
- entered?: boolean;
138
- };
130
+ activeWorktree?: ActiveWorktreeSession;
139
131
  inheritedGate?: {
140
132
  ancestorRules?: ReadonlyArray<{
141
133
  sessionId: string;
@@ -272,6 +264,8 @@ export declare function validatePendingSteer(steer: {
272
264
  text: string;
273
265
  trusted: boolean;
274
266
  };
267
+ export declare function checkpointRowMatches(cp: Checkpoint | undefined, scope: string, status: Checkpoint["status"]): cp is Checkpoint;
268
+ export declare function checkpointOccMatches(cp: Checkpoint, expect?: ResolveExpectation): boolean;
275
269
  export type CheckpointFaultMode = "resolve-after-commit" | "resolve-before-commit";
276
270
  export declare class InMemoryCheckpointStore implements CheckpointStore {
277
271
  private cps;
@@ -25,6 +25,7 @@ const MAX_TOUCHED_PATHS = 8;
25
25
  const MAX_DIGEST_KEYS = 16;
26
26
  const MAX_DIGEST_SCAN_KEYS = 256;
27
27
  export const MAX_TOOL_INPUT_PREVIEW_CHARS = 512;
28
+ export const MAX_PENDING_STEER_CHARS = 16_000;
28
29
  function isPlainRecord(x) {
29
30
  try {
30
31
  if (x === null || typeof x !== "object" || Array.isArray(x))
@@ -130,26 +131,40 @@ export const MAX_SUPPORTED_CHECKPOINT_VERSION = 4;
130
131
  export function checkpointVersionOf(cp) {
131
132
  return cp.version ?? 0;
132
133
  }
134
+ function finiteBound(v) {
135
+ return v !== undefined && Number.isFinite(v) ? v : undefined;
136
+ }
137
+ function finiteSpend(v) {
138
+ return v !== undefined && Number.isFinite(v) ? v : undefined;
139
+ }
133
140
  export function debitLedger(prior, slice, total, opts) {
134
141
  return {
135
- totalBudgetMicroUsd: prior?.totalBudgetMicroUsd ?? total?.totalBudgetMicroUsd,
136
- totalWalltimeSec: prior?.totalWalltimeSec ?? total?.totalWalltimeSec,
137
- spentMicroUsd: (prior?.spentMicroUsd ?? 0) + Math.max(0, slice.costMicroUsd),
138
- spentTokens: (prior?.spentTokens ?? 0) + Math.max(0, slice.tokens),
139
- spentTurns: (prior?.spentTurns ?? 0) + Math.max(0, slice.turns),
140
- spentWalltimeMs: (prior?.spentWalltimeMs ?? 0) + Math.max(0, slice.walltimeMs ?? 0),
142
+ totalBudgetMicroUsd: finiteBound(prior?.totalBudgetMicroUsd) ?? finiteBound(total?.totalBudgetMicroUsd),
143
+ totalWalltimeSec: finiteBound(prior?.totalWalltimeSec) ?? finiteBound(total?.totalWalltimeSec),
144
+ spentMicroUsd: (finiteSpend(prior?.spentMicroUsd) ?? 0) + Math.max(0, finiteSpend(slice.costMicroUsd) ?? 0),
145
+ spentTokens: (finiteSpend(prior?.spentTokens) ?? 0) + Math.max(0, finiteSpend(slice.tokens) ?? 0),
146
+ spentTurns: (finiteSpend(prior?.spentTurns) ?? 0) + Math.max(0, finiteSpend(slice.turns) ?? 0),
147
+ spentWalltimeMs: (finiteSpend(prior?.spentWalltimeMs) ?? 0) + Math.max(0, finiteSpend(slice.walltimeMs) ?? 0),
141
148
  sliceCount: (prior?.sliceCount ?? 0) + (opts?.countSlice === false ? 0 : 1),
142
149
  };
143
150
  }
144
151
  export function remainingBudgetMicroUsd(ledger) {
145
- if (ledger?.totalBudgetMicroUsd === undefined)
152
+ const total = ledger?.totalBudgetMicroUsd;
153
+ if (total === undefined || !Number.isFinite(total))
146
154
  return undefined;
147
- return Math.max(0, ledger.totalBudgetMicroUsd - ledger.spentMicroUsd);
155
+ const spent = ledger?.spentMicroUsd;
156
+ if (spent !== undefined && !Number.isFinite(spent))
157
+ return 0;
158
+ return Math.max(0, total - (spent ?? 0));
148
159
  }
149
160
  export function remainingWalltimeMs(ledger) {
150
- if (ledger?.totalWalltimeSec === undefined)
161
+ const total = ledger?.totalWalltimeSec;
162
+ if (total === undefined || !Number.isFinite(total))
151
163
  return undefined;
152
- return Math.max(0, ledger.totalWalltimeSec * 1000 - (ledger.spentWalltimeMs ?? 0));
164
+ const spent = ledger?.spentWalltimeMs;
165
+ if (spent !== undefined && !Number.isFinite(spent))
166
+ return 0;
167
+ return Math.max(0, total * 1000 - (spent ?? 0));
153
168
  }
154
169
  export function summarizeCheckpoint(cp) {
155
170
  const riskDescriptor = cp.gate.kind === "human" || cp.gate.kind === "irreversible_ask" ? cp.gate.riskDescriptor : undefined;
@@ -217,8 +232,17 @@ export function validatePendingSteer(steer) {
217
232
  if (/[\u0000-\u0008\u000b\u000c\u000e-\u001f\u007f]/.test(steer.text)) {
218
233
  throw new CheckpointError("steering.invalid_content", "steering text must not contain control characters");
219
234
  }
235
+ if (steer.text.length > MAX_PENDING_STEER_CHARS) {
236
+ throw new CheckpointError("steering.invalid_content", `steering text must be at most ${MAX_PENDING_STEER_CHARS} characters (got ${steer.text.length})`);
237
+ }
220
238
  return { text: steer.text, trusted: steer.trusted };
221
239
  }
240
+ export function checkpointRowMatches(cp, scope, status) {
241
+ return cp !== undefined && cp.scope === scope && cp.status === status;
242
+ }
243
+ export function checkpointOccMatches(cp, expect) {
244
+ return expect === undefined || (cp.rev ?? 0) === expect.rev;
245
+ }
222
246
  export class InMemoryCheckpointStore {
223
247
  cps = new Map();
224
248
  fault = null;
@@ -238,10 +262,10 @@ export class InMemoryCheckpointStore {
238
262
  throw new Error("injected fault: resolve before commit");
239
263
  }
240
264
  const cp = this.cps.get(token);
241
- if (!cp || cp.scope !== scope || cp.status !== "pending") {
265
+ if (!checkpointRowMatches(cp, scope, "pending")) {
242
266
  return false;
243
267
  }
244
- if (expect !== undefined && (cp.rev ?? 0) !== expect.rev) {
268
+ if (!checkpointOccMatches(cp, expect)) {
245
269
  return false;
246
270
  }
247
271
  cp.status = "resolved";
@@ -258,7 +282,7 @@ export class InMemoryCheckpointStore {
258
282
  }
259
283
  async reopen(token, scope, reason) {
260
284
  const cp = this.cps.get(token);
261
- if (!cp || cp.scope !== scope || cp.status !== "resolved") {
285
+ if (!checkpointRowMatches(cp, scope, "resolved")) {
262
286
  return false;
263
287
  }
264
288
  cp.status = "pending";
@@ -269,7 +293,7 @@ export class InMemoryCheckpointStore {
269
293
  async setPendingSteer(token, scope, steer) {
270
294
  const clean = validatePendingSteer(steer);
271
295
  const cp = this.cps.get(token);
272
- if (!cp || cp.scope !== scope || cp.status !== "pending") {
296
+ if (!checkpointRowMatches(cp, scope, "pending")) {
273
297
  return false;
274
298
  }
275
299
  cp.state.pendingSteer = clean;
@@ -277,7 +301,7 @@ export class InMemoryCheckpointStore {
277
301
  }
278
302
  async expire(token, scope) {
279
303
  const cp = this.cps.get(token);
280
- if (!cp || cp.scope !== scope || cp.status !== "pending") {
304
+ if (!checkpointRowMatches(cp, scope, "pending")) {
281
305
  return false;
282
306
  }
283
307
  cp.status = "expired";
@@ -3,9 +3,9 @@ import { runMemoryConsolidation, DEFAULT_CONSOLIDATION_BAND, DEFAULT_CONSOLIDATI
3
3
  function maxMarker(a, b) {
4
4
  return a === undefined || b > a ? b : a;
5
5
  }
6
- function cappedHighWater(seen, failed) {
6
+ function cappedHighWater(seen, blockers) {
7
7
  let floor;
8
- for (const id of failed)
8
+ for (const id of blockers)
9
9
  if (floor === undefined || id < floor)
10
10
  floor = id;
11
11
  let hw;
@@ -66,7 +66,8 @@ export async function consolidateScope(scope, deps, opts = {}) {
66
66
  });
67
67
  }
68
68
  const seen = [...batchIds, ...(stats?.addedIds ?? [])];
69
- const highWater = cappedHighWater(seen, stats?.failedIds ?? []);
69
+ const unfedTail = pending.slice(maxNotes).map((h) => h.id);
70
+ const highWater = cappedHighWater(seen, [...(stats?.failedIds ?? []), ...unfedTail]);
70
71
  if (highWater !== undefined && highWater !== cursor && (cursor === undefined || highWater > cursor)) {
71
72
  await store.setConsolidationCursor(scope, highWater);
72
73
  }
@@ -114,5 +114,8 @@ export function clearStaleToolResults(messages, opts) {
114
114
  }
115
115
  export function editBudget(model) {
116
116
  const window = model.autoCompactTokens ?? model.contextTokens ?? model.contextWindow;
117
+ if (!Number.isFinite(window) || window <= 0) {
118
+ return Number.POSITIVE_INFINITY;
119
+ }
117
120
  return contextEditFrontier(window);
118
121
  }
@@ -116,5 +116,8 @@ export function dropOrphanToolResults(messages) {
116
116
  }
117
117
  export function guardBudget(model) {
118
118
  const window = model.contextTokens ?? model.contextWindow;
119
+ if (!Number.isFinite(window) || window <= 0) {
120
+ return Number.POSITIVE_INFINITY;
121
+ }
119
122
  return Math.max(window - GUARD_HEADROOM_TOKENS, Math.floor(window * GUARD_FRACTION));
120
123
  }
@@ -1,3 +1,5 @@
1
+ declare function relUnder(root: string, child: string): string | null;
2
+ declare function joinUnder(root: string, rel: string): string;
1
3
  import type { ExecutionEnv } from "../internal/harness-types.js";
2
4
  export interface FileSnapshotError {
3
5
  code: "too_large" | "enumerate_failed" | "read_failed" | "restore_failed" | "not_found" | "aborted";
@@ -51,3 +53,9 @@ export declare class InMemoryFileSnapshotStore implements FileSnapshotStore {
51
53
  importManifest(scope: string, key: string, manifest: Map<string, string>, srcGetBlob: (hash: string) => Promise<Uint8Array | undefined>): Promise<FileSnapshotResult>;
52
54
  reap(scope: string, keepKeys: string[]): Promise<number>;
53
55
  }
56
+ export declare const __testables: {
57
+ joinUnder: typeof joinUnder;
58
+ relUnder: typeof relUnder;
59
+ stripTrailSep: (p: string) => string;
60
+ };
61
+ export {};
@@ -1,8 +1,8 @@
1
1
  import { createHash } from "node:crypto";
2
+ import { isAbsolutePathForm, isWinFormPath } from "../tools/fs/safety.js";
2
3
  const normSlash = (p) => p.replace(/\\/g, "/");
3
- const stripTrailSep = (p) => p.replace(/[\\/]+$/, "");
4
+ const stripTrailSep = (p) => (isWinFormPath(p) ? p.replace(/[\\/]+$/, "") : p.replace(/\/+$/, ""));
4
5
  const foldDrive = (p) => (/^[A-Za-z]:(\/|$)/.test(p) ? p[0].toLowerCase() + p.slice(1) : p);
5
- const isWinFormPath = (p) => /^[A-Za-z]:[\\/]/.test(p) || p.startsWith("\\\\") || (!p.startsWith("/") && p.includes("\\"));
6
6
  function relUnder(root, child) {
7
7
  const fold = isWinFormPath(root) ? normSlash : (p) => p;
8
8
  const r = foldDrive(fold(stripTrailSep(root)));
@@ -18,7 +18,7 @@ function joinUnder(root, rel) {
18
18
  return stripTrailSep(root) + (win ? "\\" : "/") + (win ? rel.replace(/\//g, "\\") : rel);
19
19
  }
20
20
  function isUnsafeRelPath(relN) {
21
- return relN.startsWith("/") || /^[A-Za-z]:/.test(relN) || relN.split("/").includes("..");
21
+ return isAbsolutePathForm(relN) || /^[A-Za-z]:/.test(relN) || relN.split("/").includes("..");
22
22
  }
23
23
  export const DEFAULT_SNAPSHOT_BOUNDS = {
24
24
  maxFiles: 10000,
@@ -256,3 +256,4 @@ export class InMemoryFileSnapshotStore {
256
256
  return removed;
257
257
  }
258
258
  }
259
+ export const __testables = { joinUnder, relUnder, stripTrailSep };
@@ -12,4 +12,9 @@ export declare function addWorktree(baseEnv: ExecutionEnv, opts: AddWorktreeOpti
12
12
  };
13
13
  worktreeDir: string;
14
14
  }>;
15
- export declare function pruneWorktrees(baseEnv: ExecutionEnv, repoRoot: string): Promise<void>;
15
+ export declare function pruneWorktrees(baseEnv: ExecutionEnv, repoRoot: string): Promise<{
16
+ ok: true;
17
+ } | {
18
+ ok: false;
19
+ detail: string;
20
+ }>;
@@ -59,5 +59,16 @@ export async function addWorktree(baseEnv, opts) {
59
59
  return { env, worktreeDir };
60
60
  }
61
61
  export async function pruneWorktrees(baseEnv, repoRoot) {
62
- await baseEnv.exec("git worktree prune", { cwd: repoRoot });
62
+ try {
63
+ const r = await baseEnv.exec("git worktree prune", { cwd: repoRoot });
64
+ if (r.ok && r.value.exitCode === 0)
65
+ return { ok: true };
66
+ return {
67
+ ok: false,
68
+ detail: r.ok ? (r.value.stderr || r.value.stdout || `exit ${r.value.exitCode}`).trim() : String(r.error),
69
+ };
70
+ }
71
+ catch (err) {
72
+ return { ok: false, detail: String(err) };
73
+ }
63
74
  }