session-orchestrator 5.10.0 → 5.11.0

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 (55) hide show
  1. package/.claude-plugin/marketplace.json +1 -1
  2. package/.claude-plugin/plugin.json +1 -1
  3. package/.codex-plugin/plugin.json +1 -1
  4. package/.cursor-plugin/plugin.json +1 -1
  5. package/AGENTS.md +2 -0
  6. package/CHANGELOG.md +32 -0
  7. package/README.md +9 -8
  8. package/SECURITY.md +1 -1
  9. package/agents/code-implementer.md +7 -1
  10. package/docs/agent-authoring.md +2 -0
  11. package/docs/events-schema.md +2 -1
  12. package/docs/fleet-checkin-audit.md +172 -0
  13. package/docs/role-profile-contract.md +139 -0
  14. package/docs/scope-collision-guard.md +3 -0
  15. package/docs/session-config-reference.md +1 -1
  16. package/evals/instruction-ablation/run.mjs +11 -6
  17. package/hooks/_lib/hook-import-set.json +27 -2
  18. package/hooks/_lib/orchestrator-ignore.mjs +112 -0
  19. package/hooks/hooks-codex.json +1 -1
  20. package/hooks/hooks.json +1 -1
  21. package/hooks/on-session-end.mjs +8 -1
  22. package/hooks/on-session-start.mjs +7 -0
  23. package/hooks/post-bash-write-verify.mjs +16 -3
  24. package/hooks/pre-task-scope-disjoint.mjs +12 -0
  25. package/hooks/subagent-telemetry.mjs +8 -3
  26. package/hooks/wave-scope-commit-guard.mjs +4 -1
  27. package/package.json +1 -1
  28. package/rules/opt-in-stack/security-web.md +1 -1
  29. package/scripts/emit-session.mjs +13 -0
  30. package/scripts/fleet-checkin-audit.mjs +138 -0
  31. package/scripts/lib/common.mjs +2 -4
  32. package/scripts/lib/context-overhead-format.mjs +57 -0
  33. package/scripts/lib/fleet-checkin-audit.mjs +244 -0
  34. package/scripts/lib/memory-proposals/sink.mjs +79 -17
  35. package/scripts/lib/memory-proposals/store.mjs +245 -9
  36. package/scripts/lib/project-hygiene.mjs +37 -0
  37. package/scripts/lib/role-profile/schema.mjs +219 -0
  38. package/scripts/lib/session-lock.mjs +57 -10
  39. package/scripts/lib/session-record-repair.mjs +63 -30
  40. package/scripts/lib/session-schema/constants.mjs +8 -0
  41. package/scripts/lib/session-schema/validator.mjs +33 -0
  42. package/scripts/lib/subagents-schema.mjs +9 -0
  43. package/scripts/lib/telemetry/pricing.mjs +11 -0
  44. package/scripts/measure-context-overhead.sh +4 -21
  45. package/scripts/release.mjs +359 -11
  46. package/scripts/run-quality-gate.mjs +47 -0
  47. package/scripts/validate-wave-scope.mjs +59 -6
  48. package/skills/claude-md-drift-check/checker.mjs +8 -6
  49. package/skills/npm-publish/SKILL.md +3 -3
  50. package/skills/release/SKILL.md +2 -2
  51. package/skills/repo-audit/SKILL.md +1 -1
  52. package/skills/session-end/metrics-collection.md +2 -2
  53. package/skills/session-end/session-metrics-write.md +1 -1
  54. package/skills/session-start/references/phase-1-2-session-lock.md +1 -1
  55. package/skills/wave-executor/circuit-breaker.md +1 -0
@@ -243,7 +243,7 @@ function runCell(variant, kase, run) {
243
243
  if (DRY) return { variant: variant.id, case: kase.id, run, dry: true, workdir: wd };
244
244
 
245
245
  let usage = {};
246
- let cost = 0;
246
+ let cost = null;
247
247
  let text = '';
248
248
  let error = null;
249
249
  try {
@@ -255,7 +255,9 @@ function runCell(variant, kase, run) {
255
255
  });
256
256
  const j = JSON.parse(out);
257
257
  usage = j.usage || {};
258
- cost = j.total_cost_usd || 0;
258
+ // Missing or non-numeric cost stays unknown; a reported 0 stays 0 (#1547).
259
+ const c = j.total_cost_usd;
260
+ cost = typeof c === 'number' && Number.isFinite(c) && c >= 0 ? c : null;
259
261
  text = typeof j.result === 'string' ? j.result : '';
260
262
  } catch (e) {
261
263
  error = String(e.message || e).slice(0, 300);
@@ -275,7 +277,7 @@ function runCell(variant, kase, run) {
275
277
  error,
276
278
  context_tokens: ctx,
277
279
  output_tokens: usage.output_tokens || 0,
278
- cost_usd: Number(cost.toFixed(4)),
280
+ cost_usd: cost === null ? null : Number(cost.toFixed(4)),
279
281
  oracles,
280
282
  reply_excerpt: text.slice(0, 300),
281
283
  workdir: wd,
@@ -300,7 +302,7 @@ for (const kase of cases) {
300
302
  const rec = runCell(variant, kase, run);
301
303
  records.push(rec);
302
304
  if (!DRY) appendFileSync(outFile, JSON.stringify(rec) + '\n');
303
- process.stderr.write(rec.dry ? 'built\n' : `${rec.pass ? 'PASS' : 'FAIL'} (${rec.cost_usd ?? 0} USD)\n`);
305
+ process.stderr.write(rec.dry ? 'built\n' : `${rec.pass ? 'PASS' : 'FAIL'} (${rec.cost_usd ?? 'unknown'} USD)\n`);
304
306
  }
305
307
  }
306
308
  }
@@ -324,13 +326,16 @@ for (const kase of cases) {
324
326
  console.log(row.join('\t'));
325
327
  }
326
328
 
327
- const spend = records.reduce((a, r) => a + (r.cost_usd || 0), 0);
329
+ // One unknown cell makes the total unknown; a partial sum would read as complete.
330
+ const spend = records.some((r) => r.cost_usd === null)
331
+ ? 'unknown'
332
+ : records.reduce((a, r) => a + r.cost_usd, 0).toFixed(2);
328
333
  const ctxAvg = (id) => {
329
334
  const c = records.filter((r) => r.variant === id && r.context_tokens);
330
335
  return c.length ? Math.round(c.reduce((a, r) => a + r.context_tokens, 0) / c.length) : 0;
331
336
  };
332
337
  console.log('\n=== mean context tokens by variant ===');
333
338
  for (const v of variants) console.log(`${v.id}\t${ctxAvg(v.id)}`);
334
- console.log(`\ntotal spend: USD ${spend.toFixed(2)}`);
339
+ console.log(`\ntotal spend: USD ${spend}`);
335
340
  console.log(`results: ${outFile}`);
336
341
  console.log(`workdirs: ${ROOT} (throwaway; inspect failures there, remove when done)`);
@@ -1,6 +1,6 @@
1
1
  {
2
- "generated_at": "2026-10-05T10:45:41.038Z",
3
- "head": "94dc4a0cf182a06492feb217162d290e9fd0e921",
2
+ "generated_at": "2026-10-10T12:04:56.821Z",
3
+ "head": "f4f6200c067d28025bf70e34296cd5ebabcfc7f8",
4
4
  "entries": [
5
5
  {
6
6
  "file": "hooks/_lib/atomic-json.mjs",
@@ -34,6 +34,12 @@
34
34
  "on-session-end.mjs"
35
35
  ]
36
36
  },
37
+ {
38
+ "file": "hooks/_lib/orchestrator-ignore.mjs",
39
+ "reachable_from": [
40
+ "on-session-start.mjs"
41
+ ]
42
+ },
37
43
  {
38
44
  "file": "hooks/_lib/profile-gate.mjs",
39
45
  "reachable_from": [
@@ -1158,6 +1164,19 @@
1158
1164
  "post-edit-validate.mjs"
1159
1165
  ]
1160
1166
  },
1167
+ {
1168
+ "file": "scripts/lib/native-command.mjs",
1169
+ "reachable_from": [
1170
+ "on-session-end.mjs",
1171
+ "on-session-start.mjs",
1172
+ "on-stop.mjs",
1173
+ "post-bash-issue-budget-refund.mjs",
1174
+ "post-edit-validate.mjs",
1175
+ "post-subagent-discovery-validator.mjs",
1176
+ "pre-bash-issue-budget.mjs",
1177
+ "pre-bash-templates-first.mjs"
1178
+ ]
1179
+ },
1161
1180
  {
1162
1181
  "file": "scripts/lib/orphan-reaper/defaults.mjs",
1163
1182
  "reachable_from": [
@@ -1633,6 +1652,12 @@
1633
1652
  "skill-invocation-telemetry.mjs"
1634
1653
  ]
1635
1654
  },
1655
+ {
1656
+ "file": "scripts/lib/telemetry/pricing.mjs",
1657
+ "reachable_from": [
1658
+ "subagent-telemetry.mjs"
1659
+ ]
1660
+ },
1636
1661
  {
1637
1662
  "file": "scripts/lib/telemetry/queue.mjs",
1638
1663
  "reachable_from": [
@@ -0,0 +1,112 @@
1
+ /**
2
+ * orchestrator-ignore.mjs — keep the plugin's runtime state out of `git status`
3
+ * in a repo that has never been bootstrapped (#1515 Pkt 5).
4
+ *
5
+ * Measured 2026-10-08 @ 644743cc: one SessionStart in a fresh `git init` repo
6
+ * left five untracked files under `.orchestrator/` (session.lock,
7
+ * current-session.json, host.json, runtime/lock-owner-proof.json,
8
+ * metrics/events.jsonl) and nothing ignored them. Bootstrapped repos get their
9
+ * runtime ignore lines from `skills/bootstrap/_shared-template.md`
10
+ * § store-lock-ignore; an un-bootstrapped one gets nothing.
11
+ *
12
+ * Root, not per writer (BV-003): 23 hook-reachable modules call mkdir, and
13
+ * `.orchestrator/` is created by whichever runs first. SessionStart is the first
14
+ * plugin hook of every session (hooks/hooks.json), so the one call in
15
+ * on-session-start.mjs covers the session. Because SessionStart runs async, a
16
+ * PreToolUse hook MAY have created the directory a moment earlier — which is
17
+ * why the decision keys on "nothing under .orchestrator/ is tracked" (one
18
+ * `git ls-files`), not on "the directory was just created".
19
+ *
20
+ * Invariants:
21
+ * - A repo WITH `.orchestrator/bootstrap.lock` is never given the file. If it
22
+ * carries a file this helper wrote (byte-identical), the file is removed:
23
+ * after bootstrap, steering/, peers/, policy/ and the ledgers are durable
24
+ * project data and must not stay ignored.
25
+ * - An existing `.orchestrator/.gitignore` is never overwritten.
26
+ * - Any tracked path under `.orchestrator/`, or git failing (no repo, no git
27
+ * binary), means: write nothing.
28
+ * - Never throws; returns a reason string for tests and diagnostics.
29
+ *
30
+ * The negations are the exact `.orchestrator/` paths a bootstrap run stages
31
+ * with `git add -- <file>` (BOOTSTRAP_FILES in skills/bootstrap/*-template.md,
32
+ * measured 2026-10-08). The Bootstrap Gate runs bootstrap in the SAME session
33
+ * this file was written, before the next SessionStart can remove it, and git
34
+ * refuses `git add` on an explicitly named ignored path. NAMED CEILING
35
+ * (BV-004): the list is hand-copied; revisit when a template adds a new
36
+ * `BOOTSTRAP_FILES+=(.orchestrator/...)` entry.
37
+ */
38
+
39
+ import { readFile, writeFile, link, unlink, mkdir } from 'node:fs/promises';
40
+ import { existsSync, lstatSync } from 'node:fs';
41
+ import { execFile } from 'node:child_process';
42
+ import path from 'node:path';
43
+
44
+ /** Directories that hold durable project data, never plugin runtime output. */
45
+ const PROJECT_DATA_DIRS = ['steering', 'peers', 'policy'];
46
+
47
+ export const GENERATED_IGNORE = [
48
+ '# Written by session-orchestrator (#1515): this repo has no .orchestrator/bootstrap.lock,',
49
+ '# so the plugin runtime state below stays out of git status. Removed automatically',
50
+ '# at the first session start after /bootstrap.',
51
+ '*',
52
+ '!*/',
53
+ '!/bootstrap.lock',
54
+ '!/metrics/learnings.jsonl',
55
+ '!/metrics/sessions.jsonl',
56
+ '!/metrics/README.md',
57
+ '!/policy/quality-gates.json',
58
+ '',
59
+ ].join('\n');
60
+
61
+ function trackedUnderOrchestrator(repoRoot) {
62
+ return new Promise((resolve) => {
63
+ execFile('git', ['ls-files', '-z', '--', '.orchestrator'], { cwd: repoRoot, timeout: 2000 }, (err, stdout) => {
64
+ if (err) return resolve(null); // unknown → caller writes nothing
65
+ resolve(stdout.length > 0);
66
+ });
67
+ });
68
+ }
69
+
70
+ /**
71
+ * @param {string} repoRoot
72
+ * @returns {Promise<'written'|'removed'|'bootstrapped'|'present'|'project-data'|'symlink'|'tracked'|'no-git'|'error'>}
73
+ */
74
+ export async function ensureOrchestratorIgnore(repoRoot) {
75
+ try {
76
+ const dir = path.join(repoRoot, '.orchestrator');
77
+ const target = path.join(dir, '.gitignore');
78
+ if (existsSync(path.join(dir, 'bootstrap.lock'))) {
79
+ if (existsSync(target)) {
80
+ const current = await readFile(target, 'utf8').catch(() => null);
81
+ if (current === GENERATED_IGNORE) {
82
+ await unlink(target);
83
+ return 'removed';
84
+ }
85
+ }
86
+ return 'bootstrapped';
87
+ }
88
+ if (existsSync(target)) return 'present';
89
+ // Durable project data (bootstrap/_shared-template.md #store-lock-ignore)
90
+ // the user may hold untracked on purpose: ignoring it would hide it from
91
+ // git status without a word. Only plugin runtime output may be swept.
92
+ if (PROJECT_DATA_DIRS.some((d) => existsSync(path.join(dir, d)))) return 'project-data';
93
+ // A symlinked .orchestrator would put the file outside the repo.
94
+ if (existsSync(dir) && lstatSync(dir).isSymbolicLink()) return 'symlink';
95
+ const tracked = await trackedUnderOrchestrator(repoRoot);
96
+ if (tracked === null) return 'no-git';
97
+ if (tracked) return 'tracked';
98
+ await mkdir(dir, { recursive: true });
99
+ // tmp + link: atomic, and link() fails with EEXIST instead of replacing a
100
+ // .gitignore another process created in between (rename would overwrite).
101
+ const tmp = `${target}.tmp-${process.pid}`;
102
+ await writeFile(tmp, GENERATED_IGNORE, 'utf8');
103
+ try {
104
+ await link(tmp, target);
105
+ } finally {
106
+ await unlink(tmp).catch(() => {});
107
+ }
108
+ return 'written';
109
+ } catch (err) {
110
+ return err && err.code === 'EEXIST' ? 'present' : 'error';
111
+ }
112
+ }
@@ -7,7 +7,7 @@
7
7
  "hooks": [
8
8
  {
9
9
  "type": "command",
10
- "command": "echo '🎯 Session Orchestrator v5.10.0 — /session [housekeeping|feature|deep] | /plan [new|feature|retro] | /discovery [scope] | /evolve [analyze|review|list]'",
10
+ "command": "echo '🎯 Session Orchestrator v5.11.0 — /session [housekeeping|feature|deep] | /plan [new|feature|retro] | /discovery [scope] | /evolve [analyze|review|list]'",
11
11
  "async": false
12
12
  },
13
13
  {
package/hooks/hooks.json CHANGED
@@ -6,7 +6,7 @@
6
6
  "hooks": [
7
7
  {
8
8
  "type": "command",
9
- "command": "echo '🎯 Session Orchestrator v5.10.0 — /session [housekeeping|feature|deep] | /plan [new|feature|retro] | /discovery [scope] | /evolve [analyze|review|list]'",
9
+ "command": "echo '🎯 Session Orchestrator v5.11.0 — /session [housekeeping|feature|deep] | /plan [new|feature|retro] | /discovery [scope] | /evolve [analyze|review|list]'",
10
10
  "async": false
11
11
  },
12
12
  {
@@ -43,6 +43,7 @@ import {
43
43
  release,
44
44
  loadOwnerProof,
45
45
  OWNER_PROOF_RELPATH,
46
+ isLockOwnedByProof,
46
47
  } from '../scripts/lib/session-lock.mjs';
47
48
  import { parseSessionId } from '../scripts/lib/session-id.mjs';
48
49
  import { isMainModule } from '../scripts/lib/is-main-module.mjs';
@@ -789,8 +790,14 @@ async function main() {
789
790
  // millisecond started_at cannot match any FUTURE lock, so it
790
791
  // self-invalidates), this just avoids the stale artifact. ENOENT
791
792
  // (no proof was ever written) lands in the same swallow.
793
+ // Only when the proof still describes the lock we released: since
794
+ // #1541 every acquire() rewrites it, so a successor that acquired
795
+ // during the awaited emit above owns a fresh proof we must not
796
+ // remove. A narrow window between this read and the unlink remains.
792
797
  try {
793
- await fs.unlink(path.join(projectRoot, OWNER_PROOF_RELPATH));
798
+ if (isLockOwnedByProof(lock, loadOwnerProof({ repoRoot: projectRoot }))) {
799
+ await fs.unlink(path.join(projectRoot, OWNER_PROOF_RELPATH));
800
+ }
794
801
  } catch { /* best-effort — a leftover proof is self-invalidating */ }
795
802
  }
796
803
  } else {
@@ -36,6 +36,7 @@ import { randomUUID } from 'node:crypto';
36
36
  import path from 'node:path';
37
37
 
38
38
  import { shouldRunHook } from './_lib/profile-gate.mjs';
39
+ import { ensureOrchestratorIgnore } from './_lib/orchestrator-ignore.mjs';
39
40
 
40
41
  import { emitEvent, eventsFilePath } from '../scripts/lib/events.mjs';
41
42
  import { readNavigatorLease } from '../scripts/lib/fleet-protocol.mjs';
@@ -781,6 +782,12 @@ function maybeQueueConsentNudge() {
781
782
  async function main() {
782
783
  const projectRoot = resolveProjectDir();
783
784
 
785
+ // #1515 Pkt 5 — before this hook's own writers (host.json, session.lock,
786
+ // events.jsonl) run: an un-bootstrapped repo gets `.orchestrator/.gitignore`
787
+ // so the runtime state never shows up in `git status`. Bootstrapped repos
788
+ // pay two existsSync and nothing else. Never throws.
789
+ await ensureOrchestratorIgnore(projectRoot);
790
+
784
791
  // Read optional stdin payload in parallel with git info so we don't stall
785
792
  // the hook. Both are best-effort and the promise races against a 500 ms cap
786
793
  // inside readStdinJson itself.
@@ -1299,10 +1299,23 @@ async function main() {
1299
1299
  ? []
1300
1300
  : readPeerScopeRecords(path.dirname(scopePath), scope.wave, sidecarRaw);
1301
1301
  if (sidecarBinding.notify) {
1302
- messages.push(formatSidecarMismatchNotice(
1303
- path.relative(repoRoot, path.join(path.dirname(scopePath), 'filescopes', `wave-${scope.wave}.scopes.json`)),
1304
- ));
1302
+ const relSidecar = path.relative(
1303
+ repoRoot,
1304
+ path.join(path.dirname(scopePath), 'filescopes', `wave-${scope.wave}.scopes.json`),
1305
+ );
1306
+ messages.push(formatSidecarMismatchNotice(relSidecar));
1305
1307
  warn = true;
1308
+ // #1514 point 6 (HR-105) — the notice alone left its firing rate unmeasurable.
1309
+ // Same once-per-new-hash dedupe as the notice (`bindPeerSidecar().notify`);
1310
+ // awaited so the append cannot be lost to the process exiting.
1311
+ try {
1312
+ const { emitEvent } = await import('../scripts/lib/events.mjs');
1313
+ await emitEvent(
1314
+ 'orchestrator.scope.sidecar_mismatch',
1315
+ { hook: 'post-bash-write-verify', sidecar: relSidecar.split(path.sep).join('/'), wave: scope.wave },
1316
+ { repoRoot: context.manifestRoot },
1317
+ );
1318
+ } catch { /* observability is best-effort — never blocks the notice */ }
1306
1319
  }
1307
1320
  const violations = [];
1308
1321
  for (const relPath of report) {
@@ -301,6 +301,18 @@ const DISPATCH_TOOL = 'Agent';
301
301
  * A flat sibling, not a directory, on purpose: bootstrap's `store-lock-ignore`
302
302
  * step (#1495) ignores `.orchestrator/wave-dispatch-scopes.*` in consumer repos,
303
303
  * which covers this name; a directory would surface untracked in each of them.
304
+ *
305
+ * NAMED CEILING (BV-004, #1504 point 7a): nothing ever deletes a per-session
306
+ * file — the hook only writes its own, and no sweep exists (a sweep would delete
307
+ * files that may be a live peer's state, PSA-003). Each FILE is bounded: it holds
308
+ * one `waveKey`'s agents, reset on the next wave. The DIRECTORY is not: it grows
309
+ * by one file (plus a transient `.lock`) per session that dispatched an agent.
310
+ * Measured 2026-10-08 in this repo's main checkout: 7 files, largest 2918 bytes,
311
+ * oldest 2026-10-03 — negligible. Fine under ~500 files; the hook never lists
312
+ * the directory, so the count costs disk, not latency.
313
+ * Revisit-Trigger: `ls .orchestrator/wave-dispatch-scopes.*.json | wc -l`
314
+ * passes 500 in any repo, OR a reader starts enumerating these files — then
315
+ * build an age-gated sweep that skips any session with a live `session.lock`.
304
316
  */
305
317
  const LEDGER_DIR_REL = '.orchestrator';
306
318
  const LEDGER_STEM = 'wave-dispatch-scopes';
@@ -113,6 +113,7 @@ import { appendSubagent } from '../scripts/lib/subagents-schema.mjs';
113
113
  import { getProjectDir } from '../scripts/lib/platform.mjs';
114
114
  import { resolveSubagentSidecar } from './_lib/subagent-paths.mjs';
115
115
  import { readTailWindow } from '../scripts/lib/tail-window.mjs';
116
+ import { COST_BASIS } from '../scripts/lib/telemetry/pricing.mjs';
116
117
 
117
118
  // ---------------------------------------------------------------------------
118
119
  // Constants
@@ -859,14 +860,18 @@ async function main() {
859
860
  // the rollup can price it part by part; absent on every other record.
860
861
  if (modelsUsage !== null) record.models_usage = modelsUsage;
861
862
 
862
- // Cost is best-effort / forward-compat (#624): the native transcript does
863
- // NOT expose total_cost_usd today, so this is null in practice. No rate
864
- // table is applied HERE — use the native cost only, default null when absent.
863
+ // Cost is best-effort (#624): retain the native payload value when valid,
864
+ // including a genuine zero; use null when absent. No rate table is applied
865
+ // HERE. Provenance (#1543) requires an explicit recognized payload basis;
866
+ // a native number alone says nothing about billing or API list prices.
865
867
  const totalCostUsd =
866
868
  typeof input.total_cost_usd === 'number' && Number.isFinite(input.total_cost_usd) && input.total_cost_usd >= 0
867
869
  ? input.total_cost_usd
868
870
  : null;
869
871
  record.total_cost_usd = totalCostUsd;
872
+ record.cost_basis = totalCostUsd !== null && input.cost_basis === COST_BASIS
873
+ ? input.cost_basis
874
+ : null;
870
875
 
871
876
  // OTel alias — #411 additive. `gen_ai.usage.input_tokens` stays the RAW
872
877
  // UNCACHED value (OTel semantic), which is why it is deliberately NOT equal
@@ -336,7 +336,10 @@ async function main() {
336
336
  // when every candidate is foreign the first comes back and
337
337
  // `scopeCommitVerdict` passes it as before. The git process inherits the
338
338
  // committing session's env; there is no hook payload here, so the env tier is
339
- // the only process-local identity.
339
+ // the only process-local identity. CEILING (#1504 point 7b): a shell that
340
+ // exports a foreign CLAUDE_CODE_SESSION_ID makes this session's manifest read
341
+ // `foreign` and the guard stands down: the same power `--no-verify` grants, so it
342
+ // stays named, not fixed (docs/scope-collision-guard.md § 6 item 12).
340
343
  const ownIds = new Set(readProcessLocalSessionIds());
341
344
 
342
345
  // #1511 point d — ALSO read the manifest at the SESSION root the scope guards
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "session-orchestrator",
3
- "version": "5.10.0",
3
+ "version": "5.11.0",
4
4
  "description": "Your coding agents get an agreed plan, a check after each round of changes and a handover to the next session. Plan, Go, Close: /session reads your repo and agrees the scope, /go runs the work in waves with your tests and typecheck after each implementation wave, /close runs the full gate, verifies, commits and files the rest as issues. Scope and destructive-command guards, crash-safe resume, learnings you approve. For Claude Code, Codex CLI, Cursor and Pi.",
5
5
  "type": "module",
6
6
  "homepage": "https://session-orchestrator.com",
@@ -19,7 +19,7 @@ tier: wave-only
19
19
 
20
20
  CSRF, rate limiting, CSP, and transport security. Core security rules are in `security.md` (always-on).
21
21
 
22
- ## CSRF Protection (SEC-005)
22
+ ## CSRF Protection (SEC-018)
23
23
  - **Next.js 15+ Server Actions**: Built-in CSRF protection via Origin header validation against Host header. No manual CSRF token needed.
24
24
  - **Manual CSRF tokens required for**: custom API routes (`/api/*`), form submissions to external endpoints, any non-Server-Action mutation endpoint.
25
25
  - For manual CSRF token validation, use `crypto.timingSafeEqual()` for timing-safe comparison.
@@ -39,6 +39,7 @@ import {
39
39
  import { parseStateMd, readSessionProfile, resolveStateMdPath } from './lib/state-md.mjs';
40
40
  import { readProcessLocalSessionIds } from './lib/session-identity/own-session.mjs';
41
41
  import { rollupSessionTokens } from './lib/session-token-rollup.mjs';
42
+ import { COST_BASIS, PRICING_TABLE_DATE } from './lib/telemetry/pricing.mjs';
42
43
  import { serializeSessionLineChecked } from './lib/session-schema/serializer.mjs';
43
44
  import {
44
45
  validateSession,
@@ -542,6 +543,18 @@ async function main() {
542
543
  }
543
544
  merged[key] = rollup[key];
544
545
  }
546
+ // #1470 Pkt 4/6 — label the merged cost fields. A dollar figure is the
547
+ // API list-price equivalent, never a payment (`cost_basis`, only beside
548
+ // a written `total_cost_usd`). The coverage counters, too, depend on
549
+ // which models the table knows, so `pricing_table_date` names the table
550
+ // that produced them — read from pricing.mjs, the same module the rollup
551
+ // priced with, never a second copy of the date.
552
+ if ('total_cost_usd' in merged && !hasOwn('cost_basis')) {
553
+ merged.cost_basis = COST_BASIS;
554
+ }
555
+ if ('cost_records_total' in merged && !hasOwn('pricing_table_date')) {
556
+ merged.pricing_table_date = PRICING_TABLE_DATE;
557
+ }
545
558
  // Enrichment must never be the reason the ledger gains no line: validate
546
559
  // the enriched record first and fall back to the un-enriched one. Blamed
547
560
  // on the rollup only when the un-enriched record validates on its own —
@@ -0,0 +1,138 @@
1
+ #!/usr/bin/env node
2
+ /** Read-only CLI for explicitly supplied, stable snapshot copies (#1462). */
3
+ import { constants, promises as fs } from 'node:fs';
4
+ import path from 'node:path';
5
+ import { TextDecoder } from 'node:util';
6
+ import { auditCheckins } from './lib/fleet-checkin-audit.mjs';
7
+ import { isMainModule } from './lib/is-main-module.mjs';
8
+
9
+ const HELP = `Usage: fleet-checkin-audit --input-root DIR --registry FILE --checkins FILE --repo KEY [--repo KEY] --at UTC [--json]
10
+ All five input flags are required. --repo may repeat with distinct explicit keys.
11
+ --at is an exact UTC instant (YYYY-MM-DDTHH:MM:SS[.sss]Z).
12
+ --json emits one JSON report; --help and --version are standalone commands.
13
+ Supply complete registry/check-in snapshot copies with capture times matching --at.
14
+ Only explicit regular files inside --input-root are read; symlinks are refused.
15
+ Use stable trusted copies: simultaneous parent-directory renames are not atomic.
16
+ No discovery, live configuration, referenced-repo reads, events or input writes.
17
+ Exit 0: measured, including findings; 1: invalid arguments, paths or JSON;
18
+ 2: unavailable inputs/system I/O; 3: incomplete or not-measurable snapshot evidence.
19
+ stdout contains the report; stderr contains only diagnostics.
20
+ `;
21
+ const failure = (category, exit, errno) => Object.assign(new Error(category), { category, exit, errno });
22
+
23
+ function argumentsFor(argv) {
24
+ const args = { repos: [], json: false };
25
+ const values = new Map([['--input-root', 'root'], ['--registry', 'registry'], ['--checkins', 'checkins'], ['--at', 'at']]);
26
+ for (let i = 0; i < argv.length; i++) {
27
+ const flag = argv[i];
28
+ if (flag === '--json') { args.json = true; continue; }
29
+ if (flag === '--help' || flag === '--version') {
30
+ if (argv.length !== 1) throw failure('invalid-arguments', 1);
31
+ args.action = flag;
32
+ return args;
33
+ }
34
+ if (flag !== '--repo' && !values.has(flag)) throw failure('invalid-arguments', 1);
35
+ const value = argv[++i];
36
+ if (!value || value.startsWith('--')) throw failure('invalid-arguments', 1);
37
+ if (flag === '--repo') args.repos.push(value);
38
+ else {
39
+ const key = values.get(flag);
40
+ if (args[key] !== undefined) throw failure('duplicate-argument', 1);
41
+ args[key] = value;
42
+ }
43
+ }
44
+ if (!args.root || !args.registry || !args.checkins || !args.at || !args.repos.length) throw failure('missing-arguments', 1);
45
+ if (new Set(args.repos).size !== args.repos.length || args.repos.some((repo) => repo.trim() !== repo)) {
46
+ throw failure('invalid-repo-selection', 1);
47
+ }
48
+ const instant = Date.parse(args.at);
49
+ if (!/^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d{3})?Z$/.test(args.at)
50
+ || !Number.isFinite(instant)
51
+ || new Date(instant).toISOString() !== args.at.replace(/Z$/, args.at.includes('.') ? 'Z' : '.000Z')) {
52
+ throw failure('invalid-measurement-time', 1);
53
+ }
54
+ return args;
55
+ }
56
+
57
+ /** Inspect every component before reading contents; parent rename races remain
58
+ * outside this portable path-based reader's guarantee (use stable copies). */
59
+ async function inspect(root, name) {
60
+ const target = path.resolve(root, name);
61
+ const relative = path.relative(root, target);
62
+ if (!relative || relative === '..' || relative.startsWith(`..${path.sep}`) || path.isAbsolute(relative)) {
63
+ throw failure('path-outside-input-root', 1);
64
+ }
65
+ const parsed = path.parse(target);
66
+ let current = parsed.root;
67
+ let stat;
68
+ const components = target.slice(parsed.root.length).split(path.sep);
69
+ for (let i = 0; i < components.length; i++) {
70
+ current = path.join(current, components[i]);
71
+ stat = await fs.lstat(current);
72
+ if (stat.isSymbolicLink()) throw failure('symlink-input', 1);
73
+ if (i < components.length - 1 && !stat.isDirectory()) throw failure('non-directory-parent', 1);
74
+ }
75
+ if (!stat?.isFile()) throw failure('non-regular-input', 1);
76
+ return { target, stat };
77
+ }
78
+
79
+ async function readSnapshot({ target, stat }) {
80
+ // The parent check above and this open are not atomic: a same-user process can swap a parent
81
+ // directory in between. Fine while the operator names the snapshot files and the result grants
82
+ // nothing; revisit (#1549) once paths come from a foreign owner or the result gates a decision.
83
+ if (typeof constants.O_NOFOLLOW !== 'number' || typeof constants.O_NONBLOCK !== 'number') {
84
+ throw failure('unsupported-safe-open', 2);
85
+ }
86
+ const handle = await fs.open(target, constants.O_RDONLY | constants.O_NOFOLLOW | constants.O_NONBLOCK);
87
+ let bytes;
88
+ try {
89
+ const actual = await handle.stat();
90
+ if (!actual.isFile() || actual.dev !== stat.dev || actual.ino !== stat.ino) throw failure('changed-input-file', 1);
91
+ bytes = await handle.readFile();
92
+ } finally { await handle.close(); }
93
+ try { return JSON.parse(new TextDecoder('utf-8', { fatal: true }).decode(bytes)); }
94
+ catch { throw failure('malformed-json-or-utf8', 1); }
95
+ }
96
+
97
+ function render(report) {
98
+ if (report.error) return `not-measurable: ${report.error.category}\n`;
99
+ return [`status: ${report.status}`, ...Object.entries(report.metrics).map(([key, metric]) =>
100
+ `${key}: ${metric.status === 'measured' ? metric.value : `not-measurable (${metric.reasons.join(', ')})`}`)].join('\n') + '\n';
101
+ }
102
+
103
+ /** The real CLI boundary; callers may route its output to their own streams. */
104
+ export async function main(argv = process.argv.slice(2), io = { stdout: process.stdout, stderr: process.stderr }) {
105
+ let args;
106
+ try {
107
+ args = argumentsFor(argv);
108
+ if (args.action === '--help') { io.stdout.write(HELP); return 0; }
109
+ if (args.action === '--version') {
110
+ const manifest = JSON.parse(await fs.readFile(new URL('../package.json', import.meta.url), 'utf8'));
111
+ io.stdout.write(`${manifest.version}\n`);
112
+ return 0;
113
+ }
114
+ const root = path.resolve(args.root);
115
+ // Inspect BOTH paths before opening either snapshot for content.
116
+ const registryFile = await inspect(root, args.registry);
117
+ const checkinFile = await inspect(root, args.checkins);
118
+ const registry = await readSnapshot(registryFile);
119
+ const checkins = await readSnapshot(checkinFile);
120
+ const report = auditCheckins({ registry, checkins, repos: args.repos, at: args.at });
121
+ io.stdout.write(args.json ? `${JSON.stringify(report)}\n` : render(report));
122
+ return report.status === 'measured' ? 0 : 3;
123
+ } catch (error) {
124
+ const category = error.category || 'input-io-error';
125
+ const errno = typeof error.code === 'string' && /^[A-Z0-9_]+$/.test(error.code) ? error.code : undefined;
126
+ const report = { version: 1, status: 'not-measurable', error: { category, ...(errno ? { errno } : {}) } };
127
+ io.stderr.write(`fleet-checkin-audit: ${category}${errno ? ` (${errno})` : ''}\n`);
128
+ io.stdout.write((args?.json || argv.includes('--json')) ? `${JSON.stringify(report)}\n` : render(report));
129
+ return error.exit || 2;
130
+ }
131
+ }
132
+
133
+ if (isMainModule(import.meta.url)) {
134
+ main().then((code) => { process.exitCode = code; }, () => {
135
+ process.stderr.write('fleet-checkin-audit: output-io-error\n');
136
+ process.exitCode = 2;
137
+ });
138
+ }
@@ -12,7 +12,7 @@ import path from 'node:path';
12
12
  import crypto from 'node:crypto';
13
13
  import { promises as fs, existsSync, statSync } from 'node:fs';
14
14
  import { fileURLToPath } from 'node:url';
15
- import { execSync } from 'node:child_process';
15
+ import { commandOnPath } from './native-command.mjs';
16
16
 
17
17
  // ---------------------------------------------------------------------------
18
18
  // Path helpers
@@ -182,9 +182,7 @@ export function warn(message) {
182
182
  * @throws {Error} when jq is not on PATH
183
183
  */
184
184
  export function requireJq() {
185
- try {
186
- execSync('command -v jq', { stdio: 'ignore', shell: true });
187
- } catch {
185
+ if (!commandOnPath('jq')) {
188
186
  throw new Error('jq is required but not installed. Install via: brew install jq');
189
187
  }
190
188
  }
@@ -0,0 +1,57 @@
1
+ /** Format Claude JSON accounting without inventing a price or its provenance. */
2
+ import { isMainModule } from './is-main-module.mjs';
3
+ import { COST_BASIS } from './telemetry/pricing.mjs';
4
+
5
+ export const CONTEXT_OVERHEAD_HEADER =
6
+ 'LABEL\tCONTEXT_TOK\tcache_create\tcache_read\toutput\tUSD\tCOST_BASIS\n';
7
+
8
+ /**
9
+ * @param {string} text JSON response text
10
+ * @param {string} label measurement label
11
+ * @returns {string} TSV row, or a recognizable PARSE-ERROR / NO-USAGE row
12
+ */
13
+ export function formatContextOverhead(text, label) {
14
+ let response;
15
+ try {
16
+ response = JSON.parse(text);
17
+ } catch {
18
+ return `${label}\tPARSE-ERROR\n`;
19
+ }
20
+ const usage = response?.usage;
21
+ // An error reply (e.g. "Not logged in") carries no usage; 0 tokens would be invented (#1547).
22
+ if (typeof usage !== 'object' || usage === null || Array.isArray(usage)) {
23
+ return `${label}\tNO-USAGE\n`;
24
+ }
25
+ const context =
26
+ (usage.input_tokens || 0) +
27
+ (usage.cache_creation_input_tokens || 0) +
28
+ (usage.cache_read_input_tokens || 0);
29
+ const cost = response?.total_cost_usd;
30
+ const validCost = typeof cost === 'number' && Number.isFinite(cost) && cost >= 0;
31
+ // Native numbers alone do not establish billing or API-list provenance.
32
+ const basis = validCost && response.cost_basis === COST_BASIS
33
+ ? response.cost_basis
34
+ : 'unknown';
35
+ return [
36
+ label, context, usage.cache_creation_input_tokens || 0,
37
+ usage.cache_read_input_tokens || 0, usage.output_tokens || 0,
38
+ validCost ? cost.toFixed(4) : 'unknown', basis,
39
+ ].join('\t') + '\n';
40
+ }
41
+
42
+ if (isMainModule(import.meta.url)) {
43
+ const args = process.argv.slice(2);
44
+ if (args.length === 1 && args[0] === '--header') {
45
+ process.stdout.write(CONTEXT_OVERHEAD_HEADER);
46
+ } else if (args.length === 2 && args[0] === '--') {
47
+ let text = '';
48
+ process.stdin.setEncoding('utf8');
49
+ process.stdin.on('data', (chunk) => { text += chunk; });
50
+ process.stdin.on('end', () => {
51
+ process.stdout.write(formatContextOverhead(text, args[1]));
52
+ });
53
+ } else {
54
+ process.stderr.write('Usage: context-overhead-format.mjs --header | -- <label> (JSON on stdin)\n');
55
+ process.exitCode = 1;
56
+ }
57
+ }