@feigi/fleet-ctl 3.23.36 → 3.23.38

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 (205) hide show
  1. package/package.json +1 -1
  2. package/scripts/cell-readout.mjs +38 -11
  3. package/scripts/main-checkout.mjs +1 -1
  4. package/scripts/member-write-guard.mjs +1 -1
  5. package/scripts/repo-root.mjs +1 -1
  6. package/skills/run-team/references/reaping.md +1 -1
  7. package/scripts/agent-test-dir-prose.test.mjs +0 -59
  8. package/scripts/agent-test.test.mjs +0 -1640
  9. package/scripts/ambient-git-vars-mjs-prose.test.mjs +0 -397
  10. package/scripts/ambient-git-vars-prose.test.mjs +0 -192
  11. package/scripts/ancestry-check-position-prose.test.mjs +0 -175
  12. package/scripts/apply-ruleset.test.mjs +0 -421
  13. package/scripts/arg-header-probes-prose.test.mjs +0 -394
  14. package/scripts/arg.test.mjs +0 -1354
  15. package/scripts/board-cli.test.mjs +0 -710
  16. package/scripts/board-identity.test.mjs +0 -429
  17. package/scripts/board-liveness-banner.test.mjs +0 -119
  18. package/scripts/board-prev-shape.test.mjs +0 -228
  19. package/scripts/board-tryparse-null.test.mjs +0 -119
  20. package/scripts/board.test.mjs +0 -3121
  21. package/scripts/brief-after-cut-prose.test.mjs +0 -53
  22. package/scripts/candidates-exit3-prose.test.mjs +0 -302
  23. package/scripts/candidates.test.mjs +0 -2312
  24. package/scripts/caveat-unification-prose.test.mjs +0 -98
  25. package/scripts/cell-readout.test.mjs +0 -615
  26. package/scripts/ci-comment-rot-prose.test.mjs +0 -2025
  27. package/scripts/ci-completes-premise-prose.test.mjs +0 -257
  28. package/scripts/ci-state-prose.test.mjs +0 -113
  29. package/scripts/ci-state.test.mjs +0 -1933
  30. package/scripts/ci-vacuous-green.test.mjs +0 -368
  31. package/scripts/citation-sweep-prose.test.mjs +0 -1712
  32. package/scripts/claim-lifecycle.test.mjs +0 -187
  33. package/scripts/claim-ticket.test.mjs +0 -1122
  34. package/scripts/closing-keyword-prose.test.mjs +0 -193
  35. package/scripts/cockpit-launch-idempotent-prose.test.mjs +0 -165
  36. package/scripts/cockpit-tick-nonblocking.test.mjs +0 -210
  37. package/scripts/compute-board.test.mjs +0 -1214
  38. package/scripts/compute-spend.test.mjs +0 -517
  39. package/scripts/consumer-technology-prose.test.mjs +0 -126
  40. package/scripts/cross-repo-citation-prose.test.mjs +0 -155
  41. package/scripts/cwd-isolation-pins.mjs +0 -83
  42. package/scripts/dedupe-guard-prose.test.mjs +0 -243
  43. package/scripts/delete-merged-branch.test.mjs +0 -576
  44. package/scripts/derive-testcmd.test.mjs +0 -595
  45. package/scripts/diff-stats.test.mjs +0 -404
  46. package/scripts/dispatch-block-golden-prose.test.mjs +0 -1130
  47. package/scripts/dispatch-block-pins-prose.test.mjs +0 -615
  48. package/scripts/dispositions-check.test.mjs +0 -936
  49. package/scripts/dispositions-gate-prose.test.mjs +0 -129
  50. package/scripts/drop-merged-label.test.mjs +0 -479
  51. package/scripts/exec-stub.mjs +0 -93
  52. package/scripts/exec-stub.test.mjs +0 -69
  53. package/scripts/expected-jobs-prose.test.mjs +0 -80
  54. package/scripts/fan-out-authorization-prose.test.mjs +0 -80
  55. package/scripts/filing-bar-seat-prose.test.mjs +0 -242
  56. package/scripts/filing-claim-bar-prose.test.mjs +0 -166
  57. package/scripts/finisher-died-prose.test.mjs +0 -31
  58. package/scripts/finisher-dispatch-premise-prose.test.mjs +0 -312
  59. package/scripts/finisher-halt-prose.test.mjs +0 -106
  60. package/scripts/finisher-name-prose.test.mjs +0 -134
  61. package/scripts/finisher-own-tree-prose.test.mjs +0 -299
  62. package/scripts/finisher-pin-race-prose.test.mjs +0 -339
  63. package/scripts/finisher-test-verdict-prose.test.mjs +0 -102
  64. package/scripts/fix-applier-correction-rules-prose.test.mjs +0 -202
  65. package/scripts/fix-applier-result-file-prose.test.mjs +0 -149
  66. package/scripts/fleet-bootstrap.test.mjs +0 -75
  67. package/scripts/fleet-heartbeat.test.mjs +0 -450
  68. package/scripts/fleet-provenance.test.mjs +0 -138
  69. package/scripts/fleet-run.test.mjs +0 -90
  70. package/scripts/fleet-state.test.mjs +0 -460
  71. package/scripts/fleet-tick-prose.test.mjs +0 -300
  72. package/scripts/fleet-tick.test.mjs +0 -2444
  73. package/scripts/force-with-lease-reason-prose.test.mjs +0 -184
  74. package/scripts/frontmatter-check.test.mjs +0 -436
  75. package/scripts/git-env.test.mjs +0 -149
  76. package/scripts/git-status-untracked-mode-sweep.test.mjs +0 -216
  77. package/scripts/gone-walk.test.mjs +0 -213
  78. package/scripts/gp-sep-invariant-prose.test.mjs +0 -188
  79. package/scripts/head-moved-after-label-prose.test.mjs +0 -233
  80. package/scripts/immutable-body-claim-prose.test.mjs +0 -204
  81. package/scripts/implementer-model-tier.test.mjs +0 -235
  82. package/scripts/inflight-citation-prose.test.mjs +0 -144
  83. package/scripts/inflight-exit2-prose.test.mjs +0 -81
  84. package/scripts/inflight-probe-guard-prose.test.mjs +0 -109
  85. package/scripts/inflight-scan-scope-prose.test.mjs +0 -57
  86. package/scripts/inflight.test.mjs +0 -2882
  87. package/scripts/injection-control-prose.test.mjs +0 -404
  88. package/scripts/install-root-audit.test.mjs +0 -464
  89. package/scripts/instrument-check-prose.test.mjs +0 -235
  90. package/scripts/instruments.test.mjs +0 -787
  91. package/scripts/is-cli.test.mjs +0 -99
  92. package/scripts/issue-tracker-prose.test.mjs +0 -71
  93. package/scripts/json.test.mjs +0 -364
  94. package/scripts/ledger-dispatch.test.mjs +0 -927
  95. package/scripts/ledger-grammar.test.mjs +0 -227
  96. package/scripts/ledger-read-require-file.test.mjs +0 -197
  97. package/scripts/ledger-rotate.test.mjs +0 -142
  98. package/scripts/ledger.test.mjs +0 -3436
  99. package/scripts/line-distance-prose.test.mjs +0 -245
  100. package/scripts/liveness-rationale-prose.test.mjs +0 -83
  101. package/scripts/locale-pin-prose.test.mjs +0 -189
  102. package/scripts/main-checkout.test.mjs +0 -332
  103. package/scripts/member-outcome-prose.test.mjs +0 -167
  104. package/scripts/member-outcomes-header.test.mjs +0 -129
  105. package/scripts/member-outcomes-prose.test.mjs +0 -63
  106. package/scripts/member-outcomes.test.mjs +0 -638
  107. package/scripts/member-prompt-prose.test.mjs +0 -165
  108. package/scripts/member-record.test.mjs +0 -541
  109. package/scripts/member-write-guard.test.mjs +0 -214
  110. package/scripts/merge-commit-probe-prose.test.mjs +0 -122
  111. package/scripts/merge-gate.test.mjs +0 -613
  112. package/scripts/muted-git-guard-sweep.test.mjs +0 -493
  113. package/scripts/net-ssh-precedence-prose.test.mjs +0 -91
  114. package/scripts/net-watchdog-prose.test.mjs +0 -79
  115. package/scripts/net.test.mjs +0 -211
  116. package/scripts/no-ci-gate-prose.test.mjs +0 -108
  117. package/scripts/no-undo-audit.test.mjs +0 -3961
  118. package/scripts/no-wave.test.mjs +0 -138
  119. package/scripts/node-floor-sweep.test.mjs +0 -611
  120. package/scripts/pointer-target-prose.test.mjs +0 -186
  121. package/scripts/pr-cost.test.mjs +0 -521
  122. package/scripts/pr-overlap.test.mjs +0 -692
  123. package/scripts/printf-die-sweep.test.mjs +0 -384
  124. package/scripts/prompt-renderer.mjs +0 -40
  125. package/scripts/prompt-renderer.test.mjs +0 -68
  126. package/scripts/prose-pin.mjs +0 -807
  127. package/scripts/prose-pin.test.mjs +0 -1034
  128. package/scripts/prove-merge.test.mjs +0 -995
  129. package/scripts/pull-table-relabel-prose.test.mjs +0 -60
  130. package/scripts/quiet-payload-prose.test.mjs +0 -222
  131. package/scripts/reap-cherry-check-prose.test.mjs +0 -80
  132. package/scripts/reap.test.mjs +0 -4432
  133. package/scripts/reaping-prose.test.mjs +0 -239
  134. package/scripts/reaping-reap-authorizers-prose.test.mjs +0 -112
  135. package/scripts/reaping-release-authorizers-prose.test.mjs +0 -85
  136. package/scripts/recipe-derivation-prose.test.mjs +0 -112
  137. package/scripts/recipe-prove.test.mjs +0 -1042
  138. package/scripts/refuter-scratch-prose.test.mjs +0 -135
  139. package/scripts/release-label-required-prose.test.mjs +0 -71
  140. package/scripts/release-label-split-prose.test.mjs +0 -184
  141. package/scripts/release-ticket-reap-pairing-prose.test.mjs +0 -63
  142. package/scripts/release-ticket.test.mjs +0 -5099
  143. package/scripts/renovate-release-contract.test.mjs +0 -379
  144. package/scripts/repo-root.test.mjs +0 -873
  145. package/scripts/reporter-output-rule-prose.test.mjs +0 -208
  146. package/scripts/review-agent-report-only.test.mjs +0 -49
  147. package/scripts/review-core-booking.test.mjs +0 -35
  148. package/scripts/review-core-cwd-isolation.test.mjs +0 -374
  149. package/scripts/review-core-reads.test.mjs +0 -478
  150. package/scripts/review-core-refuter-scratch.test.mjs +0 -148
  151. package/scripts/review-core-scratch-refusal.test.mjs +0 -64
  152. package/scripts/review-core-shared-test-run.test.mjs +0 -275
  153. package/scripts/review-core-snapshot-path.test.mjs +0 -830
  154. package/scripts/review-core-specialist-scratch.test.mjs +0 -191
  155. package/scripts/review-core-testcmd.test.mjs +0 -335
  156. package/scripts/review-core-unrun.test.mjs +0 -566
  157. package/scripts/review-core-unverified-discriminant.test.mjs +0 -208
  158. package/scripts/review-eval-resolver-path.test.mjs +0 -55
  159. package/scripts/review-host-fixture.mjs +0 -89
  160. package/scripts/review-in-run-retry.test.mjs +0 -211
  161. package/scripts/review-no-vendor-dispatch.test.mjs +0 -45
  162. package/scripts/review-path-default.test.mjs +0 -796
  163. package/scripts/review-runner-spawns.test.mjs +0 -95
  164. package/scripts/review-runner.test.mjs +0 -166
  165. package/scripts/review-tier-audit.test.mjs +0 -140
  166. package/scripts/run-artifact-pr-prose.test.mjs +0 -341
  167. package/scripts/run-merge-bot-prose.test.mjs +0 -1413
  168. package/scripts/run-team-local-hook.test.mjs +0 -442
  169. package/scripts/scratch-partition-invariant-prose.test.mjs +0 -74
  170. package/scripts/scripts-path-citation-sweep.test.mjs +0 -73
  171. package/scripts/section-between-guard-prose.test.mjs +0 -379
  172. package/scripts/select-dimensions.test.mjs +0 -674
  173. package/scripts/shared-refusal.test.mjs +0 -388
  174. package/scripts/shell-traps-prose.test.mjs +0 -504
  175. package/scripts/shortlist.test.mjs +0 -380
  176. package/scripts/slow-transport.mjs +0 -72
  177. package/scripts/snapshot-repo.test.mjs +0 -1006
  178. package/scripts/snapshot-runner-audience-prose.test.mjs +0 -130
  179. package/scripts/spend-view.test.mjs +0 -312
  180. package/scripts/staleness-qualifier-prose.test.mjs +0 -251
  181. package/scripts/staleness-verdict-prose.test.mjs +0 -142
  182. package/scripts/staleness.test.mjs +0 -893
  183. package/scripts/stalled-member-liveness-prose.test.mjs +0 -219
  184. package/scripts/stash-prohibition-prose.test.mjs +0 -313
  185. package/scripts/strip-comments.mjs +0 -44
  186. package/scripts/temp-dir.mjs +0 -47
  187. package/scripts/ticket-router.test.mjs +0 -981
  188. package/scripts/tier-check.test.mjs +0 -828
  189. package/scripts/tier-outcomes-header.test.mjs +0 -518
  190. package/scripts/tier-outcomes.test.mjs +0 -572
  191. package/scripts/tier-roles.test.mjs +0 -535
  192. package/scripts/tracker-block-copy-prose.test.mjs +0 -288
  193. package/scripts/triage-labels-gloss-prose.test.mjs +0 -97
  194. package/scripts/unattended-git-sweep.test.mjs +0 -142
  195. package/scripts/verify-sha-exit2-prose.test.mjs +0 -269
  196. package/scripts/verify-sha.test.mjs +0 -692
  197. package/scripts/watcher-degraded-block.test.mjs +0 -369
  198. package/scripts/watcher-key-prose.test.mjs +0 -106
  199. package/scripts/within-run-pair-prose.test.mjs +0 -212
  200. package/scripts/worktree-audit.test.mjs +0 -1408
  201. package/scripts/worktree-head-prose.test.mjs +0 -276
  202. package/scripts/worktree-listing-sweep.test.mjs +0 -217
  203. package/scripts/worktree-model-prose.test.mjs +0 -216
  204. package/scripts/worktree-porcelain.mjs +0 -35
  205. package/scripts/worktree-porcelain.test.mjs +0 -102
@@ -1,374 +0,0 @@
1
- // #1433. Three PRs reviewed back-to-back from one omp `eval` cell left FOUR
2
- // files modified in the checkout that cell was standing in — one specialist's
3
- // mutation-test experiment per review reaching the repository through a
4
- // RELATIVE path. Review specialists have no assigned worktree to fall back to
5
- // (unlike implementers, whose own relative-path defect is #1411), and the
6
- // controller reads every gate decision — instruments.sh, ci-state.mjs, any
7
- // local `node --test` — out of that same tree, so a modified instrument
8
- // silently changes what a later gate check reads.
9
- //
10
- // WHY THE FIX IS PROMPT PROSE AND NOT A `cwd` ARGUMENT. The preferred remedy
11
- // was an explicit scratch cwd per specialist dispatch, passed through
12
- // review-eval.mjs's `ompAgent()`. Neither dispatch primitive has one:
13
- // eval's `agent()` takes `{ agent, label, schema, schemaMode, isolated, apply,
14
- // merge, tools }` and `task`'s item shape carries no cwd either, and a
15
- // non-isolated spawn runs with the PARENT's cwd. `isolated: true` is not that
16
- // option renamed — it names no directory, it fails preflight wherever
17
- // `task.isolation.enabled` is off, and it MERGES the child's changes back into
18
- // the parent, which for a mutation-testing specialist means applying a
19
- // deliberately broken tree to the checkout. review-eval.mjs's own header holds
20
- // that determination with its sources; this file pins what the prompts say
21
- // because of it.
22
- //
23
- // THREE PARTS, PINNED SEPARATELY, because the later two are inert without the
24
- // first and each fails differently: the inherited cwd is NAMED as a tree the
25
- // agent must not write to, `pwd` fixes which directory that is, and the
26
- // `CWD-AUDIT:` line is what makes a CLEAN run say so. An audit reported only
27
- // when it finds something is indistinguishable from one never run — the same
28
- // reading `unrunReason` already applies to a shared test run reporting nothing, and
29
- // the reason "four files dirty" was found by chance rather than by a report.
30
- //
31
- // WHY THESE PINS RENDER RATHER THAN GREP, and why the slice is the pin: same
32
- // two reasons review-core-specialist-scratch.test.mjs states. The rules live
33
- // inside template literals whose backticks are written escaped (`` \` ``), so
34
- // a source-text pin passes just as happily on a prompt showing the agent a
35
- // literal `\`; and all three prompts here carry the SAME wording, so a pin
36
- // matched against the whole file would be satisfied by any one of them —
37
- // which is the defect shape this ticket reports one dispatch at a time.
38
- // Each prompt is extracted by its own `agent()` call's anchors through
39
- // prompt-renderer.mjs and rendered, so the other two are outside it by
40
- // construction. Measured, per prompt: reverting the block out of the
41
- // specialist prompt alone reds that prompt's four tests and the shared marker
42
- // pin below, leaving every refuter and snapshot test green; the mirror revert
43
- // does the same the other way, and reverting the snapshot sentence reds its
44
- // prose pin alone. Block-commenting a dispatch reds the whole file at
45
- // extraction — it does not render as live text (strip-comments.mjs).
46
- //
47
- // THE CEILING, same as the sibling scratch pins': nothing here reaches the
48
- // agent's own obedience. These pins settle what a dispatch is TOLD and what
49
- // the schema can carry back — never where a specialist actually writes.
50
- import { test } from "node:test";
51
- import assert from "node:assert/strict";
52
- import { phrase } from "./prose-pin.mjs";
53
- import { promptRenderer } from "./prompt-renderer.mjs";
54
- import { AUDIT_COMMAND, AUDIT_STATES, CWD_AUDIT_MARKER, EVERY_RUN, PWD_FIRST, REFUTER_CWD, REPORT_FIELD, SCRATCH_NAMED_REFUTER, SCRATCH_NAMED_SPECIALIST, SPECIALIST_CWD, auditLine } from "./cwd-isolation-pins.mjs";
55
- import { FINDINGS_SCHEMA, VERDICT_SCHEMA, cwdAuditFrom, runReview } from "./review-core.mjs";
56
-
57
- const FILE = "scripts/review-core.mjs";
58
- const SNAP = { path: "/scr/run-1/snapshot-abc1234", head: "abc1234", runRoot: "/scr/run-1" };
59
-
60
- // One specialist's, one refuter's and the snapshot agent's prompt. Every
61
- // argument is fixed: these pins are about what the prompts SAY, not about how
62
- // they vary, so nothing here needs to. `sharedRunNote`/`readRules`/
63
- // `usableDiff`/`environmentNote` are stubs for the sibling pins' reason —
64
- // binding them keeps a render that stops passing one throwing a
65
- // ReferenceError instead of dropping a paragraph silently.
66
- const specialist = promptRenderer({
67
- file: FILE,
68
- start: "`Review PR #${pr} (branch ${branch}) for: ",
69
- end: "label: `review:",
70
- scope: ["pr", "branch", "d", "snap", "worktree", "stats", "sharedRunNote", "sharedRun", "failureOwner", "readRules", "usableDiff", "environmentNote"],
71
- what: "review-core.mjs's specialist prompt (opening `Review PR #${pr} (branch ${branch}) for: `, labelled `review:`)",
72
- })(7, "feature/x", { key: "correctness", prompt: "does it do what it says" }, SNAP, "/repo/.worktrees/7-x", null, () => "SHARED RUN NOTE", {}, "correctness", () => "READ RULES", () => null, () => "TEST ENVIRONMENT");
73
-
74
- const refuter = promptRenderer({
75
- file: FILE,
76
- start: "`Try to REFUTE this finding from PR #",
77
- end: "{ label: `verify:",
78
- scope: ["pr", "f", "snap", "stats", "d", "i", "fi", "readRules", "usableDiff", "environmentNote"],
79
- what: "review-core.mjs's refuter prompt (opening `Try to REFUTE this finding from PR #`, labelled `verify:`)",
80
- })(7, { claim: "the guard fails open", file: "a.js", line: 12, evidence: "line 12 has no else" }, SNAP, null, { key: "correctness" }, 0, 0, () => "READ RULES", () => null, () => "TEST ENVIRONMENT");
81
-
82
- const snapshot = promptRenderer({
83
- file: FILE,
84
- start: "`In ${worktree}, cut an immutable review snapshot",
85
- end: "{ label: `snapshot${forPr}`",
86
- scope: ["worktree", "scratch", "runRootParent", "runRootPrefix", "pr", "harness"],
87
- what: "review-core.mjs's snapshot prompt (opening `In ${worktree}, cut an immutable review snapshot`, labelled `snapshot:pr<N>`)",
88
- })("/repo/.worktrees/7-x", "/scr", "/scr/pr7", "/scr/pr7/run-", 7, "omp");
89
-
90
- // --- Part 1: the inherited cwd is named, as a tree not to write to ---------
91
- // ONE contiguous span per prompt, never two assertions and never a `.{0,N}`
92
- // gap — cwd-isolation-pins.mjs holds both spans and why their three clauses
93
- // have to arrive together. Mutation-tested on each prompt: deleting any one of
94
- // the three clauses, and splicing an exception between them ("…lands THERE,
95
- // unless the command is read-only"), both red these.
96
- test("the rendered specialist prompt names the cwd it inherits and says a relative path lands there", () => {
97
- assert.match(
98
- specialist,
99
- phrase(SPECIALIST_CWD),
100
- "review-core.mjs's specialist prompt no longer names the directory it is dispatched INTO — a specialist that does not know its cwd is the controller's checkout has no reason to leave it, " +
101
- "which is how three reviews left four files modified there (#1433)",
102
- );
103
- });
104
-
105
- test("the rendered refuter prompt names the cwd it inherits and says a relative path lands there", () => {
106
- assert.match(
107
- refuter,
108
- phrase(`your shell does not start there: ${REFUTER_CWD}`),
109
- "review-core.mjs's refuter prompt no longer names the directory it is dispatched INTO — a refuter applying a mutation is the likeliest writer of the four files #1433 measured",
110
- );
111
- });
112
-
113
- // --- Part 2: `pwd` first, and the inherited directory is a no-run zone -----
114
- // Both halves in one span (PWD_FIRST, cwd-isolation-pins.mjs says why).
115
- // Mutation-tested: demoting FIRST to "early" or dropping the no-run clause
116
- // reds these.
117
- for (const [name, prompt] of [
118
- ["specialist", specialist],
119
- ["refuter", refuter],
120
- ]) {
121
- test(`the rendered ${name} prompt orders pwd as the first command and makes that directory a no-run zone`, () => {
122
- assert.match(
123
- prompt,
124
- PWD_FIRST,
125
- `review-core.mjs's ${name} prompt no longer fixes which directory it started in before running anything — the audit below then has no path to audit, and the cd rules guard a directory nothing identified (#1433)`,
126
- );
127
- });
128
- }
129
-
130
- // --- Part 2.5 (#1721): where the snapshot/scratch dir are named, per prompt
131
- // Each prompt's own claim (SCRATCH_NAMED_SPECIALIST / SCRATCH_NAMED_REFUTER,
132
- // cwd-isolation-pins.mjs says why the two differ). Regression pin: reverting
133
- // review-core.mjs's specialist prompt back to "above" while leaving this file on
134
- // "in this prompt" passed the entire suite otherwise (measured at PR #1825's
135
- // review) — nothing previously pinned past PWD_FIRST's "no-run zone from then
136
- // on" on either prompt.
137
- for (const [name, prompt, span] of [
138
- ["specialist", specialist, SCRATCH_NAMED_SPECIALIST],
139
- ["refuter", refuter, SCRATCH_NAMED_REFUTER],
140
- ]) {
141
- test(`the rendered ${name} prompt states where the snapshot and scratch dir are named as absolute paths`, () => {
142
- assert.match(
143
- prompt,
144
- span,
145
- `review-core.mjs's ${name} prompt no longer makes this claim, or worded it to match "above" when its scratch dir is actually named later (#1721)`,
146
- );
147
- });
148
- }
149
-
150
- // --- Part 3: the positive self-check, into a field the schema carries ------
151
- // The audit command and the report contract are separate spans, each
152
- // contiguous, for the reasons cwd-isolation-pins.mjs gives. Mutation-tested:
153
- // dropping `-uall`, dropping the `showUntrackedFiles=no` reason, and replacing
154
- // "every run, clean or not" with "when it printed anything" each red one of
155
- // these.
156
- test("the rendered specialist prompt audits the directory it started in, with the explicit untracked mode", () => {
157
- assert.match(
158
- specialist,
159
- phrase(AUDIT_COMMAND),
160
- "review-core.mjs's specialist prompt dropped the stray-write audit, or asks for bare `--porcelain` — which a `status.showUntrackedFiles=no` config silences into a false clean, the same false-clean class worktree-audit.sh's own `-uall` exists to deny (#1433)",
161
- );
162
- });
163
-
164
- test("the rendered refuter prompt audits the directory it started in, with the explicit untracked mode", () => {
165
- assert.match(
166
- refuter,
167
- phrase(AUDIT_COMMAND),
168
- "review-core.mjs's refuter prompt dropped the stray-write audit, or asks for bare `--porcelain` (#1433)",
169
- );
170
- });
171
-
172
- // The report contract, and the half a prose pin usually cannot reach: the
173
- // field each prompt names has to be one its SCHEMA declares. Both schemas are
174
- // `additionalProperties: false`, so a rule pointing at any other name is inert
175
- // — the agent's audit is dropped in validation and the prompt still reads
176
- // correct. The field name is EXTRACTED from the rendered prompt by
177
- // REPORT_FIELD's capture group rather than transcribed here, so the pin fails
178
- // in either direction: the prompt renaming the field, or the schema losing it.
179
- for (const [name, prompt, schema, schemaName] of [
180
- ["specialist", specialist, FINDINGS_SCHEMA, "FINDINGS_SCHEMA"],
181
- ["refuter", refuter, VERDICT_SCHEMA, "VERDICT_SCHEMA"],
182
- ]) {
183
- test(`the rendered ${name} prompt reports its audit into a field ${schemaName} actually carries, every run`, () => {
184
- const named = prompt.match(REPORT_FIELD);
185
- assert.ok(
186
- named,
187
- `review-core.mjs's ${name} prompt no longer names a report field for the CWD-AUDIT line — an audit the agent has nowhere to put is an audit the caller never sees (#1433)`,
188
- );
189
- assert.equal(schema.additionalProperties, false, `${schemaName} no longer refuses undeclared fields — this pin's premise (a wrongly-named field is DROPPED, not passed through) no longer holds; re-derive it`);
190
- assert.ok(
191
- Object.hasOwn(schema.properties, named[1]),
192
- `review-core.mjs's ${name} prompt reports the audit in \`${named[1]}\`, which ${schemaName} does not declare — \`additionalProperties: false\` drops it, so the audit dies in validation with the prompt still reading correct`,
193
- );
194
- assert.ok(
195
- schema.required.includes(named[1]),
196
- `${schemaName} no longer REQUIRES \`${named[1]}\`, so an agent that omits the audit omits the whole field and validates anyway — the report has to ride a field the schema already insists on`,
197
- );
198
- // All three states, and the always clause. Two of the states are what keep
199
- // the check POSITIVE: a prompt that spells out only `dirty` leaves a clean
200
- // run and a never-run check reporting the same nothing, and `unrepo` is
201
- // what stops a `fatal: not a git repository` — the honest answer whenever
202
- // the inherited cwd is not a checkout — being read as a reason to report
203
- // nothing at all.
204
- for (const state of AUDIT_STATES) {
205
- assert.match(
206
- prompt,
207
- auditLine(state),
208
- `review-core.mjs's ${name} prompt no longer spells the \`${state}\` form of the audit line — the state it cannot spell is the state it will not report (#1433)`,
209
- );
210
- }
211
- assert.match(
212
- prompt,
213
- EVERY_RUN,
214
- `review-core.mjs's ${name} prompt no longer demands the audit line on a CLEAN run — an audit reported only when it finds something is indistinguishable from one never run, which is exactly how #1433's four files went unrecorded`,
215
- );
216
- });
217
- }
218
-
219
- // Checked against each prompt individually (CWD_AUDIT_MARKER,
220
- // cwd-isolation-pins.mjs) — a capture group around a fixed literal with no
221
- // alternation inside it can only ever capture that same literal, so comparing
222
- // `inSpecialist[1]` against `inRefuter[1]` (the prior shape of this test)
223
- // could never fail as long as both prompts matched at all; the `assert.ok`
224
- // above already covers that.
225
- for (const [name, prompt] of [
226
- ["specialist", specialist],
227
- ["refuter", refuter],
228
- ]) {
229
- test(`the rendered ${name} prompt names the CWD-AUDIT marker as a code span`, () => {
230
- assert.match(
231
- prompt,
232
- CWD_AUDIT_MARKER,
233
- `review-core.mjs's ${name} prompt no longer names the CWD-AUDIT marker as a code span — a reader cannot tell the literal from the prose around it`,
234
- );
235
- });
236
- }
237
-
238
- // --- The snapshot dispatch, whose rule is different ------------------------
239
- // It is dispatched with the same inherited cwd, but it CREATES the scratch
240
- // tree, so it has no scratch directory to `cd` into as a first command and
241
- // gets the other half of the rule instead: every path it uses is absolute or
242
- // `-C`-anchored, so nothing relative may be added to the block.
243
- test("the rendered snapshot prompt says its own block is absolute by construction and forbids adding a relative path", () => {
244
- assert.match(
245
- snapshot,
246
- phrase(
247
- "Every path in the block above is absolute or `-C`-anchored on purpose: this dispatch carries no working directory of its own either, so you start in the controller's own checkout, and a relative path — a `tar -x` with no `-C`, a bare `git` — reads or writes THERE",
248
- ),
249
- "review-core.mjs's snapshot prompt no longer says why its paths are absolute — the agent that mints the scratch tree is the one dispatch with no scratch dir to cd into, so the rule it gets is the only one it can follow (#1433)",
250
- );
251
- });
252
-
253
- // The other direction, and the one that keeps the sentence above from being
254
- // aspirational: the block really is anchored. Every command line in the prompt
255
- // is checked, so a bare `git status` or a `tar -x` with no `-C` added later
256
- // reds HERE — where the prose claiming otherwise lives — rather than in a
257
- // review that quietly measured the controller's checkout. Command lines are
258
- // the 4-space-indented ones, which is how both blocks in this prompt are
259
- // written; the prose that NAMES `tar -x` and `git` sits at column 0 and is not
260
- // scanned, so this pin cannot be satisfied or broken by its own rule's wording.
261
- test("every command line in the snapshot prompt is -C-anchored, cd-chained, or path-free", () => {
262
- const lines = snapshot.split("\n").filter((l) => /^ {4}\S/.test(l));
263
- assert.ok(lines.length >= 15, `only ${lines.length} indented command lines found in the snapshot prompt — the block was reshaped past what this pin scans; update it or restore the block`);
264
- for (const line of lines) {
265
- const cdChained = /cd\s+"\$\w+"\s+&&/.test(line);
266
- for (const m of line.matchAll(/(?<![-\w])git\s+(?!-C\b)(\S+)/g)) {
267
- assert.ok(
268
- cdChained,
269
- `the snapshot block runs \`git ${m[1]}\` with neither \`-C\` nor a chained \`cd\` on the same line — it answers about the directory the dispatch inherited, the controller's own checkout: ${line.trim()}`,
270
- );
271
- }
272
- for (const m of line.matchAll(/(?<![-\w])tar\s+(\S+)/g)) {
273
- assert.match(
274
- line,
275
- /tar\s[^|]*-C\s/,
276
- `the snapshot block runs \`tar ${m[1]}\` with no \`-C\` — an extraction with no destination unpacks into the inherited cwd: ${line.trim()}`,
277
- );
278
- }
279
- }
280
- });
281
-
282
- // --- The specialist's audit reaching the payload (PR #1671 review, gap 1) --
283
- // Part 3 above pins that the audit is NAMED and has somewhere valid to land;
284
- // it does not pin that `runReview` ever reads it back out once it lands
285
- // there. Measured against the PR #1671 review's own reproduction: a
286
- // specialist that dutifully reported `CWD-AUDIT: dirty <checkout>` into
287
- // `scope_searched` had that fact dropped before `runReview`'s return —
288
- // `scope_searched` fed only `unrunReason`'s test-run check, which never reads
289
- // it, so the audit dead-ended inside a field nothing else looked at.
290
- // `cwdAuditFrom` extracts the line; the tests below pin that `runReview`
291
- // folds its result into the payload under `cwdAudit`, and that a missing or
292
- // misspelled line is flagged there rather than silently read as clean.
293
- test("cwdAuditFrom reads each of the audit line's states, and flags its own absence", () => {
294
- assert.deepEqual(cwdAuditFrom("grepped src/**/*.js for auth checks. CWD-AUDIT: clean /repo/.worktrees/7-x"), {
295
- state: "clean",
296
- line: "CWD-AUDIT: clean /repo/.worktrees/7-x",
297
- });
298
- assert.deepEqual(cwdAuditFrom("CWD-AUDIT: dirty /repo/.worktrees/7-x — M src/foo.js"), {
299
- state: "dirty",
300
- line: "CWD-AUDIT: dirty /repo/.worktrees/7-x — M src/foo.js",
301
- });
302
- assert.deepEqual(cwdAuditFrom("CWD-AUDIT: unrepo /repo/.worktrees/7-x"), {
303
- state: "unrepo",
304
- line: "CWD-AUDIT: unrepo /repo/.worktrees/7-x",
305
- });
306
- // No line at all.
307
- assert.deepEqual(cwdAuditFrom("grepped src/**/*.js for auth checks"), { state: "missing", line: null });
308
- // Misspelled — the ticket's own failure mode: a schema-valid string that
309
- // never actually reports, because FINDINGS_SCHEMA has no pattern over it.
310
- assert.deepEqual(cwdAuditFrom("CWD AUDIT: clean /repo/.worktrees/7-x"), { state: "missing", line: null });
311
- assert.deepEqual(cwdAuditFrom(undefined), { state: "missing", line: null });
312
- });
313
-
314
- // A minimal fake host: every fact `runReview` otherwise gets from `gh`/`git`
315
- // arrives inside the snapshot agent's own structured response — review-
316
- // core.js never shells out itself, every command lives in a dispatched
317
- // prompt (grep finds no `child_process` import in this file) — so a scripted
318
- // `agent()` stub returning one canned object per dispatch label is a
319
- // complete double for the whole pipeline, no real repository needed.
320
- function fakeReviewHost(scopeSearched) {
321
- return {
322
- agent: async (_prompt, opts) => {
323
- if (opts.label === "snapshot:pr7")
324
- return {
325
- runRoot: "/scr/pr7/run-ab12",
326
- path: "/scr/pr7/run-ab12/snapshot-abc123",
327
- head: "abc123",
328
- pathVerified: true,
329
- repoVerified: true,
330
- testCmd: "node --test",
331
- };
332
- if (opts.label === "test-run:pr7") return { exitCode: 0, tests: 5, pass: 5, fail: 0 };
333
- if (opts.label === "review:correctness:pr7")
334
- return {
335
- dimension: "correctness",
336
- scope_searched: scopeSearched,
337
- findings: [],
338
- };
339
- throw new Error(`fakeReviewHost: unexpected dispatch ${opts.label}`);
340
- },
341
- phase: () => {},
342
- log: () => {},
343
- };
344
- }
345
-
346
- test("runReview carries a specialist's dirty CWD-AUDIT into the payload's cwdAudit field", async () => {
347
- const host = fakeReviewHost("CWD-AUDIT: dirty /repo/.worktrees/7-x — M src/foo.js");
348
- const result = await runReview(host, {
349
- pr: 7,
350
- worktree: "/repo/.worktrees/7-x",
351
- scratch: "/scr",
352
- dimensions: ["correctness"],
353
- });
354
- assert.deepEqual(
355
- result.cwdAudit,
356
- [{ dimension: "correctness", state: "dirty", line: "CWD-AUDIT: dirty /repo/.worktrees/7-x — M src/foo.js" }],
357
- "a specialist that reported a dirty checkout must have that fact reach the payload — silently dropping it is the exact defect PR #1671's review found (#1433)",
358
- );
359
- });
360
-
361
- test("runReview flags a specialist's missing CWD-AUDIT line rather than reading it as clean", async () => {
362
- const host = fakeReviewHost("grepped the diff for auth checks — nothing else searched");
363
- const result = await runReview(host, {
364
- pr: 7,
365
- worktree: "/repo/.worktrees/7-x",
366
- scratch: "/scr",
367
- dimensions: ["correctness"],
368
- });
369
- assert.deepEqual(
370
- result.cwdAudit,
371
- [{ dimension: "correctness", state: "missing", line: null }],
372
- "a specialist that satisfied FINDINGS_SCHEMA without ever emitting a CWD-AUDIT line must be flagged, not read as an unreported-but-clean run",
373
- );
374
- });