rungate 0.8.2 → 0.8.4
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.
- package/bin/rungate.ts +11 -6
- package/config/prove-evidence.schema.json +21 -3
- package/gates/.gate-salt +1 -1
- package/gates/gate-executor.ts +153 -1
- package/gates/prove.test.ts +12 -6
- package/gates/schema.ts +69 -6
- package/gates/workflow.test.ts +12 -5
- package/hooks/BunTestPathGuard.hook.ts +106 -0
- package/hooks/DocHygiene.hook.ts +2 -1
- package/hooks/GateResultProtection.hook.ts +28 -7
- package/hooks/HookCurrency.hook.ts +48 -1
- package/hooks/TopLevelCdGuard.hook.ts +38 -6
- package/hooks/VerifyPhaseLock.hook.ts +25 -8
- package/hooks/WorkflowStateGuard.hook.ts +24 -7
- package/lib/ai-review-jury.ts +725 -0
- package/lib/ai-review-rounds.ts +791 -4
- package/lib/ai-review-unmeasured.ts +361 -0
- package/lib/bash-write-guard.ts +1235 -102
- package/lib/bun-test-paths.ts +253 -0
- package/lib/cli-commands.ts +252 -11
- package/lib/closing-reference-audit.ts +163 -0
- package/lib/conformity.ts +17 -1
- package/lib/create-sc.ts +45 -12
- package/lib/derived-spec.ts +206 -0
- package/lib/file-claims.ts +335 -0
- package/lib/finding-arbiter.ts +502 -28
- package/lib/finding-jury.ts +61 -5
- package/lib/finding-pr-routing.ts +333 -0
- package/lib/frontmatter.ts +16 -0
- package/lib/gate-check-audit.ts +646 -0
- package/lib/generators/agents-md.ts +29 -0
- package/lib/github.ts +112 -7
- package/lib/hook-tree-currency.ts +494 -0
- package/lib/issue-closure-audit.ts +161 -14
- package/lib/issue-create-override.ts +53 -3
- package/lib/issue-filing-gate.ts +1073 -76
- package/lib/local-review-rounds.ts +523 -0
- package/lib/organize.ts +21 -1
- package/lib/project-state-shape.ts +56 -5
- package/lib/release-preflight.ts +90 -1
- package/lib/repo-path.ts +94 -0
- package/lib/report-provenance.ts +218 -3
- package/lib/risk-tier.ts +302 -0
- package/lib/rounds-to-pass.ts +274 -0
- package/lib/rule-row-drift.ts +199 -0
- package/lib/sc-claim-forms.ts +33 -5
- package/lib/sc-guard.ts +11 -1
- package/lib/sc-id-cross-tree.ts +17 -2
- package/lib/sc-ids.ts +71 -6
- package/lib/sc-reservations.ts +158 -14
- package/lib/scaffold/ai-review-findings.ts +1369 -27
- package/lib/scaffold/ai-review-step.ts +4095 -119
- package/lib/scaffold/ai-review-unmeasured-workflow.ts +243 -0
- package/lib/scaffold/owned-files.ts +14 -0
- package/lib/scaffold/steps.ts +100 -9
- package/lib/scratch-roots.ts +93 -3
- package/lib/spec-updater.ts +22 -2
- package/lib/spec-write-ast.ts +728 -0
- package/lib/spec-write-sweep.ts +470 -0
- package/lib/test-suite-lock.ts +127 -13
- package/lib/ts-ast.ts +26 -0
- package/lib/validators/spec-validators.ts +36 -7
- package/lib/write-spec.ts +138 -0
- package/package.json +2 -1
- package/scripts/ai-review-findings.ts +41 -1
- package/scripts/arbitrate-finding.ts +330 -12
- package/scripts/audit-specs.ts +225 -24
- package/scripts/convert-spec.ts +7 -2
- package/scripts/create-sc.ts +14 -2
- package/scripts/create-spec.ts +5 -2
- package/scripts/detect-sc-drift.ts +10 -2
- package/scripts/record-ai-review-unmeasured.ts +370 -0
- package/scripts/split-spec.ts +17 -4
- package/scripts/sync-sc-status.ts +14 -2
|
@@ -28,12 +28,44 @@
|
|
|
28
28
|
* `.claude/rungate/hooks.json`, the current-state table in
|
|
29
29
|
* `specs/HOOK-ARCHITECTURE-SPEC.md`, and `.claude/settings.json`.
|
|
30
30
|
*
|
|
31
|
-
*
|
|
32
|
-
*
|
|
33
|
-
*
|
|
34
|
-
*
|
|
35
|
-
*
|
|
36
|
-
*
|
|
31
|
+
* ## NOT LIVE. User scope since #1578, and why the first classification was wrong
|
|
32
|
+
*
|
|
33
|
+
* #992 registered this at PROJECT scope, reasoning that `${CLAUDE_PROJECT_DIR}`
|
|
34
|
+
* resolves to the worktree a session is in, that every worktree carries the
|
|
35
|
+
* committed `.claude/settings.json`, and that worktree isolation is a harness
|
|
36
|
+
* concern. **The second clause is true and the conclusion does not follow.**
|
|
37
|
+
*
|
|
38
|
+
* Measured on 2026-10-24 in a worktree whose committed `.claude/settings.json`
|
|
39
|
+
* registered this hook: three bare `cd` calls were NOT refused, while a
|
|
40
|
+
* user-scoped COMP-7 refusal fired on the same `Bash` matcher in the same
|
|
41
|
+
* session. The decider returns `block` when invoked by hand from that cwd;
|
|
42
|
+
* `isHarnessProject()` resolves; `RUNGATE_CD_GUARD` is unset. The worktree
|
|
43
|
+
* carries the file. The session does not load it. So this hook has been
|
|
44
|
+
* green, registered, tested and INERT since #1050 — which is #1578, and is
|
|
45
|
+
* the purest form of *absence of evidence recorded as clean*.
|
|
46
|
+
*
|
|
47
|
+
* The scope follows from what a hook PROTECTS. This one protects the
|
|
48
|
+
* SESSION's primary working directory — the directory `isolation: 'worktree'`
|
|
49
|
+
* resolves against — and a session may be rooted anywhere while working on
|
|
50
|
+
* this repo, including via an added working directory. That is machine-wide
|
|
51
|
+
* state, so SC-518 governs it: machine-wide hooks live in user settings,
|
|
52
|
+
* because project settings only load for sessions started in that directory.
|
|
53
|
+
* It was misclassified from the start.
|
|
54
|
+
*
|
|
55
|
+
* Two consequences, both deliberate:
|
|
56
|
+
*
|
|
57
|
+
* - The repo no longer registers it in `.claude/settings.json`. SC-518 says
|
|
58
|
+
* the two scopes are not de-duplicated, so leaving it there would fire the
|
|
59
|
+
* hook twice once the real registration exists.
|
|
60
|
+
* - `test/hook-registration-reconciler.test.ts` now reports it UNMEASURED on
|
|
61
|
+
* every run, and that reading is the deliverable rather than a loss. A test
|
|
62
|
+
* saying *this cannot be verified here* beats one reporting a registration
|
|
63
|
+
* nobody loads as satisfied.
|
|
64
|
+
*
|
|
65
|
+
* **Live registration in `~/.claude/settings.json` is a human action, is
|
|
66
|
+
* outside every agent's permitted scope, and is STILL PENDING.** Merging this
|
|
67
|
+
* does not switch the guard on. Until a human does that, this hook refuses
|
|
68
|
+
* nothing.
|
|
37
69
|
*
|
|
38
70
|
* ## The rung, and what #1040 left behind
|
|
39
71
|
*
|
|
@@ -102,21 +102,38 @@ async function main() {
|
|
|
102
102
|
// inside one phase would leave the intake rate untouched.
|
|
103
103
|
const gate = await filingGate(cmd, { projectRoot: process.cwd() });
|
|
104
104
|
const unrecorded = gate.rows.filter(row => !recordAdjudication(row)).length;
|
|
105
|
+
// A refusal says the override "is recorded beside the refusal it waived".
|
|
106
|
+
// When the write failed that sentence is false, and a security control
|
|
107
|
+
// asserting a guarantee it is not providing is worse than one that says
|
|
108
|
+
// nothing — so the unrecorded notice rides ON the refusal rather than
|
|
109
|
+
// being skipped by the early exit, which is where it used to go.
|
|
110
|
+
const unrecordedNote =
|
|
111
|
+
unrecorded > 0
|
|
112
|
+
? `[FindingArbiter] ${unrecorded} adjudication(s) could NOT be written to ${ledgerPath()}, ` +
|
|
113
|
+
"so they are not in the population that prices this gate and any claim below that " +
|
|
114
|
+
"a decision was recorded does not hold for them."
|
|
115
|
+
: null;
|
|
105
116
|
|
|
117
|
+
// `gate.note` rides on the refusal for the same reason `unrecordedNote`
|
|
118
|
+
// does, and it was missed when that one was fixed: a `decision: "block"`
|
|
119
|
+
// payload has ONE text field, so anything not folded into `reason` here is
|
|
120
|
+
// discarded by the early exit and never reaches the agent.
|
|
121
|
+
//
|
|
122
|
+
// The note that mattered is the corpus-drop one. A refusal computed against
|
|
123
|
+
// a tracker three issues short is exactly where the shortfall needs saying
|
|
124
|
+
// — the agent is being told its finding is already filed, by a comparison
|
|
125
|
+
// that did not read the whole tracker — and that was the one path on which
|
|
126
|
+
// it was silently dropped. `FG-60` is the case; it went red here first.
|
|
106
127
|
if (gate.refusal !== null) {
|
|
107
|
-
|
|
128
|
+
const riders = [gate.note, unrecordedNote].filter((n): n is string => n !== null);
|
|
129
|
+
const reason = [gate.refusal, ...riders].join("\n\n");
|
|
130
|
+
console.log(JSON.stringify({ decision: "block", reason }));
|
|
108
131
|
process.exit(0);
|
|
109
132
|
}
|
|
110
133
|
// systemMessage and not stderr: stderr from a hook that exits 0 is not
|
|
111
134
|
// shown. An UNMEASURED nobody sees is a silent pass, which is the one thing
|
|
112
135
|
// lib/issue-filing-gate.ts's header says this must not become.
|
|
113
|
-
const notes = [
|
|
114
|
-
gate.note,
|
|
115
|
-
unrecorded > 0
|
|
116
|
-
? `[FindingArbiter] ${unrecorded} adjudication(s) could not be written to ${ledgerPath()}, ` +
|
|
117
|
-
`so this filing is not in the population that prices the gate.`
|
|
118
|
-
: null,
|
|
119
|
-
].filter((n): n is string => n !== null);
|
|
136
|
+
const notes = [gate.note, unrecordedNote].filter((n): n is string => n !== null);
|
|
120
137
|
if (notes.length > 0) console.log(JSON.stringify({ systemMessage: notes.join("\n\n") }));
|
|
121
138
|
process.exit(0);
|
|
122
139
|
}
|
|
@@ -19,12 +19,22 @@
|
|
|
19
19
|
* - `GateResultProtection` is authoritative for **what a permitted write may
|
|
20
20
|
* contain** — forged, erased or changed gate results, and `phase` DONE.
|
|
21
21
|
*
|
|
22
|
-
*
|
|
23
|
-
*
|
|
24
|
-
*
|
|
25
|
-
*
|
|
26
|
-
*
|
|
27
|
-
*
|
|
22
|
+
* THIS hook is the one that ships (`deployToConsumers: true`); the specific
|
|
23
|
+
* guard is OPT-IN and a scaffolded consumer does not get it. So on a consumer
|
|
24
|
+
* tree this is the ONLY guard on the artifact, and that is why its matcher
|
|
25
|
+
* being case-SENSITIVE while the specific guard's was case-folded (#1184) was
|
|
26
|
+
* a consumer-reachable gap and not a tidiness question. #1211 closed it: the
|
|
27
|
+
* matcher below case-folds, so the two predicates now agree on WHICH paths are
|
|
28
|
+
* the artifact and differ only in reach — `.includes()` here against
|
|
29
|
+
* `.endsWith()` there.
|
|
30
|
+
*
|
|
31
|
+
* The specific guard's refusals are therefore a SUBSET of this hook's. That is
|
|
32
|
+
* the safe direction and it is asserted, not assumed: this hook refusing every
|
|
33
|
+
* Write/Edit of the artifact means a consumer tree is covered for the question
|
|
34
|
+
* "may a tool call write this at all", which is the only question a consumer
|
|
35
|
+
* tree has a guard for. Subsumed is not dead — the two do not ship to the same
|
|
36
|
+
* trees, and the specific guard's refusal messages name the forgery while this
|
|
37
|
+
* one names the sanctioned writer.
|
|
28
38
|
*
|
|
29
39
|
* The contract is executed, not described:
|
|
30
40
|
* `test/workflow-state-write-contract.test.ts` runs both hooks over one corpus
|
|
@@ -47,7 +57,14 @@ async function main() {
|
|
|
47
57
|
// Write tool uses file_path; Edit tool uses file_path
|
|
48
58
|
const path = (toolInput.file_path as string) || (toolInput.path as string) || '';
|
|
49
59
|
|
|
50
|
-
|
|
60
|
+
// Case-FOLDED (#1211). `WORKFLOW-STATE.JSON` and `workflow-state.json` are one
|
|
61
|
+
// file on a case-insensitive filesystem — measured in #1184, which closed this
|
|
62
|
+
// in GateResultProtection and never carried the fix here. This is the guard
|
|
63
|
+
// that ships, so the uppercase spelling was reaching a consumer's disk with no
|
|
64
|
+
// guard at all. The population the fold newly claims is enumerated and bounded
|
|
65
|
+
// in test/workflow-state-guard-case.test.ts; it is case-variants of this name
|
|
66
|
+
// and nothing else.
|
|
67
|
+
if (!path.toLowerCase().includes('workflow-state.json')) process.exit(0);
|
|
51
68
|
|
|
52
69
|
// Block the write
|
|
53
70
|
const reason = [
|