rungate 0.8.23 → 0.8.25
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/gates/.gate-salt +1 -1
- package/gates/gate-executor.ts +423 -96
- package/gates/orchestrator.test.ts +108 -2
- package/gates/orchestrator.ts +25 -5
- package/gates/prompts/prove-reproducer.md +14 -1
- package/gates/prompts.sha256 +1 -1
- package/gates/prove.test.ts +6 -1
- package/gates/schema.ts +25 -1
- package/gates/workflow.test.ts +12 -0
- package/lib/agents-md-sections.ts +4 -1
- package/lib/ai-review-suppression.ts +26 -2
- package/lib/artifact-schema.ts +37 -0
- package/lib/bash-write-guard.ts +667 -257
- package/lib/ci-reading.ts +186 -0
- package/lib/cli-commands.ts +8 -0
- package/lib/rule-provenance.ts +54 -3
- package/lib/scaffold/ai-review-step.ts +95 -83
- package/lib/scaffold/intake-workflow.ts +120 -0
- package/lib/scaffold/marked-section.ts +75 -0
- package/lib/scaffold/owned-files.ts +12 -2
- package/lib/scaffold/steps.ts +167 -4
- package/lib/ship-mode.ts +15 -3
- package/package.json +3 -1
- package/scripts/ai-review-findings.ts +6 -2
- package/scripts/ci-reading.ts +192 -0
- package/scripts/detect-sc-drift.ts +44 -2
- package/scripts/scaffold-project.ts +13 -1
- package/config/prove-evidence.schema.json +0 -83
|
@@ -1,7 +1,10 @@
|
|
|
1
1
|
import { test, expect, describe, beforeEach, afterEach } from "bun:test";
|
|
2
2
|
import { readFileSync, writeFileSync, mkdirSync, rmSync, existsSync, statSync } from "fs";
|
|
3
3
|
import { join } from "path";
|
|
4
|
+
import { z } from "zod";
|
|
4
5
|
import { gateSaltPath } from "../lib/paths";
|
|
6
|
+
import { validateArtifactSchema } from "../lib/artifact-schema";
|
|
7
|
+
import { ShipEvidenceSchema } from "./schema";
|
|
5
8
|
import {
|
|
6
9
|
writeGateResult,
|
|
7
10
|
generateHmac,
|
|
@@ -419,8 +422,10 @@ describe("generateShipEvidence", () => {
|
|
|
419
422
|
|
|
420
423
|
const path = generateShipEvidence(SF, TMP_DIR, 432, "", 1, 0, 0);
|
|
421
424
|
|
|
422
|
-
|
|
423
|
-
|
|
425
|
+
// null means the contract refused the artifact (#2255).
|
|
426
|
+
expect(path).not.toBeNull();
|
|
427
|
+
expect(existsSync(path!)).toBe(true);
|
|
428
|
+
const evidence = JSON.parse(readFileSync(path!, "utf-8"));
|
|
424
429
|
expect(evidence.contractVersion).toBe("1.0");
|
|
425
430
|
expect(evidence.issueNumber).toBe(432);
|
|
426
431
|
expect(evidence.gateOut.status).toBe("PASS");
|
|
@@ -429,6 +434,107 @@ describe("generateShipEvidence", () => {
|
|
|
429
434
|
expect(evidence.capturedAt).toBeTruthy();
|
|
430
435
|
expect(evidence.iterationCount).toBeGreaterThanOrEqual(1);
|
|
431
436
|
expect(evidence.sizing.predicted).toBe("M");
|
|
437
|
+
// #2255 / SC-2770: the artifact meets its declared contract, not only the
|
|
438
|
+
// eight fields this case happens to name.
|
|
439
|
+
expect(validateArtifactSchema(evidence, ShipEvidenceSchema)).toEqual([]);
|
|
440
|
+
});
|
|
441
|
+
|
|
442
|
+
test("#2255 / SC-2770: a schema with one required field more than the writer emits refuses the artifact", () => {
|
|
443
|
+
writeFileSync(SF, JSON.stringify(makeState({ phase: "SHIP" }), null, 2));
|
|
444
|
+
const path = generateShipEvidence(SF, TMP_DIR, 432, "", 1, 0, 0);
|
|
445
|
+
expect(path).not.toBeNull();
|
|
446
|
+
const evidence = JSON.parse(readFileSync(path!, "utf-8"));
|
|
447
|
+
const stricter = ShipEvidenceSchema.extend({ reviewedBy: z.string() });
|
|
448
|
+
expect(validateArtifactSchema(evidence, stricter).join("\n")).toContain("reviewedBy");
|
|
449
|
+
});
|
|
450
|
+
|
|
451
|
+
test("#2255: an artifact the contract refuses is not written, and the field is named", () => {
|
|
452
|
+
// A verdict outside PASS|FAIL|SKIP is the live way the writer can emit
|
|
453
|
+
// what ShipEvidenceSchema refuses: `result: ac.verdict || "SKIP"`.
|
|
454
|
+
writeFileSync(SF, JSON.stringify(makeState({
|
|
455
|
+
phase: "SHIP",
|
|
456
|
+
acs: [{ id: "SC-1", statement: "s", verdict: "UNMEASURED", evidence: { content: "x" } }],
|
|
457
|
+
}), null, 2));
|
|
458
|
+
const errors: string[] = [];
|
|
459
|
+
const original = console.error;
|
|
460
|
+
console.error = (...a: unknown[]) => { errors.push(a.map(String).join(" ")); };
|
|
461
|
+
let path: string | null;
|
|
462
|
+
try {
|
|
463
|
+
path = generateShipEvidence(SF, TMP_DIR, 432, "", 1, 0, 0);
|
|
464
|
+
} finally {
|
|
465
|
+
console.error = original;
|
|
466
|
+
}
|
|
467
|
+
expect(path).toBeNull();
|
|
468
|
+
expect(existsSync(join(TMP_DIR, "ship-evidence.json"))).toBe(false);
|
|
469
|
+
expect(errors.join("\n")).toContain("gateOut.evidence.0.result");
|
|
470
|
+
});
|
|
471
|
+
|
|
472
|
+
test("#2255 review round 1: an OUTCOME criterion still PENDING at ship is written — a legitimate ship-time state", () => {
|
|
473
|
+
// gates/workflow.test.ts: "OUTCOME ACs may be PENDING at ship time — they
|
|
474
|
+
// require Quinn/prove for verification", and enrichment writes the literal
|
|
475
|
+
// "PENDING". The contract refused it, so validation would have dropped the
|
|
476
|
+
// evidence on every normal run with an OUTCOME criterion outstanding.
|
|
477
|
+
writeFileSync(SF, JSON.stringify(makeState({
|
|
478
|
+
phase: "SHIP",
|
|
479
|
+
acs: [{ id: "SC-1", statement: "s", type: "OUTCOME", verdict: "PENDING", evidence: null }],
|
|
480
|
+
}), null, 2));
|
|
481
|
+
const path = generateShipEvidence(SF, TMP_DIR, 432, "", 1, 0, 0);
|
|
482
|
+
expect(path).not.toBeNull();
|
|
483
|
+
const evidence = JSON.parse(readFileSync(path!, "utf-8"));
|
|
484
|
+
expect(evidence.gateOut.evidence[0].result).toBe("PENDING");
|
|
485
|
+
expect(validateArtifactSchema(evidence, ShipEvidenceSchema)).toEqual([]);
|
|
486
|
+
});
|
|
487
|
+
});
|
|
488
|
+
|
|
489
|
+
describe("#2255 review round 1: a refused artifact is never reported to the issue as shipped", () => {
|
|
490
|
+
// A loopback stub stands in for the GitHub API (lib/github.ts allows exactly
|
|
491
|
+
// GitHub or loopback). `fixture/repo` makes verifiedPush refuse rather than
|
|
492
|
+
// push from this checkout (#430). The control case is what keeps the
|
|
493
|
+
// refusal's "no requests" from being vacuous: the same stub, a valid
|
|
494
|
+
// artifact, and the label and comment arrive.
|
|
495
|
+
async function requestsDuring(acs: Array<Record<string, unknown>>): Promise<string[]> {
|
|
496
|
+
const seen: string[] = [];
|
|
497
|
+
const server = Bun.serve({
|
|
498
|
+
port: 0,
|
|
499
|
+
hostname: "127.0.0.1",
|
|
500
|
+
fetch(req) {
|
|
501
|
+
seen.push(`${req.method} ${new URL(req.url).pathname}`);
|
|
502
|
+
return Response.json({});
|
|
503
|
+
},
|
|
504
|
+
});
|
|
505
|
+
try {
|
|
506
|
+
writeFileSync(SF, JSON.stringify(makeState({ phase: "SHIP", acs }), null, 2));
|
|
507
|
+
// A child process, so the credential is SUPPLIED rather than read: the
|
|
508
|
+
// tree resolves it in one place (resolveGitHubToken), and nothing here
|
|
509
|
+
// names it as a read. Async, because this process serves the stub, and
|
|
510
|
+
// a blocking spawn would stall the server the child is calling. The
|
|
511
|
+
// child's pending requests keep it alive until they have landed.
|
|
512
|
+
const child = Bun.spawn(
|
|
513
|
+
["bun", "-e", `import { generateShipEvidence } from ${JSON.stringify(join(import.meta.dir, "orchestrator.ts"))};` +
|
|
514
|
+
`generateShipEvidence(${JSON.stringify(SF)}, ${JSON.stringify(TMP_DIR)}, 432, "fixture/repo", 1, 0, 0);`],
|
|
515
|
+
{
|
|
516
|
+
cwd: join(import.meta.dir, ".."),
|
|
517
|
+
env: { ...process.env, GH_TOKEN: "test-token-not-real", GITHUB_API_URL: `http://127.0.0.1:${server.port}` },
|
|
518
|
+
stdout: "ignore",
|
|
519
|
+
stderr: "ignore",
|
|
520
|
+
},
|
|
521
|
+
);
|
|
522
|
+
await child.exited;
|
|
523
|
+
} finally {
|
|
524
|
+
server.stop(true);
|
|
525
|
+
}
|
|
526
|
+
return seen;
|
|
527
|
+
}
|
|
528
|
+
|
|
529
|
+
test("control: a valid artifact labels the issue shipped and posts the comment", async () => {
|
|
530
|
+
const seen = await requestsDuring([{ id: "SC-1", statement: "s", verdict: "PASS", evidence: { content: "ok" } }]);
|
|
531
|
+
expect(seen.some(r => r.endsWith("/issues/432/labels"))).toBe(true);
|
|
532
|
+
expect(seen.some(r => r.endsWith("/issues/432/comments"))).toBe(true);
|
|
533
|
+
});
|
|
534
|
+
|
|
535
|
+
test("a refused artifact sends nothing: no shipped label, no Ship Complete comment", async () => {
|
|
536
|
+
const seen = await requestsDuring([{ id: "SC-1", statement: "s", verdict: "UNMEASURED", evidence: { content: "x" } }]);
|
|
537
|
+
expect(seen).toEqual([]);
|
|
432
538
|
});
|
|
433
539
|
});
|
|
434
540
|
|
package/gates/orchestrator.ts
CHANGED
|
@@ -2,7 +2,8 @@ import { readFileSync, writeFileSync, existsSync, mkdirSync, chmodSync, statSync
|
|
|
2
2
|
import { join, dirname } from "path";
|
|
3
3
|
import { createHash, createHmac, randomBytes } from "crypto";
|
|
4
4
|
import { execSync } from "child_process";
|
|
5
|
-
import { WorkflowStateSchema, ACSchema, AfkBatchPlanSchema, REJECTED_SKIP_REASONS } from "./schema";
|
|
5
|
+
import { WorkflowStateSchema, ACSchema, AfkBatchPlanSchema, REJECTED_SKIP_REASONS, ShipEvidenceSchema } from "./schema";
|
|
6
|
+
import { validateArtifactSchema } from "../lib/artifact-schema";
|
|
6
7
|
import { ZodError } from "zod";
|
|
7
8
|
import { createGitHubClient, addLabels, addComment, type GitHubClient } from "../lib/github";
|
|
8
9
|
import { computeACHash } from "../lib/workflow-security";
|
|
@@ -384,7 +385,7 @@ export function generateShipEvidence(
|
|
|
384
385
|
passes: number,
|
|
385
386
|
fails: number,
|
|
386
387
|
warns: number,
|
|
387
|
-
): string {
|
|
388
|
+
): string | null {
|
|
388
389
|
const state = readState(sf);
|
|
389
390
|
const sha = gitSha();
|
|
390
391
|
const ts = isoNow();
|
|
@@ -483,12 +484,31 @@ export function generateShipEvidence(
|
|
|
483
484
|
...(pushOutcome ? { push: pushOutcome, pushLeftWorkLocal: leftWorkLocal(pushOutcome) } : {}),
|
|
484
485
|
};
|
|
485
486
|
|
|
487
|
+
// #2255 / SC-2770: checked against its one contract before it is written,
|
|
488
|
+
// as prove-evidence.json is (#1559). An artifact the schema refuses is not
|
|
489
|
+
// written, and the fields are named. The gate's verdict is untouched: by
|
|
490
|
+
// here the ship gate has already passed and the push has happened, and the
|
|
491
|
+
// missing artifact is the visible consequence.
|
|
486
492
|
const evidencePath = join(workDir, "ship-evidence.json");
|
|
487
|
-
|
|
493
|
+
const violations = validateArtifactSchema(evidence, ShipEvidenceSchema);
|
|
494
|
+
if (violations.length > 0) {
|
|
495
|
+
console.error("ship gate: ship-evidence.json NOT written — it fails ShipEvidenceSchema (SC-2770):");
|
|
496
|
+
for (const v of violations) console.error(` - ${v}`);
|
|
497
|
+
} else {
|
|
498
|
+
writeFileSync(evidencePath, JSON.stringify(evidence, null, 2));
|
|
499
|
+
}
|
|
488
500
|
|
|
489
501
|
// GitHub: add shipped label + post comment via Octokit (best-effort, fire-and-forget)
|
|
490
502
|
// Uses POST additive label strategy (D-5) — never PUT/replace
|
|
491
|
-
|
|
503
|
+
//
|
|
504
|
+
// Not when the contract refused the artifact (#2255): labelling the issue
|
|
505
|
+
// shipped and posting "Ship Complete" beside a refusal is a public success
|
|
506
|
+
// report the run's own record contradicts, and a label is harder to take
|
|
507
|
+
// back than to withhold.
|
|
508
|
+
if (violations.length > 0 && issueRepo && issue > 0) {
|
|
509
|
+
console.error(`ship gate: #${issue} NOT labelled shipped and no Ship Complete comment posted — the evidence was not recorded`);
|
|
510
|
+
}
|
|
511
|
+
if (violations.length === 0 && issueRepo && issue > 0) {
|
|
492
512
|
try {
|
|
493
513
|
const github = createGitHubClient();
|
|
494
514
|
addLabels(github, issueRepo, issue, ["shipped"]).catch(() => {});
|
|
@@ -512,7 +532,7 @@ export function generateShipEvidence(
|
|
|
512
532
|
} catch {}
|
|
513
533
|
}
|
|
514
534
|
|
|
515
|
-
return evidencePath;
|
|
535
|
+
return violations.length > 0 ? null : evidencePath;
|
|
516
536
|
}
|
|
517
537
|
|
|
518
538
|
const GATE_PENDING_PATH = join(
|
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
doc-type: reference
|
|
3
3
|
status: active
|
|
4
4
|
owner: jason
|
|
5
|
-
updated: 2026-
|
|
5
|
+
updated: 2026-11-15
|
|
6
6
|
---
|
|
7
7
|
|
|
8
8
|
You are a prove reproducer. Your objective: reproduce the bug described
|
|
@@ -44,9 +44,22 @@ Output JSON:
|
|
|
44
44
|
"evidence": [
|
|
45
45
|
{"type": "screenshot|api-response|console", "description": "what this shows"}
|
|
46
46
|
],
|
|
47
|
+
"afterEvidence": {
|
|
48
|
+
"type": "text|api|data",
|
|
49
|
+
"path": null,
|
|
50
|
+
"capturedAt": "ISO-8601 time you observed the after-state",
|
|
51
|
+
"environment": "local|prod",
|
|
52
|
+
"description": "what the after-state showed"
|
|
53
|
+
},
|
|
47
54
|
"gaps": []
|
|
48
55
|
}
|
|
49
56
|
|
|
57
|
+
`afterEvidence` is the after-state you observed yourself. `environment` is
|
|
58
|
+
where you observed it: `local` for this checkout or a dev server, `prod` for
|
|
59
|
+
the production URL. If you observed no after-state, leave `afterEvidence`
|
|
60
|
+
out. Never fill it in to make the record complete: a PROVEN without it is
|
|
61
|
+
recorded as INCONCLUSIVE.
|
|
62
|
+
|
|
50
63
|
Verdict rules:
|
|
51
64
|
- PROVEN: bug was reproduced in before-state AND fix verified in after-state
|
|
52
65
|
- UNPROVEN: fix does not resolve the issue (evidence shows problem persists)
|
package/gates/prompts.sha256
CHANGED
|
@@ -1,3 +1,3 @@
|
|
|
1
1
|
ba56e1fcb00c7a9bc1f2b05beca701cb264785e017cd2bf54a2dd07cdde331df ac-adversary.md
|
|
2
2
|
7d4ff0ef4b787e68b59643c11ba51e62a9411f5e24509b97855a337598c67508 evidence-validator.md
|
|
3
|
-
|
|
3
|
+
6f15ff6891118f749894f55f18fa21234211f993cf6eadfb3c90561a64b91691 prove-reproducer.md
|
package/gates/prove.test.ts
CHANGED
|
@@ -1,7 +1,8 @@
|
|
|
1
1
|
import { test, expect, describe } from "bun:test";
|
|
2
2
|
import { readFileSync, existsSync, writeFileSync } from "fs";
|
|
3
3
|
import { execSync } from "child_process";
|
|
4
|
-
import { beforeEvidenceFor, comparisonSummaryFor, readBeforeState } from "./gate-executor";
|
|
4
|
+
import { beforeEvidenceFor, comparisonSummaryFor, readBeforeState, validateArtifactSchema } from "./gate-executor";
|
|
5
|
+
import { ProveEvidenceSchema } from "./schema";
|
|
5
6
|
|
|
6
7
|
const TEST_DIR = process.env.TEST_WORK_DIR || "";
|
|
7
8
|
const PROJECT_ROOT = process.env.PROJECT_ROOT || process.cwd();
|
|
@@ -145,6 +146,10 @@ describe("prove: run evidence commands from workflow-state", () => {
|
|
|
145
146
|
criteriaResults,
|
|
146
147
|
};
|
|
147
148
|
|
|
149
|
+
// #2254 / SC-2770: checked against the one contract before it is written,
|
|
150
|
+
// as B3's and prove.js's are. A required field this writer stops emitting
|
|
151
|
+
// fails here, with the field named, instead of in whatever reads it next.
|
|
152
|
+
expect(validateArtifactSchema(evidence, ProveEvidenceSchema)).toEqual([]);
|
|
148
153
|
writeFileSync(`${TEST_DIR}/prove-evidence.json`, JSON.stringify(evidence, null, 2));
|
|
149
154
|
expect(verdict).toBe("PROVEN");
|
|
150
155
|
});
|
package/gates/schema.ts
CHANGED
|
@@ -1193,7 +1193,10 @@ export const ShipEvidenceSchema = z.object({
|
|
|
1193
1193
|
status: z.enum(["PASS", "FAIL"]),
|
|
1194
1194
|
evidence: z.array(z.object({
|
|
1195
1195
|
criterion: z.string(),
|
|
1196
|
-
|
|
1196
|
+
// PENDING is a legitimate ship-time state (#2255): an OUTCOME criterion
|
|
1197
|
+
// waits for Quinn or /prove, and the verify gate passes it through
|
|
1198
|
+
// (checkVerifyPendingACs). The writer copies the criterion's verdict.
|
|
1199
|
+
result: z.enum(["PASS", "FAIL", "SKIP", "PENDING"]),
|
|
1197
1200
|
detail: z.any().optional(),
|
|
1198
1201
|
})),
|
|
1199
1202
|
}),
|
|
@@ -1252,9 +1255,26 @@ const BeforeEvidenceSchema = z.union([
|
|
|
1252
1255
|
}),
|
|
1253
1256
|
]);
|
|
1254
1257
|
|
|
1258
|
+
/**
|
|
1259
|
+
* One row of `prove-evidence.json`'s `criteriaResults` (#1559). The shape all
|
|
1260
|
+
* three writers already emit — B3's reproducer, `gates/prove.test.ts` and
|
|
1261
|
+
* `workflows/prove.js` — and the one the deleted
|
|
1262
|
+
* `config/prove-evidence.schema.json` declared. SKIP is the reproducer's
|
|
1263
|
+
* verdict for a UI criterion it has no browser to check.
|
|
1264
|
+
*/
|
|
1265
|
+
const CriterionResultSchema = z.object({
|
|
1266
|
+
scId: z.string(),
|
|
1267
|
+
verdict: z.enum(["PASS", "FAIL", "SKIP"]),
|
|
1268
|
+
evidence: z.string(),
|
|
1269
|
+
});
|
|
1270
|
+
|
|
1255
1271
|
export const ProveEvidenceSchema = z.object({
|
|
1256
1272
|
issueNumber: z.number(),
|
|
1257
1273
|
verdict: z.enum(["PROVEN", "UNPROVEN", "INCONCLUSIVE"]),
|
|
1274
|
+
// Required, but not `.min(1)`: an INCONCLUSIVE with no criteria is a record
|
|
1275
|
+
// `gates/prove.test.ts` legitimately writes. That a PROVEN needs a PASS row
|
|
1276
|
+
// is `buildProveEvidence`'s rule, because it is about two fields at once.
|
|
1277
|
+
criteriaResults: z.array(CriterionResultSchema),
|
|
1258
1278
|
commitSHA: z.string(),
|
|
1259
1279
|
capturedAt: z.string(),
|
|
1260
1280
|
beforeEvidence: BeforeEvidenceSchema,
|
|
@@ -1264,6 +1284,10 @@ export const ProveEvidenceSchema = z.object({
|
|
|
1264
1284
|
capturedAt: z.string(),
|
|
1265
1285
|
environment: z.enum(["local", "prod"]),
|
|
1266
1286
|
description: z.string(),
|
|
1287
|
+
// Who observed the after-state (#2262): `AFTER_OBSERVERS` in
|
|
1288
|
+
// gates/gate-executor.ts, which decides the types each may report. Optional
|
|
1289
|
+
// because a `none` record has no observer.
|
|
1290
|
+
observedBy: z.enum(["reproducer", "quinn"]).optional(),
|
|
1267
1291
|
}),
|
|
1268
1292
|
comparisonSummary: z.string(),
|
|
1269
1293
|
quinnVerdict: z.enum(["PASS", "FAIL", "SKIP"]).optional(),
|
package/gates/workflow.test.ts
CHANGED
|
@@ -453,6 +453,18 @@ export function verifyFanoutViolations(label: string, val: unknown, tier: unknow
|
|
|
453
453
|
`is an absence with extra steps (#126)`,
|
|
454
454
|
];
|
|
455
455
|
}
|
|
456
|
+
// #2259: the agent ran and its reply could not be read. Before, that reply
|
|
457
|
+
// left no record, which the absence branch above refused, so the ship
|
|
458
|
+
// outcome is unchanged. What changes is that the refusal names the cause
|
|
459
|
+
// instead of failing on a missing `gameable` further down.
|
|
460
|
+
if ((val as { status?: unknown }).status === "UNMEASURED") {
|
|
461
|
+
const reason = (val as { reason?: unknown }).reason;
|
|
462
|
+
return [
|
|
463
|
+
`gates.verify.${label} is UNMEASURED — the agent ran and its reply could not be read` +
|
|
464
|
+
(typeof reason === "string" && reason.trim() !== "" ? `: ${reason}` : "") +
|
|
465
|
+
`. An unread reply is not an approval (#126, #2259).`,
|
|
466
|
+
];
|
|
467
|
+
}
|
|
456
468
|
return [];
|
|
457
469
|
}
|
|
458
470
|
|
|
@@ -119,6 +119,9 @@ export interface SectionReading {
|
|
|
119
119
|
text: string;
|
|
120
120
|
/** Its non-blank lines, heading included. The unit the loss check counts. */
|
|
121
121
|
lines: string[];
|
|
122
|
+
/** Where the heading matched, as an offset into the content (#2285): the one
|
|
123
|
+
* position, so a caller never re-finds the section by searching for its text. */
|
|
124
|
+
start: number;
|
|
122
125
|
}
|
|
123
126
|
|
|
124
127
|
export interface PreserveResult {
|
|
@@ -161,7 +164,7 @@ export function readPreservedSection(
|
|
|
161
164
|
// The heading alone is not a section. One line means nothing survived the
|
|
162
165
|
// last edit, and there is no content to promise anybody.
|
|
163
166
|
if (lines.length < 2) return null;
|
|
164
|
-
return { text, lines };
|
|
167
|
+
return { text, lines, start };
|
|
165
168
|
}
|
|
166
169
|
|
|
167
170
|
/**
|
|
@@ -224,8 +224,19 @@ export function aiReviewSuppressionGate(inputs: SuppressionInputs): AiReviewSupp
|
|
|
224
224
|
const a = anchorOnlyKey(k);
|
|
225
225
|
if (a !== null) anchors.add(a);
|
|
226
226
|
}
|
|
227
|
+
// #2257: how many findings cite a line, and how many of those lines could
|
|
228
|
+
// be read. Read from the SAME calls the keys are built from, not a second
|
|
229
|
+
// pass, so the count cannot disagree with what the keys saw.
|
|
230
|
+
let located = 0;
|
|
231
|
+
let readable = 0;
|
|
227
232
|
const repeat = candidates.map(c => {
|
|
228
|
-
const
|
|
233
|
+
const cited = c as ReviewFinding;
|
|
234
|
+
const lineText = inputs.lineTextFor(cited);
|
|
235
|
+
if (typeof cited?.file === "string" && Number.isInteger(cited?.line) && (cited.line as number) >= 1) {
|
|
236
|
+
located++;
|
|
237
|
+
if (lineText !== undefined) readable++;
|
|
238
|
+
}
|
|
239
|
+
const key = carryKey(c, lineText);
|
|
229
240
|
// Unkeyable CARRIES. It cannot be shown to be a repeat, and the
|
|
230
241
|
// direction this is allowed to be wrong in is toward reporting.
|
|
231
242
|
if (key === null) return null;
|
|
@@ -282,7 +293,20 @@ export function aiReviewSuppressionGate(inputs: SuppressionInputs): AiReviewSupp
|
|
|
282
293
|
// this whole increment is a correction for.
|
|
283
294
|
return null;
|
|
284
295
|
});
|
|
285
|
-
|
|
296
|
+
// #2257. A gate that could read no cited line built every key from a line
|
|
297
|
+
// NUMBER, which never equals a key an earlier round recorded from line
|
|
298
|
+
// TEXT. "Nothing to suppress" from that gate is a reading of nothing, and
|
|
299
|
+
// it printed the same words as a gate that read every line (#1872: 0 of
|
|
300
|
+
// 420 in CI). Byte-key matches above still stand; this says the anchor
|
|
301
|
+
// half was blind.
|
|
302
|
+
const unmeasured =
|
|
303
|
+
located > 0 && readable === 0 && prior.keys.length > 0
|
|
304
|
+
? `no cited line could be read for any of this round's ${located} finding(s) that cite one, ` +
|
|
305
|
+
`so no key could match one an earlier round recorded from the line's text, and a ` +
|
|
306
|
+
`restated finding could not be recognised. The checkout the gate reads is probably not ` +
|
|
307
|
+
`the one under review (#2257).`
|
|
308
|
+
: null;
|
|
309
|
+
return { repeat, unmeasured };
|
|
286
310
|
};
|
|
287
311
|
}
|
|
288
312
|
|
|
@@ -0,0 +1,37 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Check a gate artifact against its declared schema, before it is written
|
|
3
|
+
* (#1559, #2255, SC-2770).
|
|
4
|
+
*
|
|
5
|
+
* Lives in `lib/` rather than beside its first caller in
|
|
6
|
+
* `gates/gate-executor.ts` because `gates/orchestrator.ts` needs it too, and
|
|
7
|
+
* gate-executor already imports orchestrator. `gates/gate-executor.ts`
|
|
8
|
+
* re-exports both names, so there is one definition and every existing import
|
|
9
|
+
* keeps working.
|
|
10
|
+
*/
|
|
11
|
+
|
|
12
|
+
/**
|
|
13
|
+
* Anything with zod's `safeParse` shape. Structural, so callers keep no
|
|
14
|
+
* direct zod import and a test can hand it a stricter copy of a schema.
|
|
15
|
+
*/
|
|
16
|
+
export interface ArtifactSchema {
|
|
17
|
+
safeParse(value: unknown): {
|
|
18
|
+
success: boolean;
|
|
19
|
+
error?: { issues: ReadonlyArray<{ path: ReadonlyArray<PropertyKey>; message: string }> };
|
|
20
|
+
};
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
/**
|
|
24
|
+
* One `path: message` line per violation, `[]` when it validates.
|
|
25
|
+
*
|
|
26
|
+
* `config/prove-evidence.schema.json` declared a contract nothing read, so
|
|
27
|
+
* writer and contract drifted apart: B3's artifact failed both the JSON
|
|
28
|
+
* schema and the zod one, and no check said so. A refusal that names no
|
|
29
|
+
* issue is still a refusal — an empty list there would read as a pass.
|
|
30
|
+
*/
|
|
31
|
+
export function validateArtifactSchema(artifact: unknown, schema: ArtifactSchema): string[] {
|
|
32
|
+
const result = schema.safeParse(artifact);
|
|
33
|
+
if (result.success) return [];
|
|
34
|
+
const issues = result.error?.issues ?? [];
|
|
35
|
+
if (issues.length === 0) return ["(root): the schema refused the artifact without naming a field"];
|
|
36
|
+
return issues.map((i) => `${i.path.length > 0 ? i.path.map(String).join(".") : "(root)"}: ${i.message}`);
|
|
37
|
+
}
|