mirror of
https://github.com/garrytan/gstack.git
synced 2026-05-02 03:35:09 +02:00
f479134fba
Each test drives its plan-* skill through Step 0 then asserts the review-phase AskUserQuestion count falls in [N-1, N+2] for an N=5 seeded plan, plus D19: produced plan file ends with "## GSTACK REVIEW REPORT" as its last "## " heading. plan-ceo also runs a paired-finding positive control: 2 deliberately related findings should still produce 2 distinct AUQs, not 1 batched. Periodic-tier (gate-skipped without EVALS=1, EVALS_TIER=periodic). Sequential execution by plan §D15. Each fixture is inline TypeScript content delivered as a follow-up message after the slash command, per the proven pattern at skill-e2e-plan-design-with-ui.test.ts. Calibration loop (5 runs per skill) and the manual pre-merge negative check (D7 + D12) are required before merge per plan §Verification. NOT yet run. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
225 lines
8.6 KiB
TypeScript
225 lines
8.6 KiB
TypeScript
/**
|
|
* /plan-ceo-review per-finding AskUserQuestion count (periodic, paid, real-PTY).
|
|
*
|
|
* Asserts the load-bearing rule "One issue = one AskUserQuestion call" by
|
|
* driving /plan-ceo-review against a 5-finding seeded plan and counting
|
|
* distinct review-phase AUQs. Passes when count is in [N-1, N+2].
|
|
*
|
|
* Two tests in this file:
|
|
* - 5-finding distinct fixture: count band assertion + D19 review-report-at-bottom.
|
|
* - 2-finding paired control (D12 positive control): related findings still
|
|
* produce 2 distinct AUQs, not 1 batched, when the rule is honored.
|
|
*
|
|
* Tier: periodic. Each run drives Step 0 + 11 review sections end-to-end
|
|
* (~25 min, ~$5/run). Sequential by default per plan §D15. See
|
|
* test/helpers/claude-pty-runner.ts for runPlanSkillCounting internals.
|
|
*/
|
|
|
|
import { describe, test } from 'bun:test';
|
|
import * as fs from 'node:fs';
|
|
import {
|
|
runPlanSkillCounting,
|
|
ceoStep0Boundary,
|
|
assertReviewReportAtBottom,
|
|
} from './helpers/claude-pty-runner';
|
|
|
|
const shouldRun = !!process.env.EVALS && process.env.EVALS_TIER === 'periodic';
|
|
const describeE2E = shouldRun ? describe : describe.skip;
|
|
|
|
const N_DISTINCT = 5;
|
|
const FLOOR_DISTINCT = N_DISTINCT - 1; // 4 (D11)
|
|
const CEILING_DISTINCT = N_DISTINCT + 2; // 7 (D11)
|
|
|
|
const N_PAIRED = 2;
|
|
const FLOOR_PAIRED = 2;
|
|
const CEILING_PAIRED = 4;
|
|
|
|
const PLAN_CEO_5_FINDINGS = [
|
|
'Please review this plan thoroughly. As you go, write your plan-mode plan to /tmp/gstack-test-plan-ceo.md (use Edit/Write to that exact path).',
|
|
'',
|
|
'# Plan: Payment Processing Integration',
|
|
'',
|
|
'## Architecture',
|
|
"We're adding a new `PaymentService` class that will handle Stripe webhooks.",
|
|
'This bypasses the existing `WebhookDispatcher` module — we want a clean',
|
|
'namespace separation.',
|
|
'',
|
|
'## Database access',
|
|
'The new endpoint reads `request.params.userId` directly into a raw SQL',
|
|
'fragment for the lookup query.',
|
|
'',
|
|
'## Webhook fan-out',
|
|
'On payment success we update the user record AND fire a notification email.',
|
|
'Both happen inline; no error handling on the email leg.',
|
|
'',
|
|
'## Tests',
|
|
"None planned. We'll rely on the existing integration suite catching regressions.",
|
|
'',
|
|
'## Performance',
|
|
'Each webhook lookup hits the database for the user, then fetches each',
|
|
'order in a loop.',
|
|
].join('\n');
|
|
|
|
const PLAN_CEO_2_PAIRED_FINDINGS = [
|
|
'Please review this plan thoroughly. As you go, write your plan-mode plan to /tmp/gstack-test-plan-ceo-paired.md (use Edit/Write to that exact path).',
|
|
'',
|
|
'# Plan: Payment Processing — Test Coverage',
|
|
'',
|
|
'## Tests',
|
|
'We need test coverage for `processPayment()`. Specifically:',
|
|
'1. The happy path (successful Stripe charge — assert correct receipt is generated).',
|
|
'2. The error/timeout path (Stripe returns 502 — assert retry-with-backoff fires once, then fails clean).',
|
|
'',
|
|
'Currently neither has a unit test. These are deliberately separate concerns:',
|
|
'the success path is correctness, the failure path is graceful degradation.',
|
|
].join('\n');
|
|
|
|
const PLAN_CEO_PATH = '/tmp/gstack-test-plan-ceo.md';
|
|
const PLAN_CEO_PAIRED_PATH = '/tmp/gstack-test-plan-ceo-paired.md';
|
|
|
|
describeE2E('/plan-ceo-review per-finding AskUserQuestion count (periodic)', () => {
|
|
test(
|
|
`5-finding plan emits ${FLOOR_DISTINCT}-${CEILING_DISTINCT} review-phase AskUserQuestions`,
|
|
async () => {
|
|
try {
|
|
fs.rmSync(PLAN_CEO_PATH, { force: true });
|
|
} catch {
|
|
/* best-effort */
|
|
}
|
|
|
|
const obs = await runPlanSkillCounting({
|
|
skillName: 'plan-ceo-review',
|
|
slashCommand: '/plan-ceo-review',
|
|
followUpPrompt: PLAN_CEO_5_FINDINGS,
|
|
isLastStep0AUQ: ceoStep0Boundary,
|
|
reviewCountCeiling: CEILING_DISTINCT + 1, // hard cap above assertion ceiling
|
|
cwd: process.cwd(),
|
|
timeoutMs: 1_500_000, // 25 min
|
|
env: { QUESTION_TUNING: 'false', EXPLAIN_LEVEL: 'default' },
|
|
});
|
|
|
|
try {
|
|
if (!['plan_ready', 'completion_summary', 'ceiling_reached'].includes(obs.outcome)) {
|
|
throw new Error(
|
|
`plan-ceo-review finding-count FAILED: outcome=${obs.outcome}\n` +
|
|
`step0=${obs.step0Count} review=${obs.reviewCount} elapsed=${obs.elapsedMs}ms\n` +
|
|
`fingerprints (last 8):\n` +
|
|
obs.fingerprints
|
|
.slice(-8)
|
|
.map(
|
|
(f, i) =>
|
|
` ${i}. preReview=${f.preReview} sig=${f.signature.slice(0, 12)} prompt="${f.promptSnippet.slice(0, 60)}"`,
|
|
)
|
|
.join('\n') +
|
|
`\n--- evidence (last 3KB) ---\n${obs.evidence}`,
|
|
);
|
|
}
|
|
if (obs.reviewCount < FLOOR_DISTINCT) {
|
|
throw new Error(
|
|
`BAND FAIL (below floor): reviewCount=${obs.reviewCount} < FLOOR=${FLOOR_DISTINCT}.\n` +
|
|
`Likely batching regression — agent collapsed multiple findings into fewer questions.\n` +
|
|
`Fingerprints (review-phase only):\n` +
|
|
obs.fingerprints
|
|
.filter((f) => !f.preReview)
|
|
.map((f) => ` - "${f.promptSnippet.slice(0, 80)}"`)
|
|
.join('\n'),
|
|
);
|
|
}
|
|
if (obs.reviewCount > CEILING_DISTINCT) {
|
|
throw new Error(
|
|
`BAND FAIL (above ceiling): reviewCount=${obs.reviewCount} > CEILING=${CEILING_DISTINCT}.\n` +
|
|
`Possible over-asking regression. Review-phase fingerprints:\n` +
|
|
obs.fingerprints
|
|
.filter((f) => !f.preReview)
|
|
.map((f) => ` - "${f.promptSnippet.slice(0, 80)}"`)
|
|
.join('\n'),
|
|
);
|
|
}
|
|
|
|
// D19: review report at bottom of plan file.
|
|
if (!fs.existsSync(PLAN_CEO_PATH)) {
|
|
throw new Error(
|
|
`D19 FAIL: agent did not produce expected plan file at ${PLAN_CEO_PATH}.\n` +
|
|
`Either the agent ignored the path instruction in the follow-up prompt, or\n` +
|
|
`the helper exited before the agent wrote the file. ` +
|
|
`outcome=${obs.outcome} review=${obs.reviewCount}`,
|
|
);
|
|
}
|
|
const planContent = fs.readFileSync(PLAN_CEO_PATH, 'utf-8');
|
|
const verdict = assertReviewReportAtBottom(planContent);
|
|
if (!verdict.ok) {
|
|
throw new Error(
|
|
`D19 FAIL: plan file at ${PLAN_CEO_PATH} ${verdict.reason}\n` +
|
|
(verdict.trailingHeadings
|
|
? `Trailing headings: ${verdict.trailingHeadings.join(' | ')}\n`
|
|
: '') +
|
|
`--- plan content (last 1KB) ---\n${planContent.slice(-1024)}`,
|
|
);
|
|
}
|
|
} finally {
|
|
try {
|
|
fs.rmSync(PLAN_CEO_PATH, { force: true });
|
|
} catch {
|
|
/* best-effort */
|
|
}
|
|
}
|
|
},
|
|
1_700_000,
|
|
);
|
|
|
|
test(
|
|
`paired-finding positive control: ${N_PAIRED} related findings produce ${FLOOR_PAIRED}-${CEILING_PAIRED} AskUserQuestions`,
|
|
async () => {
|
|
try {
|
|
fs.rmSync(PLAN_CEO_PAIRED_PATH, { force: true });
|
|
} catch {
|
|
/* best-effort */
|
|
}
|
|
|
|
const obs = await runPlanSkillCounting({
|
|
skillName: 'plan-ceo-review',
|
|
slashCommand: '/plan-ceo-review',
|
|
followUpPrompt: PLAN_CEO_2_PAIRED_FINDINGS,
|
|
isLastStep0AUQ: ceoStep0Boundary,
|
|
reviewCountCeiling: CEILING_PAIRED + 1,
|
|
cwd: process.cwd(),
|
|
timeoutMs: 1_500_000,
|
|
env: { QUESTION_TUNING: 'false', EXPLAIN_LEVEL: 'default' },
|
|
});
|
|
|
|
try {
|
|
if (!['plan_ready', 'completion_summary', 'ceiling_reached'].includes(obs.outcome)) {
|
|
throw new Error(
|
|
`paired-finding control FAILED: outcome=${obs.outcome}\n` +
|
|
`step0=${obs.step0Count} review=${obs.reviewCount}\n` +
|
|
`--- evidence (last 3KB) ---\n${obs.evidence}`,
|
|
);
|
|
}
|
|
if (obs.reviewCount < FLOOR_PAIRED) {
|
|
throw new Error(
|
|
`PAIRED CONTROL FAIL: reviewCount=${obs.reviewCount} < FLOOR=${FLOOR_PAIRED}.\n` +
|
|
`Two deliberately related findings were batched into <2 questions — the rule failed under D12.\n` +
|
|
`Review-phase fingerprints:\n` +
|
|
obs.fingerprints
|
|
.filter((f) => !f.preReview)
|
|
.map((f) => ` - "${f.promptSnippet.slice(0, 80)}"`)
|
|
.join('\n'),
|
|
);
|
|
}
|
|
if (obs.reviewCount > CEILING_PAIRED) {
|
|
throw new Error(
|
|
`PAIRED CONTROL FAIL: reviewCount=${obs.reviewCount} > CEILING=${CEILING_PAIRED} (over-asking on a 2-finding fixture).`,
|
|
);
|
|
}
|
|
} finally {
|
|
try {
|
|
fs.rmSync(PLAN_CEO_PAIRED_PATH, { force: true });
|
|
} catch {
|
|
/* best-effort */
|
|
}
|
|
}
|
|
},
|
|
1_700_000,
|
|
);
|
|
});
|