mirror of https://github.com/garrytan/gstack.git
150 lines
6.8 KiB
TypeScript
150 lines
6.8 KiB
TypeScript
/**
|
|
* Plan-mode-info no-op regression (gate tier, paid, real-PTY).
|
|
*
|
|
* Asserts: when a plan-review skill is invoked OUTSIDE plan mode (no
|
|
* --permission-mode plan flag, no plan-mode reminder injected), the skill
|
|
* still reaches a terminal outcome ('asked' or 'plan_ready'). This is the
|
|
* negative coverage to the per-skill plan-mode smokes — if plan-mode-keyed
|
|
* behavior ever starts misfiring for non-plan-mode sessions (e.g., gating
|
|
* questions on a phrase that isn't there, or the plan-eng/plan-design
|
|
* scope-gate auto-select-B bypass firing without plan mode), this test
|
|
* catches it.
|
|
*
|
|
* Why this matters: outside plan mode, claude doesn't render a native
|
|
* confirmation UI. The skill must drive its own AskUserQuestion. Same
|
|
* runner, same outcome contract — just `inPlanMode: false`.
|
|
*
|
|
* Coverage grew with the scope-gate bypass (plan-mode auto-select B):
|
|
* - plan-ceo-review: original preamble-misfire regression.
|
|
* - plan-eng-review / plan-design-review: the bypass must NOT fire outside
|
|
* plan mode (scopeGateAutoSelectObserved stays false), and when the run
|
|
* ends in 'asked', the question that fired must be the scope gate itself
|
|
* (outside plan mode with no named target, the gate is the FIRST
|
|
* question by contract).
|
|
* - named-target case: a pasted draft (initialPlanContent) IS an
|
|
* explicitly-named target, so the gate must NOT ask — and the review
|
|
* must actually consume the pasted content.
|
|
*
|
|
* Cost note: 4 sequential PTY runs (~3-5 min each) in the gate lane, up
|
|
* from 1 pre-bypass. Selected only when plan-ceo/eng/design or the runner
|
|
* change (see 'plan-mode-no-op' in touchfiles.ts).
|
|
*/
|
|
|
|
import { describe, test, expect } from 'bun:test';
|
|
import { runPlanSkillObservation } from './helpers/claude-pty-runner';
|
|
|
|
const shouldRun = !!process.env.EVALS && process.env.EVALS_TIER === 'gate';
|
|
const describeE2E = shouldRun ? describe : describe.skip;
|
|
|
|
const PLAN_MODE_REMINDER =
|
|
'Plan mode is active. The user indicated that they do not want you to execute yet';
|
|
|
|
// Distinctive token proves the pasted target was consumed by the review —
|
|
// not just that no question fired. Nonsense-unique so it can't appear by
|
|
// coincidence in skill output.
|
|
const SEED_TOKEN = 'ZephyrLedgerWidget';
|
|
const NAMED_TARGET_SEED = `
|
|
# Plan: ${SEED_TOKEN} settings panel
|
|
|
|
## Scope
|
|
Add a ${SEED_TOKEN} settings panel with a single toggle that enables
|
|
weekly export emails. One new component, one route, one test file.
|
|
|
|
## Files
|
|
- src/components/${SEED_TOKEN}.tsx (new)
|
|
- src/routes/settings.tsx (add panel)
|
|
- test/${SEED_TOKEN}.test.tsx (new)
|
|
`;
|
|
|
|
describeE2E('plan-mode-info no-op outside plan mode (gate regression)', () => {
|
|
for (const skillName of ['plan-ceo-review', 'plan-eng-review', 'plan-design-review'] as const) {
|
|
test(`${skillName} reaches a terminal outcome outside plan mode`, async () => {
|
|
const obs = await runPlanSkillObservation({
|
|
skillName,
|
|
inPlanMode: false,
|
|
timeoutMs: 300_000,
|
|
// eng/design: force the prose-fallback path. The unconditional
|
|
// gate-must-ask assert below pins the render shape the detector
|
|
// anchors on, and only the --disallowedTools prose fallback makes
|
|
// that shape CONTRACTUAL ("use exactly this shape" in the template);
|
|
// native AskUserQuestion could render terse option labels that a
|
|
// correct run would fail on (red-team finding).
|
|
...(skillName === 'plan-ceo-review'
|
|
? {}
|
|
: { extraArgs: ['--disallowedTools', 'AskUserQuestion'] }),
|
|
});
|
|
|
|
if (obs.outcome === 'silent_write' || obs.outcome === 'exited' || obs.outcome === 'timeout') {
|
|
throw new Error(
|
|
`plan-mode no-op regression FAILED (${skillName}): outcome=${obs.outcome}\n` +
|
|
`summary: ${obs.summary}\n` +
|
|
`elapsed: ${obs.elapsedMs}ms\n` +
|
|
`--- evidence (last 2KB visible) ---\n${obs.evidence}`,
|
|
);
|
|
}
|
|
expect(['asked', 'plan_ready']).toContain(obs.outcome);
|
|
|
|
// Negative regression: the rendered output must NOT echo the plan-mode
|
|
// distinctive reminder phrase. If it does, the plan-mode preamble
|
|
// section is leaking outside plan mode.
|
|
expect(obs.evidence).not.toContain(PLAN_MODE_REMINDER);
|
|
|
|
if (skillName !== 'plan-ceo-review') {
|
|
// Scope-gate bypass must not misfire: no auto-select announcement
|
|
// outside plan mode.
|
|
expect(obs.scopeGateAutoSelectObserved ?? false).toBe(false);
|
|
// UNCONDITIONAL: outside plan mode with no named target, the gate is
|
|
// a hard STOP before any tool call, so the gate question must have
|
|
// rendered no matter which terminal outcome fired. Gating this on
|
|
// outcome === 'asked' would let a silent-bypass run that reaches
|
|
// plan_ready (isPlanReadyVisible also matches common prose) sail
|
|
// through — the exact regression this test exists to catch.
|
|
expect(obs.scopeGateQuestionObserved ?? false).toBe(true);
|
|
}
|
|
}, 360_000);
|
|
}
|
|
|
|
// Named-target exception (outside plan mode): a pasted draft IS an
|
|
// explicitly-named target, so the scope gate must NOT ask — and the
|
|
// review must consume the pasted content (seed token visible in the
|
|
// review output), proving the target was used rather than the question
|
|
// merely skipped. Also the over-trigger guard for the tightened
|
|
// "explicit-only" exception wording.
|
|
test('plan-eng-review skips the scope gate for an explicitly-pasted target', async () => {
|
|
const obs = await runPlanSkillObservation({
|
|
skillName: 'plan-eng-review',
|
|
inPlanMode: false,
|
|
initialPlanContent: NAMED_TARGET_SEED,
|
|
trackTokens: [SEED_TOKEN],
|
|
timeoutMs: 300_000,
|
|
});
|
|
|
|
if (
|
|
obs.outcome === 'wrote_findings_before_asking' ||
|
|
obs.outcome === 'silent_write' ||
|
|
obs.outcome === 'exited' ||
|
|
obs.outcome === 'timeout'
|
|
) {
|
|
throw new Error(
|
|
`named-target no-op FAILED: outcome=${obs.outcome}\n` +
|
|
`summary: ${obs.summary}\n` +
|
|
`elapsed: ${obs.elapsedMs}ms\n` +
|
|
`--- evidence (last 2KB visible) ---\n${obs.evidence}`,
|
|
);
|
|
}
|
|
expect(['asked', 'plan_ready']).toContain(obs.outcome);
|
|
expect(obs.evidence).not.toContain(PLAN_MODE_REMINDER);
|
|
|
|
// The pasted doc is the named target: gate question must not render,
|
|
// no plan-mode announcement either (we are NOT in plan mode).
|
|
expect(obs.scopeGateQuestionObserved ?? false).toBe(false);
|
|
expect(obs.scopeGateAutoSelectObserved ?? false).toBe(false);
|
|
|
|
// Target consumption via high-water token tracking over the CUMULATIVE
|
|
// buffer — the 2KB evidence tail is lossy and the plan-file fallback is
|
|
// unreachable outside plan mode (extractPlanFilePath only matches
|
|
// plan-mode save renders).
|
|
expect(obs.tokensObserved?.[SEED_TOKEN] ?? false).toBe(true);
|
|
}, 360_000);
|
|
});
|