timothy/ersatztv
1
0
Fork 0
Code Issues 51 Pull Requests 5 Actions Packages Projects Releases 9 Wiki Activity

fix(916): the cross-family review step follows the lenses instead of running beside them, and fix commits are a sha range

Browse Source 
One .NET slot's review round ran two worktree-isolated reviewers at once — the
correctness lens and, on a rubric change, the Codex fallback — and took swap
from 6.8 GB to 10.8 GB in three minutes on the 16 GB host; three slots reached
load 82. review() now awaits the lenses, then the Codex runner, then its
fallback. The finisher's fix attribution is the sha range the fixer's report
head advances (head_sha is required on every report), replacing a line-set
difference over free text that listed all eleven #563 commits as fixes. The
harness gains a case that records how many lenses were still in flight when
the cross-family agents started (must be zero); moving the fallback back into
the parallel batch reddens it in both scripts. The mechanics page and the
standing prompt state the measured cap: one .NET-building slot at a time.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015QqCpYFsKgnAnx6jVwrKiV
This commit is contained in:
timothy
2026-09-05 12:07:17 +02:00
co-authored by Claude Fable 5.1
parent ea1519871d
commit f7a0da4051
5 changed files with 77 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
web/scripts/orchestration-workflow-loop.test.mjs
+31 -7
View File
@@ -38,16 +38,30 @@ const finding = (severity) => ({ severity, file: 'f', summary: `${severity} find
// Every stub lens returns the same findings, so a finding counts once per lens (two lenses → 2).
// reviewRounds: array indexed by round-1 of {findings, lensesNull?}; landOverrides: extra fields on the finisher report;
// postRebase: findings returned by the review after a patch-changing rebase, or 'lensesNull'; fixerNull: the fixer agent dies.
function run(script, args, { reviewRounds, landOverrides = {}, postRebase = null, fixerNull = false, fixerDone = true, fixerCommits = null }) {
function run(script, args, { reviewRounds, landOverrides = {}, postRebase = null, fixerNull = false, fixerDone = true, fixerNoCommit = false }) {
const calls = [];
let round = 0;
let fixCount = 0;
let lensesInFlight = 0;
const xfamilyStartedWithLensesInFlight = [];
const agent = async (prompt, opts) => {
const label = (opts && opts.label) || '';
calls.push({ label, prompt, opts });
if (label.startsWith('impl:') || label.startsWith('fix:#')) return { done: true, summary: '', verified: '', left: '', commits: IMPL_COMMITS };
if (label.startsWith('impl:') || label.startsWith('fix:#')) return { done: true, summary: '', verified: '', left: '', commits: IMPL_COMMITS, head_sha: 'aaaaaaa' };
if (label.startsWith('review:codex:')) {
xfamilyStartedWithLensesInFlight.push(lensesInFlight);
return { verdict: 'merge', findings: [], ran: false };
}
if (label.startsWith('review:fallback:')) {
xfamilyStartedWithLensesInFlight.push(lensesInFlight);
const spec = reviewRounds[round - 1] || { findings: [] };
return { verdict: 'merge', findings: spec.findings };
}
if (label.startsWith('review:')) {
const spec = reviewRounds[round - 1] || { findings: [] };
lensesInFlight += 1;
await new Promise((resolve) => setTimeout(resolve, 5));
lensesInFlight -= 1;
if (spec.lensesNull) return null;
return { verdict: 'merge', findings: spec.findings };
}
@@ -55,7 +69,7 @@ function run(script, args, { reviewRounds, landOverrides = {}, postRebase = null
fixCount += 1;
if (fixerNull) return null;
if (!fixerDone) return { done: false, summary: 'stopped', verified: '', left: 'blocked', commits: IMPL_COMMITS };
return { done: true, summary: '', verified: '', left: '', commits: fixerCommits !== null ? fixerCommits : `f${fixCount}xxxxx fix round ${fixCount}\n${IMPL_COMMITS}` };
return { done: true, summary: '', verified: '', left: '', commits: fixerNoCommit ? IMPL_COMMITS : `f${fixCount}xxxxx fix round ${fixCount}\n${IMPL_COMMITS}`, head_sha: fixerNoCommit ? 'aaaaaaa' : `f${fixCount}xxxxx` };
}
if (label.startsWith('land:')) return { done: true, summary: '', verified: '', left: '', commits: '', pr_url: 'http://pr/1', head_sha: 'deadbeef', patch_changed: false, ...landOverrides };
throw new Error(`unexpected agent label ${label}`);
@@ -67,7 +81,7 @@ function run(script, args, { reviewRounds, landOverrides = {}, postRebase = null
}
return Promise.all(thunks.map((t) => t().catch(() => null)));
};
return script(args, agent, parallel, () => {}, () => {}).then((result) => ({ result, calls, fixCount }));
return script(args, agent, parallel, () => {}, () => {}).then((result) => ({ result, calls, fixCount, xfamilyStartedWithLensesInFlight }));
}
const cases = [
@@ -94,10 +108,10 @@ describe.each(cases)('%s review loop', (name, args) => {
expect(result.error).toBeUndefined();
expect(fixCount).toBe(1);
const hist = finisherPrompt(calls);
expect(hist).toContain('round 1: 2 lens(es); 0 blocking, 2 should-fix, 0 nit; answered by fix commit(s): f1xxxxx fix round 1');
expect(hist).toContain('round 1: 2 lens(es); 0 blocking, 2 should-fix, 0 nit; answered by the fix commit(s) in git log --oneline aaaaaaa..f1xxxxx');
expect(hist).toContain('round 2: 2 lens(es); 0 blocking, 0 should-fix, 0 nit; clean — loop ended');
expect(hist).not.toContain(IMPL_COMMITS);
expect(hist).toContain('what each of these fix commits changed');
expect(hist).toContain('for exactly the commits git log --oneline lists in these ranges: aaaaaaa..f1xxxxx');
});
it('a blocking finding that survives two fix rounds stops before the push', async () => {
@@ -147,7 +161,7 @@ describe.each(cases)('%s review loop', (name, args) => {
});
it('findings answered without a new commit are described as such, never as a clean round one', async () => {
const { calls } = await run(script, args, { reviewRounds: [{ findings: [finding('should-fix')] }, { findings: [] }], fixerCommits: IMPL_COMMITS });
const { calls } = await run(script, args, { reviewRounds: [{ findings: [finding('should-fix')] }, { findings: [] }], fixerNoCommit: true });
const p = finisherPrompt(calls);
expect(p).toContain('answered without a new commit');
expect(p).toContain('every finding was answered without a new commit');
@@ -159,6 +173,16 @@ describe.each(cases)('%s review loop', (name, args) => {
expect(result.error).toMatch(/^the post-rebase review round produced no reviews/);
});
it('on a rubric-class change the cross-family step starts only after both lenses have returned', async () => {
const rubricArgs = { ...args, risk: 'rubric' };
const { result, calls, xfamilyStartedWithLensesInFlight } = await run(script, rubricArgs, { reviewRounds: [{ findings: [] }] });
expect(result.error).toBeUndefined();
expect(calls.filter((c) => c.label.startsWith('review:codex:')).length).toBe(1);
expect(calls.filter((c) => c.label.startsWith('review:fallback:')).length).toBe(1);
expect(xfamilyStartedWithLensesInFlight).toEqual([0, 0]);
expect(finisherPrompt(calls)).toContain('substituted a cold same-family review-only agent');
});
it('a should-fix on the pushed head after a patch-changing rebase is an error', async () => {
const { result } = await run(script, args, {
reviewRounds: [{ findings: [] }],