fix(ingest): verify repair outcomes and reject dangling join targets (#292)

One ingest integration hiccup no longer discards a whole source:

- Replace the duplicated gate-repair and textual-resolver loops with one
  shared constrained-repair loop whose success criterion is re-running
  the failed check (verify), not whether the agent edited files. Verify
  failures feed the retry prompt; maxAttempts is 2.
- Let the resolver declare a conflicting patch redundant: a verified
  no-change resolution is accepted as subsumed instead of failing the
  source (duplicate wiki-page creation from parallel work units).
- Carry per-source validation errors through validateWuTouchedSources
  into gate messages and work-unit failure reasons instead of
  discarding them.
- Move join-neighbor expansion into the shared validation path so
  work-unit validation and integration gates check the same set.
- Reject joins whose target resolves to no source, at sl_write time and
  in the gates, attributed to the declaring source. Resolution mirrors
  the Python engine exactly (case-sensitive name within the
  connection), with a case-mismatch hint for the writing agent.
This commit is contained in:
Andrey Avtomonov 2026-06-11 14:39:51 +02:00 committed by GitHub
parent 00cdf2de90
commit a278d2f7d0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 1119 additions and 705 deletions

View file

@ -221,6 +221,7 @@ describe('integrateWorkUnitPatch', () => {
touchedPaths: ['wiki/global/a.md'],
});
await writeFile(join(configDir, 'wiki/global/a.md'), 'accepted\nproposal\n', 'utf-8');
await expect(context.verify(['wiki/global/a.md'])).resolves.toEqual({ ok: true });
return {
status: 'repaired' as const,
attempts: 1,
@ -336,6 +337,7 @@ describe('integrateWorkUnitPatch', () => {
touchedPaths: ['wiki/global/a.md'],
});
await writeFile(join(configDir, 'wiki/global/a.md'), 'repaired semantic ref\n', 'utf-8');
await expect(context.verify(['wiki/global/a.md'])).resolves.toEqual({ ok: true });
return {
status: 'repaired' as const,
attempts: 1,
@ -402,71 +404,56 @@ describe('integrateWorkUnitPatch', () => {
await expect(readFile(join(configDir, 'wiki/global/a.md'), 'utf-8')).resolves.toBe('old\n');
});
it('repairs a semantic gate failure after a textual conflict is resolved', async () => {
const { homeDir, configDir, git } = await makeRepo();
await mkdir(join(configDir, 'wiki/global'), { recursive: true });
await writeFile(join(configDir, 'wiki/global/a.md'), 'base\n', 'utf-8');
await git.commitFiles(['wiki/global/a.md'], 'base page', 'System User', 'system@example.com');
const conflictBase = await git.revParseHead();
it('accepts a redundant duplicate-creation patch as subsumed without committing', async () => {
// Regression: two work units each emitted a creation patch for the same
// wiki page. The second creation patch conflicts with the page already in
// the tree; the resolver verifies a no-change resolution and the source
// must not fail.
const { homeDir, configDir, git, baseSha } = await makeRepo();
await writeFile(join(configDir, 'wiki/global/b.md'), 'page from the first work unit\n', 'utf-8');
await git.commitFiles(['wiki/global/b.md'], 'first creation', 'System User', 'system@example.com');
const acceptedHead = await git.revParseHead();
await writeFile(join(configDir, 'wiki/global/a.md'), 'accepted\n', 'utf-8');
await git.commitFiles(['wiki/global/a.md'], 'accepted edit', 'System User', 'system@example.com');
const childDir = join(homeDir, 'child-conflict-repair');
await git.addWorktree(childDir, 'child-conflict-repair', conflictBase);
const childDir = join(homeDir, 'child-duplicate');
await git.addWorktree(childDir, 'child-duplicate', baseSha);
const childGit = git.forWorktree(childDir);
await writeFile(join(childDir, 'wiki/global/a.md'), 'proposal\n', 'utf-8');
await childGit.commitFiles(['wiki/global/a.md'], 'proposal edit', 'System User', 'system@example.com');
const patchPath = join(homeDir, 'proposal-repair.patch');
await childGit.writeBinaryNoRenamePatch(conflictBase, 'HEAD', patchPath);
await writeFile(join(childDir, 'wiki/global/b.md'), 'duplicate page from the second work unit\n', 'utf-8');
await childGit.commitFiles(['wiki/global/b.md'], 'second creation', 'System User', 'system@example.com');
const patchPath = join(homeDir, 'duplicate-creation.patch');
await childGit.writeBinaryNoRenamePatch(baseSha, 'HEAD', patchPath);
const trace = new FileIngestTraceWriter({
tracePath: join(homeDir, '.ktx/ingest-traces/job-resolver-repair/trace.jsonl'),
jobId: 'job-resolver-repair',
tracePath: join(homeDir, '.ktx/ingest-traces/job-subsumed/trace.jsonl'),
jobId: 'job-subsumed',
connectionId: 'warehouse',
sourceKey: 'metabase',
sourceKey: 'notion',
level: 'trace',
});
// Gate fails on the resolver's merged tree, then passes after the repair edit.
const validateAppliedTree = vi
.fn()
.mockRejectedValueOnce(
new Error('final artifact gates failed:\narr-definition: unknown sl_refs entity mart_arr_daily.arr_dollars'),
)
.mockResolvedValueOnce(undefined);
const repairGateFailure = vi.fn(async (context: { unitKey: string; touchedPaths: string[] }) => {
expect(context).toMatchObject({ unitKey: 'wu-conflict-repair', touchedPaths: ['wiki/global/a.md'] });
await writeFile(join(configDir, 'wiki/global/a.md'), 'accepted\nproposal repaired\n', 'utf-8');
return { status: 'repaired' as const, attempts: 1, changedPaths: ['wiki/global/a.md'] };
});
const result = await integrateWorkUnitPatch({
unitKey: 'wu-conflict-repair',
unitKey: 'wu-duplicate',
patchPath,
integrationGit: git,
trace,
author: { name: 'System User', email: 'system@example.com' },
slDisallowed: false,
allowedTargetConnectionIds: new Set(['warehouse']),
validateAppliedTree,
resolveTextualConflict: vi.fn(async () => {
await writeFile(join(configDir, 'wiki/global/a.md'), 'accepted\nproposal\n', 'utf-8');
return { status: 'repaired' as const, attempts: 1, changedPaths: ['wiki/global/a.md'] };
validateAppliedTree: vi.fn(async () => {}),
resolveTextualConflict: vi.fn(async (context) => {
await expect(context.verify([])).resolves.toEqual({ ok: true });
return { status: 'repaired' as const, attempts: 1, changedPaths: [] };
}),
repairGateFailure,
});
expect(result).toMatchObject({
status: 'accepted',
touchedPaths: ['wiki/global/a.md'],
textualResolution: { status: 'repaired' },
gateRepair: { status: 'repaired', attempts: 1, changedPaths: ['wiki/global/a.md'] },
touchedPaths: [],
textualResolution: { status: 'repaired', attempts: 1, changedPaths: [] },
});
expect(validateAppliedTree).toHaveBeenCalledTimes(2);
expect(repairGateFailure).toHaveBeenCalledOnce();
await expect(readFile(join(configDir, 'wiki/global/a.md'), 'utf-8')).resolves.toBe('accepted\nproposal repaired\n');
await expect(readFile(trace.tracePath, 'utf-8')).resolves.toContain('patch_accepted_after_textual_resolution');
expect(await git.revParseHead()).toBe(acceptedHead);
await expect(readFile(join(configDir, 'wiki/global/b.md'), 'utf-8')).resolves.toBe(
'page from the first work unit\n',
);
await expect(readFile(trace.tracePath, 'utf-8')).resolves.toContain('patch_subsumed_after_textual_resolution');
});
});