FE-1254: Repair failed slice verification

[kostandinang]

Why

A recoverable worker-authored verification failure halted the run immediately instead of giving the worker a bounded chance to repair it.

What

Add a finite verify-to-repair loop in the same slice workspace. Repair state is crash-durable and replayable across serial and parallel execution, and only a passing verification cycle can integrate.

How to test

• Full suite: 2664 passed, 2 pre-existing skips
• npm run check
• npm run build

[kostandinang]

• FE-1254: Repair failed slice verification #370 👈 (View in Graphite)
• FE-1241: Add isolated brownfield comparison cases #362 : 1 other dependent PR (#369 )
• next

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

High Risk
Touches core executor orchestration, Petri topology, run metadata shape, and crash-recovery paths for serial and parallel slice execution—incorrect repair authority could cause duplicate effects or unsafe worker prompts.

Overview
Failed slice verification no longer halts immediately when repair budget remains. The executor runs a finite repair ladder (up to three repair cycles, three agent/verify attempts per stage) in the same slice workspace, with cycle-qualified Petri places/transitions and sliceRepairHistory replacing flat sliceAttemptHistory.

Repair authority is canonical and crash-safe: bounded verify diagnostics are written to durable repair context (digest-checked, fail-closed validation), durable-file primitives fsync parent chains for journal/marking/context writes, and serial vs parallel paths persist pending repair in run.json or marking.json before materialization. Recovery does not replay the source verifier; the agent runner gets cycle, validated repair context (untrusted stderr/stdout only), and durable result writes.

Observers/RPC expose activeSliceCycle and phases like repair_pending; integration only proceeds from a passed verdict in the active cycle. Tests cover multi-repair drives, corruption cases, and parallel sibling behavior.

^{Reviewed by Cursor Bugbot for commit abb40e4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[graphite-app]

Merge activity

• Jul 23, 1:19 AM UTC: This pull request can not be added to the Graphite merge queue. Please try rebasing and resubmitting to merge when ready.
• Jul 23, 1:19 AM UTC: Graphite disabled "merge when ready" on this PR due to: a merge conflict with the target branch; resolve the conflict and try again..

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f399cbb. Configure here.}