Classify failing verification outputs and apply bounded repairs to agent runs.
npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-repair
Please help me install this Agent Skill. Skill: loop-repair Source: https://github.com/SollanSystems/loop-engineer/tree/main/skills/loop-repair Command: npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-repair
This Skill helps recover agent runs that have failed verification, avoiding endless thrashing by forcing a disciplined, evidence-based repair process.
Use the loop-repair skill to inspect the failing verification output, classify the failure mode, apply the smallest bounded fix, and rerun the same check.
High-intent search queries and answers about installing and using this skill.
To fix failing tests without thrashing, apply a bounded repair process that classifies the failure, makes the smallest possible diff, and reruns the same check. This enforces a disciplined, evidence-based recovery for failed agent runs.
Handling broken verification outcomes requires classifying failures into deterministic errors, flaky checks, or spec gaps, then applying targeted minimal diffs. This structured approach confirms whether the repair actually improved verification.
To stop an agent from weakening tests, use a safety framework that refuses test manipulation and caps retry attempts. It escalates unresolvable failures to replan, revert, approval, or termination instead of allowing scope creep.
Yes, a bounded repair workflow supports flaky checks and regressions by distinguishing them from deterministic failures. It applies one targeted repair and records whether the change was productive before attempting independent re-verification.
You should escalate a failing loop repair when retry attempts hit the cap or the failure mode indicates an environmental blocker. The process escalates to replan, revert, approval, or termination to avoid unproductive scope creep.
derHakenTrace data flow and apply a four-phase process to fix root causes.
MohamedAbdallah-14Diagnose asset_validate() failures and map them to repair primitives within a retry budget.
timurgaleevAggregate test outcomes, lint warnings, and dead code signals into a single health score.
MoonLadderStudiosOrchestrate GitHub pull request resolution with bounded retries and structured artifacts.
