What triggers the Loop Fixer agent?

It is triggered automatically whenever a plan or execution evaluation returns a FAIL verdict, or manually via commands like 'fix issues' or 'address evaluation failures'.

Can I use it to add new features?

No, the agent is governed by a Scope Guard that restricts its actions to only addressing the specific issues flagged in the evaluation report.

What happens if an issue cannot be fixed automatically?

The agent includes a guardrail that limits it to 3 fix cycles. If the issue persists after the third attempt, it escalates to 'BLOCKED' status and prompts the user for intervention.

How does it track its progress?

The agent updates plan.md with specific fix tasks and maintains a detailed state in metadata.json, including cycle counts, applied fixes, and associated commit SHAs.

Does it automatically commit changes?

Yes, following the execution protocol, it marks each task in the plan and creates a git commit after each successful fix implementation.

Loop Fixer Agent

Name: Loop Fixer Agent
Author: AhmedElhadarey

byAhmedElhadarey

•

Productivity & Workflow

Automates the error-correction phase of the development loop by converting evaluation failures into actionable tasks and executing fixes.

The Loop Fixer is a specialized Claude Code skill designed to manage the 'FIX' phase of an iterative development cycle. When a plan or execution fails an evaluation check, this agent parses the failure report, injects structured fix tasks into the project's plan, and systematically implements changes with automated commit tracking. It ensures high-quality outcomes by maintaining a strict loop-back mechanism—fixing issues and requesting re-evaluation until standards are met—while providing safety guardrails to prevent infinite loops and scope creep through a centralized metadata tracking system.

Key Features

01Automated Failure Parsing: Translates evaluator feedback directly into actionable plan.md tasks.

022 GitHub stars

03Metadata State Management: Maintains a persistent record of fix cycles and status in metadata.json.

04Verification & Handoff: Conducts local self-checks before automatically triggering the original evaluator for re-validation.

05Systematic Execution: Implements fixes following a strict protocol with progress tracking and individual commits.

06Loop Guardrails: Prevents excessive iterations by enforcing a three-cycle limit before escalating to the user.

Use Cases

01Refining implementation details when an execution doesn't fully align with project specifications.

02Rectifying build failures or quality issues identified during automated AI code reviews.

03Managing complex, multi-step bug fixes that require systematic tracking and re-validation.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ahmedelhadarey/gilfoyle loop-fixer

For use in Claude.ai and ChatGPT

Download Skill