Convert evaluator failure reports into fix tasks and re-run evaluations.
npx skills add https://github.com/AhmedElhadarey/Gilfoyle --skill loop-fixer
Please help me install this Agent Skill. Skill: loop-fixer Source: https://github.com/AhmedElhadarey/Gilfoyle/tree/main/skills/loop-fixer Command: npx skills add https://github.com/AhmedElhadarey/Gilfoyle --skill loop-fixer
Evaluates and fixes the loop-back scenario in an automated Evaluate-Loop by translating evaluator failure reports into concrete fix tasks, applying them, and triggering a re-evaluation to resume progress.
Instruct the agent to fix the latest evaluation failure by generating and applying fixes, then re-run the evaluator.
High-intent search queries and answers about installing and using this skill.
To automatically fix failed evaluations, the system parses evaluator failure reports, translates them into concrete fix tasks within plan.md, applies the changes, commits them, and triggers a re-evaluation cycle to resume progress.
Automated evaluation fix loops enforce guardrails by limiting the process to a maximum of three fix cycles and ensuring all applied fixes strictly align with the original evaluator instructions to prevent runaway loops.
You can generate fix tasks from evaluator failure reports by parsing the output from loop-plan-evaluator or loop-execution-evaluator and appending actionable items under a dedicated Fix Phase in your plan.md file.
Version control integration tracks applied fixes by updating task statuses, recording commit SHAs for state tracking, and committing the modified plan.md file to preserve progress across multiple re-evaluation cycles.
If an automated evaluation loop fails to progress after applying fixes, you should verify that the generated tasks align with evaluator instructions and check if the maximum limit of three fix cycles has been reached.
Yes, you can use loop-plan-evaluator failure reports to automatically update plan.md by converting the reports into specific fix tasks, applying the corrections, and committing the changes to reflect current progress.
cexllOrchestrate multi-backend AI code tasks with pluggable backends and file references.
Ibrahim-3dParses failure reports and executes bounded fix cycles for evaluations.
zxdxjtuIteratively generate and evaluate fixes until feature gaps converge.
paultyngDiagnose failing GitHub Actions runs, apply fixes, and re-trigger CI.
