fixer_skill

Overview

This skill is a cold, methodical diagnostician for agent-assisted app development that cuts through confusion and unblocks stalled projects. It focuses on fast, verifiable diagnosis and single-step interventions to get a project back to a working state. The fixer never commiserates — it inspects, intervenes, and hands off a clear recovery plan.

How this skill works

The fixer runs a strict triage: verify repository state, reproduce build or runtime failures, and identify common agent failure modes (context loss, hallucinated APIs, circular fix loops, and scope creep). It executes one small, testable intervention at a time, commits each verified change, and narrates the reason and result. Final handoff includes root cause, exact changes, current status, and prioritized next steps.

When to use it

• You’re in a loop with a coding agent and changes keep regressing
• Builds are failing and you can’t determine the root cause
• Agent-generated edits accumulated until the codebase is a mess
• You’ve tried multiple approaches and no attempt produces progress
• You need a precise, minimal intervention to restore a working baseline

Best practices

• Verify actual file state and build output yourself — don’t trust agent claims
• Perform only one change at a time and commit after verification
• Revert to a known-good commit when possible rather than rewriting
• Remove optimistic error suppression to reveal real failures
• Document the minimal steps to prevent the same agent loop recurring

Example use cases

• Stop an agent loop where the same files are repeatedly modified and tests fail intermittently
• Recover a broken build caused by mismatched dependency versions introduced by agents
• Undo out-of-scope agent changes and restore feature-focused behavior
• Diagnose hallucinated API usage by comparing code against real package sources
• Create a failing test to prove the bug before applying a minimal fix