learn-from-mistake

Root cause analysis identifies the triggering context or thought leading to agent errors. This Skill systematically investigates conversations, pinpoints the mistake's origin, designs preventive fixes like prompt updates or validation hooks, then reproduces the error to verify the prevention works before deploying changes.

Yes. The Skill enforces a phase-driven workflow—INVESTIGATE, PREVENT, FIX—with phase-state tracking and git history restoration. It automates root cause investigation, applies targeted fixes to agent configurations, and blocks premature fixes until prevention is verified through reproduction testing.

Reproduction testing attempts to trigger the original mistake after applying preventive measures. By recreating the exact conditions and confirming the agent no longer fails, you validate that the fix prevents recurrence rather than just masking the symptom.

Yes. Git enables history restoration for content recovery and workflow tracking, while jq handles structured data parsing during root cause investigation. Both are required dependencies for the Skill's phase-driven analysis pipeline.

Manual review is time-consuming and often fails to prevent recurrence. This Skill automates the learning loop—systematically investigating conversations, applying targeted preventive fixes, and validating them through reproduction—reducing rework and ensuring consistent error prevention.

Yes. The Skill is designed for post-mortem review of recurring mistakes like incorrect file placement. It identifies ambiguous instructions as root cause, creates validation hooks like pre-write checks to block violations, then tests whether the agent still attempts the error.