What problem does it solve? AI agents repeat the same mistakes because nobody systematically reviews their execution history and converts failure patterns into durable instruction improvements. This Skill turns an agent's recent issue history into reviewed, evidence-backed coaching proposals instead of ad-hoc rewrites. ## Core Features & Use Cases - Trajectory-Based Failure Clustering: Pulls an agent's recent issues and comments via the Todero API, then clusters failures into a named taxonomy (verifier-miss, stale-context, late-escalation, and more) with at least two quoted evidence tuples per cluster. - Minimal Diff Proposals: Routes each cluster to the right surface — AGENTS.md updates capped at +20% growth, reusable skills capped at 15KB, or tool descriptions capped at 500 characters — and drafts the actual diffs and SKILL.md files. - Gated Application: Publishes proposals on a reflection issue with a request_confirmation interaction showing the diff; changes are applied only in a separate follow-up run after acceptance, never hot-swapped. - Use Case: A manager notices an engineer agent repeatedly marks issues done without re-running failing tests. Run this Skill against that agent to get a replay-benchmarked proposal adding one verification rule to its AGENTS.md, gated behind human approval. ## Quick Start Ask the agent to reflect on target agent <agentId> over the last 10 closed issues and produce a coaching proposal with diffs for review.
