What problem does it solve? When working with a coding agent, failures and repetitive manual work accumulate without a systematic way to turn them into durable improvements. This Skill scans what actually happened — one failed run's artifacts or a window of session logs — and recommends which AI-layer primitive (rule, skill, hook, subagent, MCP, automation) would prevent the failure or encode the repeated work. ## Core Features & Use Cases - Reactive run analysis: Point it at one run's artifacts (plan, implementation report, RCA, PR body, diff) plus a symptom you noticed, and it identifies the smallest durable change that would have prevented the problem. - Proactive log scanning: Point it at weeks of session logs (Claude Code, Codex, PI, etc.) and a theme you care about, and it surfaces recurring patterns worth encoding, ranked by frequency and value. - Agent-agnostic primitive mapping: It first reads the agent's own capability docs, then maps each finding to the best-fit extension point in the house style of your existing rules and skills. - Use Case: After a coding agent ignored your conventions in a PR, run the scan on that run's artifacts with the symptom described; receive a self-contained HTML report proposing, for example, a PreToolUse hook or a new CLAUDE.md rule that would have caught it. ## Quick Start Ask the agent to run the opportunity scan on the artifacts from your last failed run, describing the symptom you noticed, and have it produce the HTML report of recommended changes.
