from ai-launchpad-marketplace124
Analyze a session's skill invocations to surface performance issues and recommend or apply improvements using the skill-creator workflow.
Skill Retro processes session transcripts to analyze how installed skills were invoked and whether they behaved as intended. It automates preprocessing of logs, spawns analysis agents to identify findings, and guides remediation using skill-creator sub-agents so skills can be improved or fixed.
Use Skill Retro at the end of an agent session when you want a structured retrospective: if you suspect skills misfired, returned incorrect output, or you want to systematically improve skill documentation and scripts based on real usage.
Designed for Claude-style skill workflows and sub-agent orchestration but applicable to any agent environment supporting script execution and sub-agent spawning (Claude Code, Sonnet-style subagents, Cursor).
Skill Retro analyzes Claude Code session JSONL logs to evaluate skill invocation performance and recommend improvements via skill-creator. The single script (preprocess.mjs) is well-written Node.js but was skipped by the runner due to .mjs extension. SKILL.md is detailed with clear multi-step process, error handling guidance, and good progressive disclosure. No security concerns — reads only local session files with no network calls.
node (script is .mjs, not .py/.sh, so runner skipped it)Well-designed skill for a specific Claude Code workflow. The .mjs script is clean and production-quality. The SKILL.md process is thorough (6 steps) with good error handling and user confirmation gates. Minor shell injection surface in preprocess.mjs path args but low risk since it's run locally by the agent. The skill is narrowly useful — only relevant for Claude Code users doing skill retrospectives.
