aurorascharff/agent-friction-skill

A pair of skills for AI agents to record developer-experience friction during agentic runs. It includes an active skill, friction-log, which the agent runs on demand to narrate and write a structured markdown log in real time, and a passive skill, friction-report, which scans the conversation at session end and drafts a review if it detects problems. Who uses it: agent developers, harnesses, and AI agents that need to capture and surface build failures, misleading errors, documentation gaps, or workflow pain points. Key features: - Active logging: real-time narration, chronological tool timeline, header, prompt, summary, severity-coded log, and action items split into Docs, Framework, and DX/Research, plus post-review skill feedback. - Passive reporting: end-of-session scan that posts a draft for human review and opens a review URL; silent if no friction. - Viewer: browser-based renderer for collapsible, severity-coded logs; logs stay client-side via URL fragment. - Installable via npx; includes documentation and examples in the repo.