What is agent-friction-skill?

This is a skill, a structured set of instructions, built for AI coding agents. When an AI agent works on a development task with this skill active, it logs friction in real time: moments where it gets stuck, makes a mistake, needs better documentation, or runs into a tool that does not work the way it should. The result is a structured friction log that captures the agent's development experience as it happens, rather than relying on the agent to recall problems afterward from memory. The skill instructs the agent to record a chronological list of every tool call it makes, fold in any messages the user sends mid task, and then produce a formatted log in markdown once the task is done. The log includes a header with the date, model, and agent tool used, the original prompt along with any follow-up clarifying exchanges, a chronological tool call timeline, a summary of the overall experience and the biggest pain point, and action items split into three categories: documentation gaps, framework or code issues, and open research questions. Each log entry is marked with a color coded severity indicator so the most serious problems stand out at a glance. The intended audience is developers and teams who want to understand, in a structured way, where AI agents struggle during real coding workflows, so they can improve documentation, fix confusing error messages, or adjust tooling defaults. A finished log can be pasted into a companion web viewer at agent-friction-skill.vercel.app, which renders it as a collapsible, severity coded layout with the biggest pain points surfaced first. The viewer keeps everything in the browser, encoding the log into a shareable URL fragment rather than uploading anything to a server, so a log can be shared with a teammate as a link without exposing the underlying task details elsewhere.