What problem does it solve? Agent sessions accumulate friction — misunderstandings, backtracks, and wasted wall-clock waits — that disappears when the session ends. This Skill mines captured conversation logs or the live session for those friction signals and converts them into durable rules, defaults, and conventions in the repo's local packs so the next run needs less human steering. ## Core Features & Use Cases - Friction signal detection: Identifies misunderstandings, suboptimal agent actions, and long wall-time waits, then converts each into a concrete fix such as a default, convention, footgun note, or parallelization rule. - Measured wall-time analysis: Computes real idle-wait versus compute time from the log's own timestamps and tool_use/tool_result pairs rather than estimating. - Dual-window review: Reads both the newest 24 hours of captures and the oldest 24 hours still on the conversation-logs branch for a hindsight re-read. - Provenance comments: Posts a short summarized comment on the originating PR or issue for each rule that lands, never pasting raw transcript. - Use Case: After a week of agent sessions captured on the conversation-logs branch, run a retrospective that turns repeated user corrections into standing conventions and recurring user questions into proactive self-checks. ## Quick Start Ask the agent to extract lessons from the latest captures on the conversation-logs branch and land them as a pull request for review.
