What problem does it solve? When an AI agent implements a ticket, the reasoning behind its choices — which review findings it accepted or declined, and which interpretation of an ambiguous ticket it picked — normally dies with the conversation. This Skill captures that judgement in a committed review-points.md file so human reviewers can see what was decided and how confident the agent was. ## Core Features & Use Cases - Guided implementation flow: Implements a ticket, commits the feature alone, then runs /code-review over that commit without --fix so accept/decline decisions are preserved. - Structured decision record: Writes review-points.md at the repo root with Fixed, Ignored, and Assumptions sections, each entry anchored to a file:line and each assumption carrying a confidence score in [0, 1]. - Commit trailer conventions: Emits Review-Points, Implements, and Claude-Session trailers so downstream scripts can attribute implementation and review commits and compute phase costs. - Use Case: A developer types /implement-ticket 37; the agent builds the feature, reviews its own work, commits both the code and a review-points.md explaining which findings it declined and which ticket readings it was unsure about. ## Quick Start Ask the agent to run /implement-ticket with a ticket number, for example by typing /implement-ticket 37 in the repository where the work should happen.
