Claude Skills by dev-geon

Apply the findings of a code review to the working tree: fix every Failed finding, Warnings only when safe or requested, and verify each fix resolves the finding's evidence — resolved findings are reported as Pass. Use this skill whenever a review report exists (from the veriloop skill or any structured review) and the user asks to act on it — "apply the review findings", "fix what the review found", "리뷰 지적사항 반영해줘", "리뷰에서 나온 문제 고쳐줘" — and as the fix step inside the run-review-loop skill.

Use this skill to produce the written specification for a concrete piece of upcoming development work when none exists yet. The spec gates in the veriloop and run-review-loop skills route here whenever they find no referenced spec; and users can invoke it directly in any language: "write a spec for this first", "draft the spec before we start", "명세서 작성해줘", "명세부터 만들자", "이 작업 명세로 정리해줘". It analyzes the target codebase first — house rules, conventions, workflow, and the code the goal touches — t...

Initialize Veriloop in the current repository by creating optional per-role model configuration, strict blind review defaults, and the findings ledger. Use when the user asks to set up, configure, initialize, or choose developer, fixer, reviewer, or gate models for the review loop. Do not use merely to run a review or start a loop.

Run a bounded, specification-grounded develop-review-fix loop on a repository until executable acceptance checks pass and independent code review has no failed findings. Use when the user asks to implement or repair work in a review loop, iterate until clean, run the Veriloop workflow, or invokes $run-review-loop. Requires an explicit goal, a confirmed written specification, executable acceptance checks, and scope bounds. Do not use for a one-off review or for open-ended make-it-better requests.

Review concrete recent code changes before commit, merge, or deployment: a diff, uncommitted work, branch, PR, or just-finished feature written by a user, coding agent, IDE assistant, subagent, or automated task. Trigger on review, check, verify, inspect, final pass, 검토, 리뷰, 확인, or 훑어보기, including requests focused only on regression, performance, metered-service cloud cost, readability, or repository conventions. Ground the review in a confirmed written specification; when none exists, route ...