swatkinson-toolkit

🐊 Claudecodile fixer. Addresses the reviewer's open inline comments (applying their suggested fixes) in the worktree, re-verifies, and reports. Spawned by the /claudecodile-review loop. Never commits or pushes.

🐊 Claudecodile code reviewer. Reviews a PR's diff, posts P#-tagged inline comments WITH suggested fixes, resolves the review threads it confirms fixed, and maintains one rating comment (0–5, where 5/5 = no P0/P1 and every in-scope P2/P3 fixed). Spawned each review round by the /claudecodile-review loop. Posts/resolves comments only — never edits code, commits, or pushes.

handle-it bug investigator. Spawned for unclear bugs (symptom / error log / perf regression with no known root cause). Runs /diagnose in the worktree — reproduce → root-cause → fix → regression-test — and reports. Never commits or pushes.

handle-it implementer for features and clear bugs. Spawned by the handle-it orchestrator to implement one Linear issue in an isolated worktree via /ship, verify, and report. Never commits or pushes.

handle-it test runner. Runs the automatable items from a PR's Test plan (bun run check/test, build, focused suites) in the worktree and reports pass/fail per item. Does NOT edit the PR, tick checkboxes, commit, or push.

🐊 Claudecodile in-house code review. Runs an iterate-until-5/5 review⇄fix loop on a PR — an Opus reviewer posts P#-tagged inline comments with suggested fixes and maintains one rating comment; a Sonnet fixer applies them; loops until 5/5 (no P0/P1). Use when the user invokes /claudecodile-review on a PR, or when another skill (e.g. /handle-it) delegates its review phase here.

End-to-end orchestrator that takes a Linear issue (or a freeform feature/project) and drives it through planning → implementation → draft PR → an in-house review⇄fix loop (rated to 5/5) → conflict-free CI + preview → an auto-tester → your manual review (PR stays a draft) → ready-for-review on your approval, with a live status table and Linear blocking awareness. Use when the user invokes /handle-it, or asks to take an issue/feature all the way to a review-ready PR.

Live dashboard of your active Linear issues. Gathers your In Progress / In Review / blocked issues, fans out Haiku scanner subagents to read each issue's handle-it status block (or derive status from worktree + PR ground truth when there's no block), and compiles ONE consolidated glance-table — then keeps it refreshed on a loop. Use when the user invokes /issue-watcher, or asks for a status board / overview of all their in-flight issues.

Retrospective self-evaluation of a skill you just ran. Reconstructs the skill's intended goals, reconstructs what ACTUALLY happened from the session transcript (errors, retries, stalls, hangs, user interventions, deviations), scores the run against a 6-dimension rubric, root-causes each friction point, and proposes concrete edits to the skill file plus ways the operator could run it better. Use at the end of a skill run, when the user invokes /skill-evaluate, or asks "how well did that skill go", "how could this skill be improved", "evaluate that run", "what slowed that down".