grinchenkoedu
OrganizationClaude Code skills for everyday development: plan, implement, review, resolve PR comments, verify, and write a repo's CLAUDE.md. Repository-agnostic, economical by design.
Categories
Indexed Skills (9)
implement
Build a task step by step in the current session — from a plan file, a markdown brief, or a sentence. Works through ordered steps, ticking each one off in the task file as it lands, so an interrupted run resumes exactly where it stopped instead of starting over.
plan
Turn a request — a sentence you type, or a markdown brief — into a grounded plan you can hand to /gku:implement. Works out what is really being asked (bug, feature, question, data fix), checks it against the actual code and data, and writes an ordered plan with acceptance criteria. Plans only; writes no production code.
pr-resolve
Work through the review comments on your own pull request, from any directory — pass a URL and it resolves the repository, reusing a local clone or making one. Every finding gets a verdict before any code changes (agree, disagree with evidence, or ask you), then fixes land one commit per finding, get pushed, and each thread gets a reply. Never fixes blindly; reviewers and bots are sometimes wrong.
pr-review
Deep review of someone else's pull request in an isolated worktree — hunts the failures that automated review misses (broken error paths, infinite loops, races, blast radius, claim-vs-code drift), deduplicates against the bot's comments, and gives a short verdict in chat. Read-only; never edits, pushes or posts.
review
Review your own changes before you push or open a pull request — a severity-rated list of what to fix, checked against the repository's own conventions, a lint pass over the changed files, its tests, a security pass over the well-known flaws (injection, escaping, login and permission checks, CSRF, secrets, files, outbound requests), and the failure modes that green tests miss. Ends by naming the next step. Review-only; never edits, commits or pushes.
verify
Check that a change actually works — run the project's tests, then drive the real thing (the command, the page, the function) and confirm the effect, and confirm its guards hold (login, permission, CSRF token, escaping, paths) with one minimal local probe each. Reports a clear verdict, and says honestly what could not be checked and why, instead of quietly skipping it.
fix
Fix what is wrong — the findings from a review still in this conversation, a review report file, or a symptom you describe in a sentence. A described symptom gets investigated first, the way /gku:plan investigates, until the cause is proven; then it acts instead of writing a plan. Re-checks every finding against the current code before touching it, lands the smallest change per finding, and re-runs the tests. Blockers and warnings by default; nits only when you ask.
init
Write or refresh this repository's CLAUDE.md — detected build, test and run commands plus the conventions and security rules for its family (Moodle plugin, PHP app, CMS, Python). Reads an existing file first and reports it as already fine rather than churning it; regenerating replaces only a marked block, so hand-written sections survive. Also leaves AGENTS.md as a stub pointing at CLAUDE.md.
pr
Open a pull request for the current branch, or update the one already linked to it — with a title and description written from the actual diff and the repository's own template. Checks first that the branch reads as one coherent change, and asks before opening a pull request that is really two. Never merges, never force-pushes, never commits on your behalf.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.