一键导入
review-fix
Use when you need to fetch PR review comments, fix issues in code, commit, push, and reply on GitHub. Invoke with /review-fix or /review-fix <PR-number>.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when you need to fetch PR review comments, fix issues in code, commit, push, and reply on GitHub. Invoke with /review-fix or /review-fix <PR-number>.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Przygotowuje gotowy tekst completion condition dla komendy /goal w Claude Code (v2.1.139+), dostosowany do implementacji OpenSpec change. Bierze nazwę changu z openspec/changes/<name>/ i produkuje sformułowanie, które ewaluator (mały model, Haiku) potrafi sprawdzić z samego transkryptu rozmowy. Triggeruje WYŁĄCZNIE na komendę /prepare-openspec-goal — nie auto-triggeruj na keywords typu 'goal', 'openspec', 'change'. Używaj gdy user pisze /prepare-openspec-goal <change-name> i chce sformułować warunek zakończenia dla autonomicznej pętli /goal implementującej openspec change.
Use when you want to automatically iterate between Claude and Copilot PR reviews — fix comments, push, wait for Copilot re-review, repeat until stable. Invoke with /review-loop <PR-number> <openspec-change-name>.
LinkedIn post creator and content strategist for personal brand building. Use when the user wants to write a LinkedIn post, draft LinkedIn content, brainstorm post ideas for LinkedIn, optimize an existing LinkedIn post, create hooks, or build a content calendar for LinkedIn. Also trigger when the user mentions 'post na LinkedIn', 'LI', 'personal brand', 'ghostwriter', or wants to share a professional insight, case study, behind-the-scenes story, or opinion piece on LinkedIn. Trigger on Polish requests like 'napisz post na LinkedIn', 'pomysl na post', 'hook', 'zasieg na LinkedIn', 'seria postow'. Do NOT trigger for blog posts, website copy, email drafts, PowerPoint presentations, or social media graphics — only for LinkedIn text content. Can work standalone or be invoked by the marketing agent.
Generuje 1-2 stronicowy, agent-ready PRD dla MVP budowanego w 7 dni (Faza 1 playbooka PLSoft 7-Day MVP / AI Startup Builders Week 3). Bierze insighty z Week 1 (Problem-Solution Fit) i Week 2 (GTM/ICP/Value Proposition + LP) — dostarczone w dowolnym układzie plików (jeden dokument, osobne icp/positioning/ gtm-plan/landing-brief, folder, notatki) plus brand — i sprowadza je do jednego dokumentu prd.md, który agent (Claude Code lub Lovable/v0) rozumie i pod który generuje kod. Triggeruj na komendę /prd ORAZ gdy founder-uczestnik programu chce zrobić PRD pod MVP, spisać scope na 7 dni, przerobić output W1/W2 na dokument dla agenta, albo pyta "jak napisać PRD dla mojego MVP", "spnij to w PRD", "scope na MVP". Triggeruj też gdy ktoś chce ograniczyć/przyciąć zakres (scope) MVP do jednego core flow i spisać to w dokument pod agenta, oraz gdy osoba nietechniczna buduje aplikację w Lovable/v0 i potrzebuje PRD/spisanego zakresu pod swoje MVP. Nie triggeruj na pełne PRD dla zewnętrznego klienta z fixed-price (to software
Przygotowuje gotowy tekst completion condition dla komendy /goal w Claude Code (v2.1.139+). Bierze luźny opis zadania i produkuje sformułowanie, które ewaluator (mały model) potrafi sprawdzić z samego transkryptu rozmowy. Triggeruje WYŁĄCZNIE na komendę /prepare-goal — nie auto-triggeruj na keywords typu 'goal', 'cel sesji', 'cel zadania'. Używaj gdy user pisze /prepare-goal <opis> lub /prepare-goal i chce sformułować warunek zakończenia dla autonomicznej pętli /goal.
Financial advisor and fractional CFO for business analysis. Use when: analyzing cash flow, runway, burn rate, revenue, costs, profitability, budget planning, financial forecasting, cost optimization, unit economics, break-even analysis, MRR/ARR/churn/CAC/LTV (SaaS metrics), P&L review, financial reports, or comparing financial scenarios. Also use for investment decisions, 'can we afford X' questions, expense analysis, or when the user mentions money, spending, earnings, or financial health of the company. Dyrektor finansowy, analiza finansowa, przeplywy pieniezne, rentownosc, budzet, koszty, przychody, prognoza finansowa, optymalizacja kosztowa, runway, ile mamy kasy, ile wydalismy, ile zarabiamy.
| name | review-fix |
| description | Use when you need to fetch PR review comments, fix issues in code, commit, push, and reply on GitHub. Invoke with /review-fix or /review-fix <PR-number>. |
Fetch PR review comments, fix valid issues, commit, push, and reply on GitHub — all in one command.
Input: Optionally specify a PR number (e.g., /review-fix 55). If omitted, detect from current branch via gh pr view.
Detect PR and repo
Detect owner/repo from the git remote:
gh repo view --json nameWithOwner --jq .nameWithOwner
If a PR number was provided as argument, use it. Otherwise detect from current branch:
gh pr view --json number,url,headRefName
Note: gh pr view without a number only works if the current branch has an associated PR. If it fails (e.g., you're on main), prompt the user for the PR number — do NOT guess.
Store the PR branch name (headRefName) — you'll need it for pushing later.
Fetch review comments
Fetch ALL review comments (the GitHub API paginates at 30 per page by default — you MUST use --paginate or you will miss comments on long-running PRs):
gh api --paginate "repos/{owner}/{repo}/pulls/{pr}/comments?per_page=100" > pr-comments.json
--paginate with gh api automatically follows Link: rel="next" headers and concatenates pages into a single JSON array. per_page=100 is the max — keeps the round-trips low.
Sanity check: after fetching, log total: <c.length>. If the PR thread shows more comments than that in the GitHub UI, the fetch missed pages — re-run with --paginate.
Important: Do NOT pipe through jq — it may not be installed. Parse the JSON output directly with Node (require('./pr-comments.json')).
Filtering logic:
git config user.login or match against GitHub username from gh api userin_reply_to_id: comments with in_reply_to_id are replies, those without are top-levelTriage each comment
For each comment, read the referenced file and surrounding code, then classify:
For OUTDATED and DISAGREE, do NOT change code — just prepare a reply.
Apply fixes
For each FIXABLE comment:
suggestion code block is provided in the comment, verify it's correct before applyingVerify
Run pnpm typecheck after all fixes are applied. If it fails, fix the issue before committing.
Commit and push
Stage only files that were changed by fixes. Create a single commit:
fix: address PR review feedback
- <brief description of each fix>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Push to the PR's branch (use headRefName from step 1):
git push origin HEAD:<headRefName>
Reply to each comment
CRITICAL: Send replies ONE AT A TIME, sequentially. Do NOT send multiple replies in parallel — if one fails, the rest get cancelled and you have to redo them.
Use the GitHub API to reply in-thread:
gh api repos/{owner}/{repo}/pulls/{pr}/comments/{comment_id}/replies -X POST -f body="<reply>"
Do NOT pipe the output through jq or other tools that may not be installed. Redirect to /dev/null or use > /dev/null 2>&1 if you don't need the response.
Reply guidelines:
Summary
Print a table:
| # | Comment | Action | Reply |
|---|---------|--------|-------|
| 1 | timer leak in postWithRetry | FIXED | Fixed in abc1234 |
| 2 | outdated fire-and-forget | OUTDATED | Refactored since review |