[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6BSVFF1CZAR7AW8WBH4TK9J":3,"skill-same:01M6BSVFF1CZAR7AW8WBH4TK9J":66},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":33,"linkOnly":34,"markdownContent":35,"markdownBody":36,"frontMatter":37,"refName":38,"commitSha":39,"skillDirectory":40,"forksCount":41,"repositoryHtmlUrl":42,"repositoryHomepageUrl":6,"ownerHtmlUrl":43,"ownerBlogUrl":44,"ownerDescription":6,"updatedAt":45,"install":46,"resources":59},"01M6BSVFF1CZAR7AW8WBH4TK9J","codex-review",null,"Run a code review using the local OpenAI `codex` CLI. Use ONLY when the user names codex explicitly — \"codex review\", \"run codex on this branch\", \"what does codex say\". Do NOT use for generic review requests like \"review my changes\" or \"look for bugs\"; those belong to \u002Fcode-review, which stays local. This one uploads the diff to OpenAI.","gurdiga\u002Fdotfiles","gurdiga","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F53922?v=4","01M6BSV943TWTZSMZK5D2TRR1F","Vlad GURDIGA","01M6BSV97XQEB3TMGQ0FFS6FM1",3,"MIT","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcodex-review\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000007","代码审查","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32],"review","codex","codex review","code review","cli","run","code","code-review","openai","2026-11-10 10:12:30",false,"---\nname: codex-review\ndescription: Run a code review using the local OpenAI `codex` CLI. Use ONLY when the user names codex explicitly — \"codex review\", \"run codex on this branch\", \"what does codex say\". Do NOT use for generic review requests like \"review my changes\" or \"look for bugs\"; those belong to \u002Fcode-review, which stays local. This one uploads the diff to OpenAI.\n---\n\n# Codex Review\n\n`codex` is OpenAI's review CLI, run locally against a diff in the current repo. **It uploads that diff to OpenAI**, so it runs only when the user has asked for it by name. Offering it is fine where the model-routing rules call for a second opinion, but an offer isn't a request: say that it uploads the diff, and wait for the user to name Codex. Never reach for it as a general-purpose reviewer, and never run it on a repo the user hasn't pointed you at.\n\nIf `codex` isn't on PATH, say so and stop — don't offer to install it.\n\n## Pick the scope\n\nThree mutually exclusive modes:\n\n| Flag | Reviews |\n|------|---------|\n| `--base \u003CBRANCH>` | This branch against a base — the usual case |\n| `--uncommitted` | Staged, unstaged, and untracked changes |\n| `--commit \u003CSHA>` | One commit's changes |\n\nMatch the flag to what the user actually wants reviewed. \"Review my branch\" is `--base`; \"review my uncommitted changes\" is `--uncommitted`. There is no staged-only mode: `--uncommitted` also uploads unstaged and untracked files, so when the user asks for only what's staged, say that and ask before running it.\n\n## Every scope flag conflicts with `[PROMPT]`\n\nThe usage line reads `codex review [OPTIONS] [PROMPT]`, which suggests you can pass custom instructions alongside a scope. You can't — all three scope flags are mutually exclusive with the positional prompt:\n\n```\nerror: the argument '--base \u003CBRANCH>' cannot be used with '[PROMPT]'\n```\n\nVerified through codex-cli 0.147.0. Take the scope flag: reviewing the correct diff matters more than steering the review. A bare `codex review \"instructions\"` is the only way to get custom instructions, and it gives up scope control to do it.\n\n## Get the base branch right\n\nTwo ways to end up reviewing already-merged code, both of which waste a full run:\n\n1. **Local branch refs go stale.** Use the remote ref (`origin\u002Fmain`), never the local one (`main`) — local branches only move when you pull them.\n2. **The remote ref is stale too, until you fetch.** Run `git fetch origin` first, every time.\n\nSquash merges sharpen this: when a branch is cut from another branch that later gets squash-merged, the merged work reappears as unmerged commits in the diff, and the review fills up with findings about code that already landed.\n\n```bash\ngit fetch origin\nBASE=$(git symbolic-ref --short refs\u002Fremotes\u002Forigin\u002FHEAD 2>\u002Fdev\u002Fnull || echo origin\u002Fmain)\n```\n\nIf `origin\u002FHEAD` isn't set — common in clones that never ran `git remote set-head` — that fallback is a guess. Confirm the real default branch with the user rather than reviewing against the wrong one.\n\n## Run it in the background\n\nIt streams to stdout and runs for several minutes — well past the default Bash timeout — so run it with `run_in_background: true` rather than waiting on it. You'll be re-invoked when it exits, so there's no need to poll.\n\nRedirect both streams into a file in your scratchpad directory so the transcript survives the wait:\n\n```bash\ncodex review --base \"$BASE\" > \"$OUT\" 2>&1\n```\n\nSet `OUT` to a path under your scratchpad directory. Tell the user it's running before you go quiet — a silent multi-minute gap reads as a hang.\n\n## Triage before relaying\n\nCodex output is a starting point, not a verdict to pass through. Three things to check:\n\n- **The `exec` blocks in the transcript are codex's own tool calls, not findings.** The actual verdict is at the end of the file.\n- **Check the head it reviewed against current HEAD.** It reports what it reviewed. If commits landed after the run started, findings may already be fixed — this has produced a confident P1 about code that had been corrected two commits earlier.\n- **Verify load-bearing claims against the tree yourself.** Observed failure mode: a type narrowing (`Array.isArray` on a union) reported as dead code when the TypeScript compiler actually requires it. Codex reads the diff, not the whole type graph.\n\nRelay what survives that check, and say which findings you verified versus which you're passing along unchecked.\n","# Codex Review\n\n`codex` is OpenAI's review CLI, run locally against a diff in the current repo. **It uploads that diff to OpenAI**, so it runs only when the user has asked for it by name. Offering it is fine where the model-routing rules call for a second opinion, but an offer isn't a request: say that it uploads the diff, and wait for the user to name Codex. Never reach for it as a general-purpose reviewer, and never run it on a repo the user hasn't pointed you at.\n\nIf `codex` isn't on PATH, say so and stop — don't offer to install it.\n\n## Pick the scope\n\nThree mutually exclusive modes:\n\n| Flag | Reviews |\n|------|---------|\n| `--base \u003CBRANCH>` | This branch against a base — the usual case |\n| `--uncommitted` | Staged, unstaged, and untracked changes |\n| `--commit \u003CSHA>` | One commit's changes |\n\nMatch the flag to what the user actually wants reviewed. \"Review my branch\" is `--base`; \"review my uncommitted changes\" is `--uncommitted`. There is no staged-only mode: `--uncommitted` also uploads unstaged and untracked files, so when the user asks for only what's staged, say that and ask before running it.\n\n## Every scope flag conflicts with `[PROMPT]`\n\nThe usage line reads `codex review [OPTIONS] [PROMPT]`, which suggests you can pass custom instructions alongside a scope. You can't — all three scope flags are mutually exclusive with the positional prompt:\n\n```\nerror: the argument '--base \u003CBRANCH>' cannot be used with '[PROMPT]'\n```\n\nVerified through codex-cli 0.147.0. Take the scope flag: reviewing the correct diff matters more than steering the review. A bare `codex review \"instructions\"` is the only way to get custom instructions, and it gives up scope control to do it.\n\n## Get the base branch right\n\nTwo ways to end up reviewing already-merged code, both of which waste a full run:\n\n1. **Local branch refs go stale.** Use the remote ref (`origin\u002Fmain`), never the local one (`main`) — local branches only move when you pull them.\n2. **The remote ref is stale too, until you fetch.** Run `git fetch origin` first, every time.\n\nSquash merges sharpen this: when a branch is cut from another branch that later gets squash-merged, the merged work reappears as unmerged commits in the diff, and the review fills up with findings about code that already landed.\n\n```bash\ngit fetch origin\nBASE=$(git symbolic-ref --short refs\u002Fremotes\u002Forigin\u002FHEAD 2>\u002Fdev\u002Fnull || echo origin\u002Fmain)\n```\n\nIf `origin\u002FHEAD` isn't set — common in clones that never ran `git remote set-head` — that fallback is a guess. Confirm the real default branch with the user rather than reviewing against the wrong one.\n\n## Run it in the background\n\nIt streams to stdout and runs for several minutes — well past the default Bash timeout — so run it with `run_in_background: true` rather than waiting on it. You'll be re-invoked when it exits, so there's no need to poll.\n\nRedirect both streams into a file in your scratchpad directory so the transcript survives the wait:\n\n```bash\ncodex review --base \"$BASE\" > \"$OUT\" 2>&1\n```\n\nSet `OUT` to a path under your scratchpad directory. Tell the user it's running before you go quiet — a silent multi-minute gap reads as a hang.\n\n## Triage before relaying\n\nCodex output is a starting point, not a verdict to pass through. Three things to check:\n\n- **The `exec` blocks in the transcript are codex's own tool calls, not findings.** The actual verdict is at the end of the file.\n- **Check the head it reviewed against current HEAD.** It reports what it reviewed. If commits landed after the run started, findings may already be fixed — this has produced a confident P1 about code that had been corrected two commits earlier.\n- **Verify load-bearing claims against the tree yourself.** Observed failure mode: a type narrowing (`Array.isArray` on a union) reported as dead code when the TypeScript compiler actually requires it. Codex reads the diff, not the whole type graph.\n\nRelay what survives that check, and say which findings you verified versus which you're passing along unchecked.\n",{"name":5,"description":7},"main","f2d447b5b931f057c19dc6770b9fcf651d3101e3",".claude\u002Fskills\u002Fcodex-review",0,"https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles","https:\u002F\u002Fgithub.com\u002Fgurdiga","https:\u002F\u002Fgurdiga.com","2026-10-10 19:25:43",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add gurdiga\u002Fdotfiles --skill codex-review",{"tool":52,"command":53},"bunx","bunx skills add gurdiga\u002Fdotfiles --skill codex-review",{"tool":55,"command":56},"pnpm","pnpm dlx skills add gurdiga\u002Fdotfiles --skill codex-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcodex-review\u002FSKILL.md",true,[60],{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"big.txt","REFERENCE","参考文档",300000,"https:\u002F\u002Fgithub.com\u002Fcanyonroad\u002Fagentsh\u002Fblob\u002Fmain\u002Finternal\u002Fskillcheck\u002Ftestdata\u002Fskills\u002Foversized\u002Fbig.txt",[67,80,90,97,107,124],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":76,"publishedAt":33},"01M6BSVWMJQ6KYFASV0SKX1F1C","write-skill","Create or improve a Claude Code Agent Skill (SKILL.md). Use when the user asks to capture a workflow as a skill, create a reusable skill, or turn something we just did together into a skill.","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fwrite-skill\u002FSKILL.md",[73],{"categoryNo":74,"categoryName":75,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[77,30,78,79],"write skill","create","workflow",{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":33},"01M6BSVV49EKDRGZQ4KHGZWE1S","ssh-controlmaster","Use a persistent SSH connection via ControlMaster when running multiple commands against the same remote host, to avoid repeated authentication overhead. Use when about to run several ssh commands in a row against one host.","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fssh-controlmaster\u002FSKILL.md",[],[87,88,89,29],"ssh controlmaster","ssh","authentication",{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":33},"01M6BSVSDTV9EY5E9E792AWV86","reflect","Analyze recent session transcripts for the current project and suggest CLAUDE.md improvements. Invoke manually once a month or after a significant stretch of work.","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freflect\u002FSKILL.md",[],[92],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":33},"01M6BSVQFP801VXHHWC1DETG0N","grill-me","Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions \"grill me\".","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fgrill-me\u002FSKILL.md",[],[104,105,106],"grill me","test","design",{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":33},"01M6BSVN9CX51YXKGPSQFJP3EJ","git-message-style","Vlad's personal conventions for writing commit messages and PR descriptions — subject shape, body rules, vocabulary, and PR body structure. Use when writing a commit message or a PR description in one of Vlad's personal repos (not a team repo with its own convention).","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fgit-message-style\u002FSKILL.md",[],[114,115,116,117,118,119,120,121,122,123],"git message style","git","writing","messages","commit message","pr description","pr","repos","commit","personal",{"skillNo":125,"name":126,"nameZh":6,"description":127,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":128,"categories":129,"keywords":130,"publishedAt":33},"01M6BSVJQRMR2V4VSFE5CRY4RJ","gem-security-update","Fix Dependabot security alerts for Ruby gems. Use when the user shares a GitHub Dependabot URL, asks to fix gem vulnerabilities, or mentions a gem CVE needing a version bump.","https:\u002F\u002Fgithub.com\u002Fgurdiga\u002Fdotfiles\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fgem-security-update\u002FSKILL.md",[],[131,132,133,134,135,136,137,138],"security","gem security update","update","cve","github","vulnerabilities","security alerts","dependabot"]