[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79RKWW6S7Z6FQD0HYYZQJVC":3,"skill-same:01M79RKWW6S7Z6FQD0HYYZQJVC":67},{"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":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":31,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M79RKWW6S7Z6FQD0HYYZQJVC","code-review",null,"Reviews a pull request or diff in the mergeCraft repository and produces typed, evidence-backed findings with a blocking verdict. Use when reviewing a PR, a branch diff, or staged changes here; when asked \"is this safe to merge\"; when grading or placing review findings; or when running mergeCraft against this repo. Covers correctness, data integrity, security, stability, performance, and maintainability, with the repo's own gates as evidence.","alexhawat\u002FmergeCraft","alexhawat","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1254687?v=4","01M79RKRMA3JJ8JZFXZ1YBDT0D","Alex Hawat","01M79RKRNE7VHRR1W30AGW74C2",12,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: code-review\ndescription: >-\n  Reviews a pull request or diff in the mergeCraft repository and produces typed,\n  evidence-backed findings with a blocking verdict. Use when reviewing a PR, a\n  branch diff, or staged changes here; when asked \"is this safe to merge\";\n  when grading or placing review findings; or when running mergeCraft against\n  this repo. Covers correctness, data integrity, security, stability,\n  performance, and maintainability, with the repo's own gates as evidence.\nlicense: Apache-2.0\nmetadata:\n  mergecraft.doctrine-version: \"1\"\n---\n\n# Reviewing code in mergeCraft\n\nA mergeCraft review is a falsifiable audit of one diff: typed findings anchored to\nevidence, graded on three axes, filtered hard, then one terminal verdict. A good\nreview is not the one with the most comments — it is the one whose blocking\nconcerns survive verification and whose silence means the author can merge.\n\nThis file is **self-sufficient**. A reviewer that reads only `SKILL.md` can run a\ncorrect, complete review of this repository. The `references\u002F` files add depth;\nthey are never prerequisites. Nothing in this skill tree links outside\n`.github\u002Fskills\u002Fcode-review\u002F`.\n\n> **For humans:** `docs\u002FREVIEW-CHECKS.md` is the consumer-facing check catalog;\n> `docs\u002FREVIEW-DOCTRINE.md` is the rationale and rejected alternatives. This\n> directory is the agent-facing instruction bundle mergeCraft injects at review time.\n\n## The four rules\n\nThese four are absolute. Everything else in this file is judgement. Each rule\nstates its reason so you can generalise to cases this file never anticipated.\n\n1. **No finding without evidence you can point at.** A claim must cite a file and\n   line, a diff hunk, tool output, or a retrieved artifact. Without an anchor it\n   is a question — ask it in the summary or drop it. Unfalsifiable comments teach\n   authors to ignore this reviewer.\n\n2. **Never re-raise a withdrawn finding.** The section\n   `## Withdrawn review findings (known non-issues)` in the learnings file is\n   binding. Re-litigating a refuted finding is worse than missing a real one,\n   because it proves this reviewer does not remember prior pushback.\n\n3. **Every blocking concern in the summary exists as a findings row.** Prose in\n   the summary is read by no gate. If it blocks merge, it must appear in the\n   structured `findings` array with severity, category, and evidence.\n\n4. **Exactly one terminal verdict.** Call `approve` or `request_changes` once,\n   after findings survive verification. The verdict follows from the findings —\n   never the other way around.\n\n## Review progress\n\nCopy this into your response and check items off as you go.\n\n```\n- [ ] 1. Scope    — full diff read end-to-end; coverage checklist built\n- [ ] 2. Evidence — repo gates + analyzers run; withdrawn findings read\n- [ ] 3. Triage   — what kind of change is this; trivial or not\n- [ ] 4. Lenses   — load-bearing questions named, or none\n- [ ] 5. Findings — drafted, evidence attached, graded\n- [ ] 6. Filter   — drop list applied; blocking findings verified\n- [ ] 7. Verdict  — one terminal call\n```\n\n## 1. Scope\n\nRead the **complete diff** before anything else. Use the diff table of contents and\nfile line ranges as a coverage checklist. A reviewer that samples the diff writes\nfindings about the part it sampled and stays silent about the rest — and silence\nreads as approval.\n\nThen pull **targeted** context: the touched functions, their callers, the tests\nthat exercise them, the interfaces they cross. Not the whole repository. More\ncontext measurably makes frontier models worse at this task; stay on the blast\nradius.\n\n## 2. Evidence before opinion\n\nRun the repo's own gates first and quote them. \"`make lint` fails on this file\"\nis a finding nobody argues; \"consider sorting this\" is one everybody does.\n\nThree outcomes mean **no signal** — report them as skipped, never as findings:\n\n- the tool is not in your toolset at all;\n- the tool returns `ran: false`;\n- a gate status is `unavailable`, `declared-but-cannot-run`, or `timed_out`.\n\nOnly `failed` is a finding. Never substitute your own linter, formatter, or\ninterpreter to fill a gap — a gate run under the wrong toolchain version\nmanufactures findings.\n\nRead `## Withdrawn review findings (known non-issues)` **now**, not at drafting\ntime. It changes what you bother investigating.\n\n## 3. Triage\n\nName what kind of change this is: which domain, which seams, which external\ncontracts, which user-visible surfaces.\n\n**Genuinely trivial — skip to the verdict:** doc typo, whitespace-only,\ncomment-only, lockfile or generated-code regeneration, mechanical rename whose\nonly effect is import paths, low-risk dependency patch bump.\n\n**Looks trivial but is not — small diff, large blast radius:**\n\n- one-line changes to SQL, regex, auth, billing, permissions, or signature\n  verification;\n- a flipped feature-flag default, retry\u002Ftimeout constant, or money\u002Ftax constant;\n- a changed HTTP method, redirect URL, response code, or comparison operator;\n- a renamed public API surface or new direct dependency;\n- a \"typo fix\" in user-facing copy that changes meaning;\n- a semantic one-liner buried in a formatting-only diff.\n\nRead the shape, not the line count.\n\n## 4. Lenses\n\nName the load-bearing questions you cannot resolve yourself. A question is\nload-bearing only when its answer could change the verdict, and falsifiable only\nwhen evidence could settle it. \"Another look for confidence\" is neither.\n\nTwo framings:\n\n- **Themed lenses** across the whole diff (correctness, security, performance).\n- **Subsystem lenses** for high-stakes domains (auth, billing, webhooks, schema\n  migration). For those domains, lead with the subsystem lens — \"the billing lens\"\n  primes double-charge and refund-race failure modes that a generic correctness\n  pass misses.\n\nRun `mergecraft lens list` for the registry — each entry carries its id,\ntriggers, rubric, and required evidence. Do not work from a remembered list.\n\n## 5. Findings\n\nEvery finding carries six fields before you place it:\n\n| Field | Value |\n| --- | --- |\n| Category | one of the six taxonomy categories (see references\u002Fgrading.md) |\n| Severity | Critical · Major · Minor · Trivial |\n| Effort | Quick win · Heavy lift · Low value |\n| Confidence | certain · likely · possible |\n| Evidence | exact file and line, tool output, or retrieved artifact |\n| Why | one sentence of consequence, not of mechanism |\n\n**Placement is mechanical:** `Trivial` **or** `Low value` → a bullet in the\nbody's Nitpicks list. Everything else → an inline comment at its line.\n\nThe cost of a false positive is not the minute spent reading it — it is the team\nlearning to skip every comment this reviewer ever leaves. Grade honestly;\ninflating a nit to Major to justify an inline anchor is the habit that makes a\nreviewer ignorable.\n\n**Category is a sweep, not a menu.** A PR that writes persistent state with no\nData Integrity & Atomicity finding is worth one more look before concluding there\nwas nothing there.\n\n**Hunt for non-anchored concerns too:** deletion plans for code this diff\nshadows; rollout sequencing and in-flight state; coverage the diff implies but\ndoes not add; scope questions only a human can answer. On substantial PRs at\nleast one exists — if you cannot think of any, the bar is too high, not the PR\ntoo clean.\n\nSee [references\u002Fgrading.md](references\u002Fgrading.md) for axes, placement, collateral,\nand the verification loop.\n\n## 6. What gets dropped\n\nThe drop list is most of what keeps a review readable:\n\n- Praise and style preferences the repo does not enforce.\n- Speculative and unverified claims.\n- Findings whose root cause predates this diff.\n- Anything already withdrawn.\n- Anything not actionable.\n- Bloat-shaped fixes — defensive checks for impossible cases, abstractions used\n  once, comments restating obvious code, tests asserting tautologies,\n  just-in-case guards. The bar is sound **and** correct **and** elegant; a change\n  that improves one by degrading another makes the codebase worse.\n\n**Silence is a result.** If nothing survives this filter, say so explicitly and\napprove. An empty review from a reviewer that looked hard is a valid outcome and a\ngood one. Manufacturing a finding to look diligent is the failure mode this\nsection exists to prevent. Route silence through §7 — do not stop after §6 without\na terminal verdict.\n\n## 7. Verdict\n\nThe terminal call is exactly one of `approve` or `request_changes`, with a\n`summary` and a `findings` array. Match the opening callout tier to what the\nauthor should actually do next.\n\n- **`approve`** — no blocking findings survive verification; say so plainly.\n- **`request_changes`** — at least one Critical or Major finding blocks merge;\n  every blocking concern in the summary must have a matching findings row.\n\nThe server **rejects** `request_changes` with an empty `findings` array — that is\nenforced, not a style preference. `approve` over a verifier-confirmed Critical or\nMajor blocker is also rejected.\n\nWhen silence is the outcome: call `approve` with an empty `findings` array and a\nsummary that states what you read, what gates you ran, and that nothing actionable\nsurvived the filter.\n\nSee [references\u002Ftools.md](references\u002Ftools.md) for the call in each environment.\n\n## Tools\n\nThree capability tiers share this one skill. Use the tier that matches your\nruntime; degrade steps — never skip them silently.\n\n**Tier 1 — mergeCraft MCP (full):** `mcp__mergecraft__checkout_pr` establishes\nscope and returns `diffPath`. In the same turn call\n`mcp__mergecraft__run_static_checks` and `mcp__mergecraft__run_analyzers` with\nchanged paths; when CI failed on the head, call\n`mcp__mergecraft__analyze_ci_failures`. Before publishing, hand Critical\u002FMajor\nagent findings to `mcp__mergecraft__verify_agent_findings` and record each\nverdict with `mcp__mergecraft__record_finding_verdict`. Finish with exactly one\n`mcp__mergecraft__submit_review_verdict` (`verdict`, `summary`, `findings`).\n\n**Tier 2 — Copilot \u002F hosted code review (no MCP):** Read the PR diff from the\nhost UI. Run repo gates locally when you have shell access (`make lint`,\n`make typecheck`, scoped pytest). Draft findings as inline review comments with\nthe triage tag; post an approve or request-changes review through the host. You\ncannot call `submit_review_verdict` — make the summary and inline comments\nself-contained so a human can act without mergeCraft's structured export.\n\n**Tier 3 — bare agent (file access only):** Read the diff artifact and changed\nfiles directly. Note which gates you could not run. Produce findings with the six\nfields above; end with an explicit approve or request-changes recommendation and\nthe same discipline on evidence and silence.\n\n## References\n\n- [references\u002Fchecks.md](references\u002Fchecks.md) — the check catalog by category\n- [references\u002Fgrading.md](references\u002Fgrading.md) — axes, placement, collateral, verification\n- [references\u002Frepo-traps.md](references\u002Frepo-traps.md) — mergeCraft-specific failure modes\n- [references\u002Ftools.md](references\u002Ftools.md) — what to call in each environment\n","# Reviewing code in mergeCraft\n\nA mergeCraft review is a falsifiable audit of one diff: typed findings anchored to\nevidence, graded on three axes, filtered hard, then one terminal verdict. A good\nreview is not the one with the most comments — it is the one whose blocking\nconcerns survive verification and whose silence means the author can merge.\n\nThis file is **self-sufficient**. A reviewer that reads only `SKILL.md` can run a\ncorrect, complete review of this repository. The `references\u002F` files add depth;\nthey are never prerequisites. Nothing in this skill tree links outside\n`.github\u002Fskills\u002Fcode-review\u002F`.\n\n> **For humans:** `docs\u002FREVIEW-CHECKS.md` is the consumer-facing check catalog;\n> `docs\u002FREVIEW-DOCTRINE.md` is the rationale and rejected alternatives. This\n> directory is the agent-facing instruction bundle mergeCraft injects at review time.\n\n## The four rules\n\nThese four are absolute. Everything else in this file is judgement. Each rule\nstates its reason so you can generalise to cases this file never anticipated.\n\n1. **No finding without evidence you can point at.** A claim must cite a file and\n   line, a diff hunk, tool output, or a retrieved artifact. Without an anchor it\n   is a question — ask it in the summary or drop it. Unfalsifiable comments teach\n   authors to ignore this reviewer.\n\n2. **Never re-raise a withdrawn finding.** The section\n   `## Withdrawn review findings (known non-issues)` in the learnings file is\n   binding. Re-litigating a refuted finding is worse than missing a real one,\n   because it proves this reviewer does not remember prior pushback.\n\n3. **Every blocking concern in the summary exists as a findings row.** Prose in\n   the summary is read by no gate. If it blocks merge, it must appear in the\n   structured `findings` array with severity, category, and evidence.\n\n4. **Exactly one terminal verdict.** Call `approve` or `request_changes` once,\n   after findings survive verification. The verdict follows from the findings —\n   never the other way around.\n\n## Review progress\n\nCopy this into your response and check items off as you go.\n\n```\n- [ ] 1. Scope    — full diff read end-to-end; coverage checklist built\n- [ ] 2. Evidence — repo gates + analyzers run; withdrawn findings read\n- [ ] 3. Triage   — what kind of change is this; trivial or not\n- [ ] 4. Lenses   — load-bearing questions named, or none\n- [ ] 5. Findings — drafted, evidence attached, graded\n- [ ] 6. Filter   — drop list applied; blocking findings verified\n- [ ] 7. Verdict  — one terminal call\n```\n\n## 1. Scope\n\nRead the **complete diff** before anything else. Use the diff table of contents and\nfile line ranges as a coverage checklist. A reviewer that samples the diff writes\nfindings about the part it sampled and stays silent about the rest — and silence\nreads as approval.\n\nThen pull **targeted** context: the touched functions, their callers, the tests\nthat exercise them, the interfaces they cross. Not the whole repository. More\ncontext measurably makes frontier models worse at this task; stay on the blast\nradius.\n\n## 2. Evidence before opinion\n\nRun the repo's own gates first and quote them. \"`make lint` fails on this file\"\nis a finding nobody argues; \"consider sorting this\" is one everybody does.\n\nThree outcomes mean **no signal** — report them as skipped, never as findings:\n\n- the tool is not in your toolset at all;\n- the tool returns `ran: false`;\n- a gate status is `unavailable`, `declared-but-cannot-run`, or `timed_out`.\n\nOnly `failed` is a finding. Never substitute your own linter, formatter, or\ninterpreter to fill a gap — a gate run under the wrong toolchain version\nmanufactures findings.\n\nRead `## Withdrawn review findings (known non-issues)` **now**, not at drafting\ntime. It changes what you bother investigating.\n\n## 3. Triage\n\nName what kind of change this is: which domain, which seams, which external\ncontracts, which user-visible surfaces.\n\n**Genuinely trivial — skip to the verdict:** doc typo, whitespace-only,\ncomment-only, lockfile or generated-code regeneration, mechanical rename whose\nonly effect is import paths, low-risk dependency patch bump.\n\n**Looks trivial but is not — small diff, large blast radius:**\n\n- one-line changes to SQL, regex, auth, billing, permissions, or signature\n  verification;\n- a flipped feature-flag default, retry\u002Ftimeout constant, or money\u002Ftax constant;\n- a changed HTTP method, redirect URL, response code, or comparison operator;\n- a renamed public API surface or new direct dependency;\n- a \"typo fix\" in user-facing copy that changes meaning;\n- a semantic one-liner buried in a formatting-only diff.\n\nRead the shape, not the line count.\n\n## 4. Lenses\n\nName the load-bearing questions you cannot resolve yourself. A question is\nload-bearing only when its answer could change the verdict, and falsifiable only\nwhen evidence could settle it. \"Another look for confidence\" is neither.\n\nTwo framings:\n\n- **Themed lenses** across the whole diff (correctness, security, performance).\n- **Subsystem lenses** for high-stakes domains (auth, billing, webhooks, schema\n  migration). For those domains, lead with the subsystem lens — \"the billing lens\"\n  primes double-charge and refund-race failure modes that a generic correctness\n  pass misses.\n\nRun `mergecraft lens list` for the registry — each entry carries its id,\ntriggers, rubric, and required evidence. Do not work from a remembered list.\n\n## 5. Findings\n\nEvery finding carries six fields before you place it:\n\n| Field | Value |\n| --- | --- |\n| Category | one of the six taxonomy categories (see references\u002Fgrading.md) |\n| Severity | Critical · Major · Minor · Trivial |\n| Effort | Quick win · Heavy lift · Low value |\n| Confidence | certain · likely · possible |\n| Evidence | exact file and line, tool output, or retrieved artifact |\n| Why | one sentence of consequence, not of mechanism |\n\n**Placement is mechanical:** `Trivial` **or** `Low value` → a bullet in the\nbody's Nitpicks list. Everything else → an inline comment at its line.\n\nThe cost of a false positive is not the minute spent reading it — it is the team\nlearning to skip every comment this reviewer ever leaves. Grade honestly;\ninflating a nit to Major to justify an inline anchor is the habit that makes a\nreviewer ignorable.\n\n**Category is a sweep, not a menu.** A PR that writes persistent state with no\nData Integrity & Atomicity finding is worth one more look before concluding there\nwas nothing there.\n\n**Hunt for non-anchored concerns too:** deletion plans for code this diff\nshadows; rollout sequencing and in-flight state; coverage the diff implies but\ndoes not add; scope questions only a human can answer. On substantial PRs at\nleast one exists — if you cannot think of any, the bar is too high, not the PR\ntoo clean.\n\nSee [references\u002Fgrading.md](references\u002Fgrading.md) for axes, placement, collateral,\nand the verification loop.\n\n## 6. What gets dropped\n\nThe drop list is most of what keeps a review readable:\n\n- Praise and style preferences the repo does not enforce.\n- Speculative and unverified claims.\n- Findings whose root cause predates this diff.\n- Anything already withdrawn.\n- Anything not actionable.\n- Bloat-shaped fixes — defensive checks for impossible cases, abstractions used\n  once, comments restating obvious code, tests asserting tautologies,\n  just-in-case guards. The bar is sound **and** correct **and** elegant; a change\n  that improves one by degrading another makes the codebase worse.\n\n**Silence is a result.** If nothing survives this filter, say so explicitly and\napprove. An empty review from a reviewer that looked hard is a valid outcome and a\ngood one. Manufacturing a finding to look diligent is the failure mode this\nsection exists to prevent. Route silence through §7 — do not stop after §6 without\na terminal verdict.\n\n## 7. Verdict\n\nThe terminal call is exactly one of `approve` or `request_changes`, with a\n`summary` and a `findings` array. Match the opening callout tier to what the\nauthor should actually do next.\n\n- **`approve`** — no blocking findings survive verification; say so plainly.\n- **`request_changes`** — at least one Critical or Major finding blocks merge;\n  every blocking concern in the summary must have a matching findings row.\n\nThe server **rejects** `request_changes` with an empty `findings` array — that is\nenforced, not a style preference. `approve` over a verifier-confirmed Critical or\nMajor blocker is also rejected.\n\nWhen silence is the outcome: call `approve` with an empty `findings` array and a\nsummary that states what you read, what gates you ran, and that nothing actionable\nsurvived the filter.\n\nSee [references\u002Ftools.md](references\u002Ftools.md) for the call in each environment.\n\n## Tools\n\nThree capability tiers share this one skill. Use the tier that matches your\nruntime; degrade steps — never skip them silently.\n\n**Tier 1 — mergeCraft MCP (full):** `mcp__mergecraft__checkout_pr` establishes\nscope and returns `diffPath`. In the same turn call\n`mcp__mergecraft__run_static_checks` and `mcp__mergecraft__run_analyzers` with\nchanged paths; when CI failed on the head, call\n`mcp__mergecraft__analyze_ci_failures`. Before publishing, hand Critical\u002FMajor\nagent findings to `mcp__mergecraft__verify_agent_findings` and record each\nverdict with `mcp__mergecraft__record_finding_verdict`. Finish with exactly one\n`mcp__mergecraft__submit_review_verdict` (`verdict`, `summary`, `findings`).\n\n**Tier 2 — Copilot \u002F hosted code review (no MCP):** Read the PR diff from the\nhost UI. Run repo gates locally when you have shell access (`make lint`,\n`make typecheck`, scoped pytest). Draft findings as inline review comments with\nthe triage tag; post an approve or request-changes review through the host. You\ncannot call `submit_review_verdict` — make the summary and inline comments\nself-contained so a human can act without mergeCraft's structured export.\n\n**Tier 3 — bare agent (file access only):** Read the diff artifact and changed\nfiles directly. Note which gates you could not run. Produce findings with the six\nfields above; end with an explicit approve or request-changes recommendation and\nthe same discipline on evidence and silence.\n\n## References\n\n- [references\u002Fchecks.md](references\u002Fchecks.md) — the check catalog by category\n- [references\u002Fgrading.md](references\u002Fgrading.md) — axes, placement, collateral, verification\n- [references\u002Frepo-traps.md](references\u002Frepo-traps.md) — mergeCraft-specific failure modes\n- [references\u002Ftools.md](references\u002Ftools.md) — what to call in each environment\n",{"name":5,"license":15,"metadata":24,"description":7},{"mergecraft.doctrine-version":25},"1","main","c0404bd8c54b33a576f0b66b8740122d7256d321",".github\u002Fskills\u002Fcode-review",5,"https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft#-get-started-in-3-steps","https:\u002F\u002Fgithub.com\u002Falexhawat","AI-native engineer in Amsterdam. Building exclusively with agentic AI.","2026-11-15 04:52:32",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add alexhawat\u002FmergeCraft --skill code-review",{"tool":41,"command":42},"bunx","bunx skills add alexhawat\u002FmergeCraft --skill code-review",{"tool":44,"command":45},"pnpm","pnpm dlx skills add alexhawat\u002FmergeCraft --skill code-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002FSKILL.md",true,[49,55,59,63],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fchecks.md","REFERENCE","参考文档",5156,"https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002Freferences\u002Fchecks.md",{"path":56,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Fgrading.md",4117,"https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002Freferences\u002Fgrading.md",{"path":60,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Frepo-traps.md",2734,"https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002Freferences\u002Frepo-traps.md",{"path":64,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Ftools.md",4394,"https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fcode-review\u002Freferences\u002Ftools.md",[68,76,83,88,93,98],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M79RNKDBMF6Y0DNFGHTYRMG5","git-archaeology","Investigate how code reached its current state — when a line, function, import, or whole file was changed or deleted, who removed it, and what it looked like before. Use when `git blame` came up empty, when content has been refactored away, or when you need the full evolution of a function across commits.","MIT","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fsrc\u002Fmergecraft\u002Fskills\u002Fgit-archaeology\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M79RNF1TMQWNAKKV1FA6HCN0","mergecraft","Set up, run, and troubleshoot mergeCraft — the BYOK AI PR review GitHub Action and CLI. Use when the user asks to install or configure mergeCraft, add AI PR review to a repo, run a local review, interpret mergecraft-approval status or findings, configure models in .mergecraft\u002Fconfig.yaml, use mergecraft mcp serve, or debug a failing mergeCraft workflow.","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fskills\u002Fopencode\u002Fmergecraft\u002FSKILL.md",[],[],{"skillNo":84,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M79RN8VQWTKEWT6SMCB2874Q","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fskills\u002Fopenclaw\u002Fmergecraft\u002FSKILL.md",[],[],{"skillNo":89,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":19},"01M79RN31H176QC23BJRY6FBVW","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fskills\u002Fmergecraft\u002FSKILL.md",[],[],{"skillNo":94,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M79RMYGSB312HBVMAAF3ZF8H","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fskills\u002Fhermes\u002Fmergecraft\u002FSKILL.md",[],[],{"skillNo":99,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":19},"01M79RMSR72M4GYR661R7W41WV","https:\u002F\u002Fgithub.com\u002Falexhawat\u002FmergeCraft\u002Fblob\u002Fmain\u002Fskills\u002Fgrok-bot\u002Fmergecraft\u002FSKILL.md",[],[]]