[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7BZQSWFE748936QW9Q4P4PR":3,"skill-same:01M7BZQSWFE748936QW9Q4P4PR":71},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7BZQSWFE748936QW9Q4P4PR","review-codebase",null,"Use when reviewing a whole repository rather than a diff — an inherited codebase, a pre-release sweep, a security pass, or any request to find what is wrong across a tree too large to hold in one context","zak-keown\u002Fmoe","zak-keown","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F320717241?v=4","01M7BZFXME0NW6TQX0H7AZM2KS","01M7BZFXN32VR0JN5AATRD53H4",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: review-codebase\ndescription: Use when reviewing a whole repository rather than a diff — an inherited codebase, a pre-release sweep, a security pass, or any request to find what is wrong across a tree too large to hold in one context\nargument-hint: \"[--depth shallow|medium|deep] [--verify]\"\n---\n\n# Reviewing A Codebase\n\n## Overview\n\n**You already know how to review code. What you do not do reliably is emit a\nreport another program can consume.**\n\nThat is measured, not assumed. Eleven baseline runs against this repository\nproduced severity labels, per-finding file and line, retracted findings kept\nwith their reasons, and a section naming what was examined and found clean —\nnone of it prompted. The same runs produced positional finding numbers that\nrenumber on every re-run, no frontmatter, and a coverage statement filed in a\ndifferent document from the report. One run collapsed four findings under a\nsingle `12-15. Smaller items` heading, leaving them impossible to address.\n\nSo this skill is a contract, not a lecture. Judgement is yours. The five\nelements below are the ones that were missing every time, and `fix-review`\ncannot work without them.\n\n## The output contract\n\nWrite `CODEBASE-REVIEW.md` to the root of the repository under review. It has\nexactly this shape, and every REQUIRED element earns its place by being\nsomething a later program reads.\n\n```markdown\n---\nreport: codebase-review          # REQUIRED\ngenerated: \u003CISO date>            # REQUIRED\nbase_sha: \u003Cshort sha>            # REQUIRED — what you actually read\ndepth: shallow|medium|deep       # REQUIRED\ndenominator: \u003Cn> \u003Cdefinition>    # REQUIRED — see Coverage\nfiles_opened: \u003Cn>                # REQUIRED\nfindings: {critical, high, medium, low, total}   # REQUIRED\nverified: true|false             # REQUIRED — whether --verify ran\nstatus: clean|issues_found       # REQUIRED\n---\n\n# Codebase Review — \u003Crepo>\n\n## Coverage                       \u003C!-- REQUIRED, and it comes FIRST -->\n\n## Critical\n### CR-001: \u003Ctitle>              \u003C!-- one heading per finding, never a range -->\n**File:** `path`\n**Anchor:** `symbol`, test name, or a short quoted sentence\n**Severity:** critical\n\u003Cwhat is wrong, why it is wrong, and the fix>\n\n## High\n## Medium\n## Low\n## Checked and found sound        \u003C!-- optional, and worth writing -->\n```\n\n### Finding IDs\n\n`CR-###`, zero-padded to three, assigned once at merge in severity order —\ncritical first, then high, medium, low; within a severity, by path. **Never a\npositional integer.** `fix-review` addresses findings by ID and stamps\ndispositions back against them, so an ID that shifts when a finding is added\nsilently repoints every record. A heading covering more than one finding is the\nsame defect in a worse form: it cannot be addressed at all.\n\n### Finding anchors\n\nUse the repository path plus a stable symbol, test name, or short quoted\nsentence. Never use a line number. Parallel workers can read adjacent commits,\nand the repair itself moves lines; a line offset becomes false evidence while\nthe named symbol or behavior still identifies the defect.\n\n### Severity\n\n`critical` \u002F `high` \u002F `medium` \u002F `low`. Four, fixed, stated here because\notherwise it is invented per run — baseline runs produced `High\u002FMedium\u002FLow`,\n`High\u002FMedium\u002FLow\u002FPolicy`, and one that filed credential capture on a shared host\nas High with no critical rung at all.\n\n- **critical** — exploitable now, or destroys data, or leaks a credential\n- **high** — wrong under ordinary conditions, or a security control that does\n  not hold\n- **medium** — wrong under conditions a reader would not expect to be excluded\n- **low** — real, small, and safe to defer\n\nA policy decision worth recording rather than fixing is not a finding. Put it\nunder `## Checked and found sound` or its own section, outside the ID sequence.\n\n### Coverage\n\n**The coverage section goes in the report, above the findings, and it names its\nown denominator.** Every baseline run stated coverage honestly when asked and\nthen filed it somewhere the report does not travel with. The report is what\nreaches the reader.\n\nThree runs counting \"source files\" in the same tree got 874, 935 and 943. A\nratio whose denominator is undefined is decoration, so state the rule you used:\n\n```\nDenominator: 903 files matching *.ts,*.js,*.mjs,*.cjs,*.py,*.rs,*.sh under\n`git ls-files`, excluding lockfiles, dist\u002F, and generated output.\nOpened: 31 (5 partial). Not opened: 872.\nAbsence of findings in an unopened area is evidence nobody looked, not\nevidence it is clean.\n```\n\nName the unopened areas by package or directory. \"I did not reach\n`packages\u002Fflight`\" is worth more to the reader than three more low findings.\n\n## Depth\n\n`--depth shallow|medium|deep`, default `medium`. It widens the file set and adds\nreview lenses together.\n\n| | files | lenses | model |\n|---|---|---|---|\n| `shallow` | entrypoints, plus files changed most often in git history | correctness, security | {model-default} |\n| `medium` | all tracked source after exclusions | plus tests, error handling, API contracts | {model-default} |\n| `deep` | plus tests, config, scripts, infrastructure | plus performance, coupling, dependency risk | **{model-deep}** |\n\nCredential-bearing paths — `.env`, `*.pem`, `id_rsa`, `.npmrc` and the like —\nare in scope at **every** depth regardless of extension. A committed secret is\nthe highest-severity thing a review can find and it never lives in a file with a\ncode extension, so an extension filter is exactly the wrong instrument for it.\n\nOnly `deep` escalates the model, by passing `model:` {model-deep} when dispatching\n`review-shard`. Shallow is cheaper by scope, not by capability.\n\n**This is review breadth. It is not the `patch` \u002F `change` \u002F `feature` workflow\ndepth that `brainstorming` defines** — that axis classifies a unit of work\nbefore planning it; this one sizes a review of work that already exists. The two\nnever appear in the same decision.\n\n## Running it\n\nResolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-scope.mjs} relative to this loaded document, then invoke it as:\n\n```bash\nnode \"\u003Cresolved-review-scope.mjs>\" \\\n  --depth medium --out .moe\u002Freview-shards\n```\n\nIt enumerates via `git ls-files`, applies the exclusions, groups by top-level\ndirectory, splits any group over 30 files, and writes a shard manifest plus one\nfile list per shard. The default `.moe\u002Freview-shards\u002F` workspace is\nrepository-local and self-ignoring: shard reports survive session boundaries\nwithout leaking into a broad `git add`. Do this with the script rather than by\nhand: it is the part that must be identical across runs, and reproducing it in\nprose costs tokens on every invocation to get a different answer each time.\n\nThen dispatch one `review-shard` agent per shard, in waves of at most 8\nconcurrent. Each writes its own shard report; none of them commit. The agent\ndefinition carries an off-limits list for the host it runs on; the dispatch\nprompt may add to it, never subtract.\n\nAs reports land, resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-check.mjs} relative to this loaded document and validate each one before the merge ever sees it:\n\n```bash\nnode \"\u003Cresolved-review-check.mjs>\" \\\n  --shards .moe\u002Freview-shards\n```\n\nIt applies the merge's own grammar per report the moment the report exists,\nplus the lint the merge cannot see: a `###` inside a fenced block, a\nline-number citation in a body, a `**File:**` the tree does not contain. One\nmalformed record fails the whole merge, so fix the report (or re-dispatch the\nshard) while that reviewer is still around. Pass `--require-all` for the final\npass before merging; `--shard \u003Cid>` checks one report.\n\n**If parallel dispatch is unavailable** — session policy, runtime limits, a\nharness without subagents — run the shards yourself, serially, in manifest\norder. The output is identical and only the wall clock changes. See\n`..\u002F_shared\u002Fparallel-adversarial-review.md` for the fallback this follows.\n\nFinally, resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-merge.mjs} relative to this loaded document:\n\n```bash\nnode \"\u003Cresolved-review-merge.mjs>\" \\\n  --shards .moe\u002Freview-shards --out CODEBASE-REVIEW.md\n```\n\nThe merge assigns the `CR-###` sequence, builds the frontmatter, and fails loudly\nif a shard report is missing rather than silently reporting a smaller tree. It\nalso requires every shard's machine-readable base SHA and opened-file count to\nmatch the manifest, requires `HEAD` still to be that base, and refuses a dirty\ntracked tree. Do not change source between scope and merge; restart the review\nfrom a new clean commit if the tree moves.\n\n## `--verify`\n\nWith `--verify`, every critical and high finding goes to a `verify-finding`\nagent whose job is to **refute it** — reproduce the defect or show it cannot\nhappen. Findings that survive are marked `verified: confirmed`; those that fall\nare demoted or dropped, with the refutation kept.\n\nRun the ordinary merge first so the serious findings have their stable\n`CR-###` IDs. Resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-verify-scope.mjs} relative to this loaded document, then split them out:\n\n```bash\nnode \"\u003Cresolved-review-verify-scope.mjs>\" \\\n  --shards .moe\u002Freview-shards --report CODEBASE-REVIEW.md\n```\n\nThat writes one `.moe\u002Freview-shards\u002Fverify\u002FCR-###.md` per critical and high\nfinding, plus a manifest of the ID set. Dispatch one `verify-finding` agent per\nfile, pointing it at its file. Every reply ends in a `VERDICT-JSON:` line.\nResolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-verify-record.mjs}\nrelative to this loaded document and record each reply as it arrives, from the\nsaved reply or the bare object:\n\n```bash\nnode \"\u003Cresolved-review-verify-record.mjs>\" \\\n  --shards .moe\u002Freview-shards --from-file reply.txt\nnode \"\u003Cresolved-review-verify-record.mjs>\" \\\n  --shards .moe\u002Freview-shards \\\n  '{\"id\":\"CR-001\",\"verdict\":\"confirmed\",\"evidence\":\"Reproduced from the public route.\"}'\n```\n\nValid verdicts are `confirmed`, `confirmed-lower`, `refuted`, and `unproven`.\nOnly `confirmed-lower` carries a replacement severity, and it must be lower\nthan the original. Evidence is REQUIRED for every result. The recorder refuses\nan ID outside the manifest, a verdict outside those four, a `confirmed-lower`\nthat does not lower, empty or over-long evidence, and a second verdict for an\nID unless `--replace` is passed; it rewrites\n`.moe\u002Freview-shards\u002Fverifications.json` whole on every call. When its tally\nshows nothing missing, finalize:\n\n```bash\nnode \"\u003Cresolved-review-merge.mjs>\" \\\n  --shards .moe\u002Freview-shards \\\n  --verification-results .moe\u002Freview-shards\u002Fverifications.json \\\n  --out CODEBASE-REVIEW.md\n```\n\nThe merge refuses a mismatched base, duplicate\u002Fextra results, or a missing\nverdict for any critical\u002Fhigh ID. A bare `--verified` flag is rejected: the\nreport can say `verified: true` only after consuming a complete ledger.\n\nThis follows `_shared\u002Fparallel-adversarial-review.md`, adapted: PAR pairs two\nreviewers over the same code, this pairs a challenger against one finding.\n\n**PAR says it is always-on with no opt-out. That rule is scoped to the\niterate cluster's gates**, so a flag here is not a violation of\nit. Verification costs one dispatch per serious finding, which is worth choosing\ndeliberately on a large tree.\n\n## Red flags\n\n- A finding heading covering a range (`12-15. Smaller items`)\n- Positional integers where `CR-###` belongs\n- A coverage number whose denominator is not stated next to it\n- Reporting only what you found, with no line naming what you did not open\n- Inventing a fifth severity rung, or filing a credential leak below critical\n- Merging a shard report that `review-check.mjs` has not passed\n- A verify reply with no `VERDICT-JSON:` line, or a verdict typed into the\n  ledger by hand\n- A worker that ran a repository script to prove the script touches `$HOME`\n","# Reviewing A Codebase\n\n## Overview\n\n**You already know how to review code. What you do not do reliably is emit a\nreport another program can consume.**\n\nThat is measured, not assumed. Eleven baseline runs against this repository\nproduced severity labels, per-finding file and line, retracted findings kept\nwith their reasons, and a section naming what was examined and found clean —\nnone of it prompted. The same runs produced positional finding numbers that\nrenumber on every re-run, no frontmatter, and a coverage statement filed in a\ndifferent document from the report. One run collapsed four findings under a\nsingle `12-15. Smaller items` heading, leaving them impossible to address.\n\nSo this skill is a contract, not a lecture. Judgement is yours. The five\nelements below are the ones that were missing every time, and `fix-review`\ncannot work without them.\n\n## The output contract\n\nWrite `CODEBASE-REVIEW.md` to the root of the repository under review. It has\nexactly this shape, and every REQUIRED element earns its place by being\nsomething a later program reads.\n\n```markdown\n---\nreport: codebase-review          # REQUIRED\ngenerated: \u003CISO date>            # REQUIRED\nbase_sha: \u003Cshort sha>            # REQUIRED — what you actually read\ndepth: shallow|medium|deep       # REQUIRED\ndenominator: \u003Cn> \u003Cdefinition>    # REQUIRED — see Coverage\nfiles_opened: \u003Cn>                # REQUIRED\nfindings: {critical, high, medium, low, total}   # REQUIRED\nverified: true|false             # REQUIRED — whether --verify ran\nstatus: clean|issues_found       # REQUIRED\n---\n\n# Codebase Review — \u003Crepo>\n\n## Coverage                       \u003C!-- REQUIRED, and it comes FIRST -->\n\n## Critical\n### CR-001: \u003Ctitle>              \u003C!-- one heading per finding, never a range -->\n**File:** `path`\n**Anchor:** `symbol`, test name, or a short quoted sentence\n**Severity:** critical\n\u003Cwhat is wrong, why it is wrong, and the fix>\n\n## High\n## Medium\n## Low\n## Checked and found sound        \u003C!-- optional, and worth writing -->\n```\n\n### Finding IDs\n\n`CR-###`, zero-padded to three, assigned once at merge in severity order —\ncritical first, then high, medium, low; within a severity, by path. **Never a\npositional integer.** `fix-review` addresses findings by ID and stamps\ndispositions back against them, so an ID that shifts when a finding is added\nsilently repoints every record. A heading covering more than one finding is the\nsame defect in a worse form: it cannot be addressed at all.\n\n### Finding anchors\n\nUse the repository path plus a stable symbol, test name, or short quoted\nsentence. Never use a line number. Parallel workers can read adjacent commits,\nand the repair itself moves lines; a line offset becomes false evidence while\nthe named symbol or behavior still identifies the defect.\n\n### Severity\n\n`critical` \u002F `high` \u002F `medium` \u002F `low`. Four, fixed, stated here because\notherwise it is invented per run — baseline runs produced `High\u002FMedium\u002FLow`,\n`High\u002FMedium\u002FLow\u002FPolicy`, and one that filed credential capture on a shared host\nas High with no critical rung at all.\n\n- **critical** — exploitable now, or destroys data, or leaks a credential\n- **high** — wrong under ordinary conditions, or a security control that does\n  not hold\n- **medium** — wrong under conditions a reader would not expect to be excluded\n- **low** — real, small, and safe to defer\n\nA policy decision worth recording rather than fixing is not a finding. Put it\nunder `## Checked and found sound` or its own section, outside the ID sequence.\n\n### Coverage\n\n**The coverage section goes in the report, above the findings, and it names its\nown denominator.** Every baseline run stated coverage honestly when asked and\nthen filed it somewhere the report does not travel with. The report is what\nreaches the reader.\n\nThree runs counting \"source files\" in the same tree got 874, 935 and 943. A\nratio whose denominator is undefined is decoration, so state the rule you used:\n\n```\nDenominator: 903 files matching *.ts,*.js,*.mjs,*.cjs,*.py,*.rs,*.sh under\n`git ls-files`, excluding lockfiles, dist\u002F, and generated output.\nOpened: 31 (5 partial). Not opened: 872.\nAbsence of findings in an unopened area is evidence nobody looked, not\nevidence it is clean.\n```\n\nName the unopened areas by package or directory. \"I did not reach\n`packages\u002Fflight`\" is worth more to the reader than three more low findings.\n\n## Depth\n\n`--depth shallow|medium|deep`, default `medium`. It widens the file set and adds\nreview lenses together.\n\n| | files | lenses | model |\n|---|---|---|---|\n| `shallow` | entrypoints, plus files changed most often in git history | correctness, security | {model-default} |\n| `medium` | all tracked source after exclusions | plus tests, error handling, API contracts | {model-default} |\n| `deep` | plus tests, config, scripts, infrastructure | plus performance, coupling, dependency risk | **{model-deep}** |\n\nCredential-bearing paths — `.env`, `*.pem`, `id_rsa`, `.npmrc` and the like —\nare in scope at **every** depth regardless of extension. A committed secret is\nthe highest-severity thing a review can find and it never lives in a file with a\ncode extension, so an extension filter is exactly the wrong instrument for it.\n\nOnly `deep` escalates the model, by passing `model:` {model-deep} when dispatching\n`review-shard`. Shallow is cheaper by scope, not by capability.\n\n**This is review breadth. It is not the `patch` \u002F `change` \u002F `feature` workflow\ndepth that `brainstorming` defines** — that axis classifies a unit of work\nbefore planning it; this one sizes a review of work that already exists. The two\nnever appear in the same decision.\n\n## Running it\n\nResolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-scope.mjs} relative to this loaded document, then invoke it as:\n\n```bash\nnode \"\u003Cresolved-review-scope.mjs>\" \\\n  --depth medium --out .moe\u002Freview-shards\n```\n\nIt enumerates via `git ls-files`, applies the exclusions, groups by top-level\ndirectory, splits any group over 30 files, and writes a shard manifest plus one\nfile list per shard. The default `.moe\u002Freview-shards\u002F` workspace is\nrepository-local and self-ignoring: shard reports survive session boundaries\nwithout leaking into a broad `git add`. Do this with the script rather than by\nhand: it is the part that must be identical across runs, and reproducing it in\nprose costs tokens on every invocation to get a different answer each time.\n\nThen dispatch one `review-shard` agent per shard, in waves of at most 8\nconcurrent. Each writes its own shard report; none of them commit. The agent\ndefinition carries an off-limits list for the host it runs on; the dispatch\nprompt may add to it, never subtract.\n\nAs reports land, resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-check.mjs} relative to this loaded document and validate each one before the merge ever sees it:\n\n```bash\nnode \"\u003Cresolved-review-check.mjs>\" \\\n  --shards .moe\u002Freview-shards\n```\n\nIt applies the merge's own grammar per report the moment the report exists,\nplus the lint the merge cannot see: a `###` inside a fenced block, a\nline-number citation in a body, a `**File:**` the tree does not contain. One\nmalformed record fails the whole merge, so fix the report (or re-dispatch the\nshard) while that reviewer is still around. Pass `--require-all` for the final\npass before merging; `--shard \u003Cid>` checks one report.\n\n**If parallel dispatch is unavailable** — session policy, runtime limits, a\nharness without subagents — run the shards yourself, serially, in manifest\norder. The output is identical and only the wall clock changes. See\n`..\u002F_shared\u002Fparallel-adversarial-review.md` for the fallback this follows.\n\nFinally, resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-merge.mjs} relative to this loaded document:\n\n```bash\nnode \"\u003Cresolved-review-merge.mjs>\" \\\n  --shards .moe\u002Freview-shards --out CODEBASE-REVIEW.md\n```\n\nThe merge assigns the `CR-###` sequence, builds the frontmatter, and fails loudly\nif a shard report is missing rather than silently reporting a smaller tree. It\nalso requires every shard's machine-readable base SHA and opened-file count to\nmatch the manifest, requires `HEAD` still to be that base, and refuses a dirty\ntracked tree. Do not change source between scope and merge; restart the review\nfrom a new clean commit if the tree moves.\n\n## `--verify`\n\nWith `--verify`, every critical and high finding goes to a `verify-finding`\nagent whose job is to **refute it** — reproduce the defect or show it cannot\nhappen. Findings that survive are marked `verified: confirmed`; those that fall\nare demoted or dropped, with the refutation kept.\n\nRun the ordinary merge first so the serious findings have their stable\n`CR-###` IDs. Resolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-verify-scope.mjs} relative to this loaded document, then split them out:\n\n```bash\nnode \"\u003Cresolved-review-verify-scope.mjs>\" \\\n  --shards .moe\u002Freview-shards --report CODEBASE-REVIEW.md\n```\n\nThat writes one `.moe\u002Freview-shards\u002Fverify\u002FCR-###.md` per critical and high\nfinding, plus a manifest of the ID set. Dispatch one `verify-finding` agent per\nfile, pointing it at its file. Every reply ends in a `VERDICT-JSON:` line.\nResolve {resource:skills\u002Freview-codebase\u002Fscripts\u002Freview-verify-record.mjs}\nrelative to this loaded document and record each reply as it arrives, from the\nsaved reply or the bare object:\n\n```bash\nnode \"\u003Cresolved-review-verify-record.mjs>\" \\\n  --shards .moe\u002Freview-shards --from-file reply.txt\nnode \"\u003Cresolved-review-verify-record.mjs>\" \\\n  --shards .moe\u002Freview-shards \\\n  '{\"id\":\"CR-001\",\"verdict\":\"confirmed\",\"evidence\":\"Reproduced from the public route.\"}'\n```\n\nValid verdicts are `confirmed`, `confirmed-lower`, `refuted`, and `unproven`.\nOnly `confirmed-lower` carries a replacement severity, and it must be lower\nthan the original. Evidence is REQUIRED for every result. The recorder refuses\nan ID outside the manifest, a verdict outside those four, a `confirmed-lower`\nthat does not lower, empty or over-long evidence, and a second verdict for an\nID unless `--replace` is passed; it rewrites\n`.moe\u002Freview-shards\u002Fverifications.json` whole on every call. When its tally\nshows nothing missing, finalize:\n\n```bash\nnode \"\u003Cresolved-review-merge.mjs>\" \\\n  --shards .moe\u002Freview-shards \\\n  --verification-results .moe\u002Freview-shards\u002Fverifications.json \\\n  --out CODEBASE-REVIEW.md\n```\n\nThe merge refuses a mismatched base, duplicate\u002Fextra results, or a missing\nverdict for any critical\u002Fhigh ID. A bare `--verified` flag is rejected: the\nreport can say `verified: true` only after consuming a complete ledger.\n\nThis follows `_shared\u002Fparallel-adversarial-review.md`, adapted: PAR pairs two\nreviewers over the same code, this pairs a challenger against one finding.\n\n**PAR says it is always-on with no opt-out. That rule is scoped to the\niterate cluster's gates**, so a flag here is not a violation of\nit. Verification costs one dispatch per serious finding, which is worth choosing\ndeliberately on a large tree.\n\n## Red flags\n\n- A finding heading covering a range (`12-15. Smaller items`)\n- Positional integers where `CR-###` belongs\n- A coverage number whose denominator is not stated next to it\n- Reporting only what you found, with no line naming what you did not open\n- Inventing a fifth severity rung, or filing a credential leak below critical\n- Merging a shard report that `review-check.mjs` has not passed\n- A verify reply with no `VERDICT-JSON:` line, or a verdict typed into the\n  ledger by hand\n- A worker that ran a repository script to prove the script touches `$HOME`\n",{"name":5,"description":7,"argument-hint":23},"[--depth shallow|medium|deep] [--verify]","main","728894f3f85fbd4e788f015ddfd72bb6155d5dd5","packages\u002Fcore\u002Fskills\u002Freview-codebase",1,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe","https:\u002F\u002Fgithub.com\u002Fzak-keown","2026-09-05 06:29:06",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add zak-keown\u002Fmoe --skill review-codebase",{"tool":37,"command":38},"bunx","bunx skills add zak-keown\u002Fmoe --skill review-codebase",{"tool":40,"command":41},"pnpm","pnpm dlx skills add zak-keown\u002Fmoe --skill review-codebase","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002FSKILL.md",true,[45,51,55,59,63,67],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"scripts\u002Freview-check.mjs","SCRIPT","脚本",4607,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-check.mjs",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"scripts\u002Freview-merge.mjs",11598,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-merge.mjs",{"path":56,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":57,"sourceHtmlUrl":58},"scripts\u002Freview-report.mjs",3101,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-report.mjs",{"path":60,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":61,"sourceHtmlUrl":62},"scripts\u002Freview-scope.mjs",9235,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-scope.mjs",{"path":64,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Freview-verify-record.mjs",4816,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-verify-record.mjs",{"path":68,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Freview-verify-scope.mjs",3535,"https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fskills\u002Freview-codebase\u002Fscripts\u002Freview-verify-scope.mjs",[72,79,86,93,100,107],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":18},"01M7C1Z5362Y8NVRZ69HWP0ZD0","write-skill","Use when creating new skills, editing existing skills, or verifying skills work before deployment","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fwrite-skill\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":18},"01M7C1YZ9966WK9R6ZRPCGWF3G","write-plan","Use when you have a spec or requirements for a multi-step task, before touching code","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fwrite-plan\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":18},"01M7C1YMR4SJD7BPBBPXZCVGVH","write-clearly","Apply Strunk's timeless writing rules to ANY prose humans will read—documentation, commit messages, error messages, explanations, reports, or UI text. Makes your writing clearer, stronger, and more professional.","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fwrite-clearly\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":18},"01M7C1YH9805YSM0N20VQAVSXF","windows-vm","Create, manage, or connect to a headless Windows 11 VM running in Docker with SSH access. Use when the user wants to spin up, stop, restart, or SSH into a Windows VM.","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fwindows-vm\u002FSKILL.md",[],[],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":18},"01M7C1YD87SCNEJ2C9B8HFR2NQ","verify-completion","Use when about to claim work is complete, fixed, or passing, before committing or creating MRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fverify-completion\u002FSKILL.md",[],[],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":18},"01M7C1YARDEHT0TJ9BRG30ZCF4","using-moe","Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions","https:\u002F\u002Fgithub.com\u002Fzak-keown\u002Fmoe\u002Fblob\u002Fmain\u002Fplugins\u002Fmoe\u002Fskills\u002Fusing-moe\u002FSKILL.md",[],[]]