[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FEVJS9N71DSEFNSQ2TZW6Q":3,"skill-same:01M7FEVJS9N71DSEFNSQ2TZW6Q":72},{"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":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7FEVJS9N71DSEFNSQ2TZW6Q","gitnexus-review",null,"Review code changes with GitNexus from a GitHub PR URL or number, a branch\u002Fref or commit range, or local staged, unstaged, and untracked changes. Use when the user asks for a code review, merge-risk assessment, regression hunt, missing-test analysis, or a verdict on whether a PR, branch, commit range, or local diff is safe.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: gitnexus-review\ndescription: 'Review code changes with GitNexus from a GitHub PR URL or number, a branch\u002Fref or commit range, or local staged, unstaged, and untracked changes. Use when the user asks for a code review, merge-risk assessment, regression hunt, missing-test analysis, or a verdict on whether a PR, branch, commit range, or local diff is safe.'\n---\n\n# GitNexus review\n\nReview the requested change surface without editing source, committing, pushing,\nposting, or resolving threads. A later explicit request may authorize those\nactions. Use GitNexus for structural evidence and source inspection for proof;\nneither substitutes for the other.\n\n## Resolve the target\n\nAccept these forms:\n\n| Input                                                  | Review surface                                                              |\n| ------------------------------------------------------ | --------------------------------------------------------------------------- |\n| PR URL, `owner\u002Frepo#42`, `#42`, or bare number         | GitHub PR                                                                   |\n| `base...head`                                          | Merge-base range                                                            |\n| `base..head`                                           | Exact two-dot range                                                         |\n| Branch, tag, or commit                                 | Ref against the repository default branch                                   |\n| `local`, `staged`, `unstaged`, or working-tree wording | Local changes                                                               |\n| No target                                              | Current branch's open PR; otherwise local changes; otherwise current branch |\n\nAn explicit target always wins. Interpret a bare number as a PR only in a\nGitHub repository with working `gh` authentication; otherwise ask for a ref or\nURL. If implicit mode finds both branch commits and local changes, review them\nas two labeled surfaces rather than silently dropping or blending either one.\n\nRecord the resolved target kind, repository root, default branch, base SHA,\nhead SHA, merge-base when applicable, and included local states. Resolve the\ndefault branch from remote metadata (`refs\u002Fremotes\u002F\u003Cremote>\u002FHEAD` or GitHub\nrepository metadata); use `main` or `master` only as an explicit fallback and\nsay when doing so.\n\n### PR\n\nUse `gh pr view`\u002F`gh api` to pin the PR number, repository, title, URL, base\nref, base SHA, head ref, and head SHA. Fetch those exact commits without\nswitching the user's branch. Compute `git merge-base \u003Cbase> \u003Chead>` and use\nthat SHA as the review base: GitHub PR diffs are merge-base diffs, while\n`detect_changes(scope: \"compare\")` is a two-dot comparison.\n\nUse the local `git diff \u003Cmerge-base> \u003Chead>` as the complete diff source of\ntruth; use GitHub metadata for PR facts and review state. For fork PRs, fetch\nthe pull ref or the contributor remote instead of assuming the head branch\nexists on `origin`.\n\n### Branch, ref, or range\n\nResolve every ref to a commit before reviewing. For a branch or `A...B`, use\nthe merge-base as the comparison base. For an explicit `A..B`, honor `A` as\nthe exact base. Do not compare a feature branch directly with a moving default\nbranch tip when merge-base semantics were intended.\n\n### Local changes\n\nInspect `git status --short`, the staged diff, the unstaged diff, and every\nuntracked file. Use `detect_changes` with `staged`, `unstaged`, or `all` as\nrequested. Untracked files are not guaranteed to appear in Git diff or graph\nmapping, so read them directly and list them in the review provenance.\n\n## Align the checkout and index\n\nThe graph and diff must describe the same head. Reuse an existing worktree only\nwhen it is at the exact target SHA. Otherwise create a temporary detached\nworktree for the PR\u002Fref head, review there, and remove only that temporary\nworktree afterward. Never switch or reset the user's current worktree.\n\nCheck GitNexus status in the target worktree. If stale, run\n`node .gitnexus\u002Frun.cjs analyze --index-only` before trusting graph results\n(temporary worktrees never carry the gitignored `run.cjs` — fall back to the\ninstalled `gitnexus` CLI, then `npx gitnexus`), and include `--pdg` in that\nsame refresh when the diff plausibly touches trust or data-flow boundaries,\nso the taint pass below doesn't pay a second full analyze. Taint and\ndependence evidence needs that PDG layer: when the workflow's taint pass\nfinds it missing, rebuild with `analyze --pdg --index-only` and record the\nrebuild in provenance. For local changes, refresh the index so new or\nmodified source is represented.\nIf an exact target checkout\u002Findex cannot be established, state the limitation\nand do not claim a complete graph-backed review.\n\n## Review workflow\n\n1. Read the full diff and changed-file list. Separate generated files,\n   dependency churn, tests, and behavior changes.\n2. Run `detect_changes` against the exact surface:\n   - PR\u002Fbranch\u002F`...`: `scope: \"compare\"`, `base_ref: \u003Cmerge-base SHA>`.\n   - Explicit `A..B`: `scope: \"compare\"`, `base_ref: \u003CA SHA>` from a worktree\n     at `B`.\n   - Local: `scope: \"staged\"`, `\"unstaged\"`, or `\"all\"`.\n     Pass `worktree` when the MCP server is attached elsewhere.\n3. Run upstream `impact` with `includeTests: true` for each behaviorally changed\n   symbol. Prioritize public contracts, shared types, control flow, persistence,\n   security boundaries, and error handling; skip mechanical\u002Fgenerated changes.\n4. Inspect every direct (`d=1`) dependent that is outside the diff. A dependent\n   outside the diff is a lead, not automatically a bug—verify the changed\n   contract and caller behavior in source.\n5. Use `context` on key or ambiguous symbols and inspect affected execution\n   flows. Read the surrounding implementation and tests at cited locations.\n6. **Taint and dependence pass.** For changed code on trust or data-flow\n   boundaries — external input, persistence, process execution, network,\n   auth — run `explain` on the changed files or symbols and judge its\n   source→sink taint findings against the diff: a flow the change\n   introduces, or a sanitizer\u002Fguard the change removes, is a finding; a\n   pre-existing flow is context, not a defect of this change. When the\n   change claims to guard or sanitize something, verify with `pdg_query`:\n   what controls the changed statement, and where its values flow. This\n   needs a `--pdg` index; if one cannot be built, state that the taint pass\n   was skipped rather than implying coverage.\n7. Check whether tests exercise the changed behavior, boundary conditions, and\n   affected flows. Run focused read-only validation when practical. When the\n   diff refreshes a committed baseline, fingerprint, or golden, re-run the\n   exact CI check command against the head instead of trusting the committed\n   value — a stale artifact is invisible in the diff and fails only in CI.\n8. Reconcile graph evidence with the raw diff. New files, dynamic dispatch,\n   configuration, reflection, and untracked content may require direct review\n   even when graph results are empty. Version and invalidation constants are\n   review surface: when the diff changes what gets emitted or persisted,\n   verify every schema\u002Fversion constant gating caches, incremental\n   writebacks, and fingerprint baselines was bumped or regenerated — in\n   GitNexus itself, for example: graph DDL needs no manual bump, because\n   `SCHEMA_FINGERPRINT` (`gitnexus\u002Fsrc\u002Fcore\u002Flbug\u002Fschema.ts`) is derived\n   from `NODE_SCHEMA_QUERIES` + `REL_SCHEMA_QUERIES` and moves on its own;\n   the check there is whether the diff changed any string in those arrays,\n   and, if it added a new DDL array, whether that array was folded into the\n   fingerprint. The hand-maintained ritual still applies where no\n   declarative artifact describes the invalidated set: the parse-store\n   `SCHEMA_BUMP` and both bench fingerprint sets still need an explicit\n   bump, re-checked against the base branch right before merge. Semantic\n   changes that leave the DDL untouched are outside the fingerprint; they\n   rely on the analyzer runner-identity receipt in the index metadata.\n\n## Expert lenses\n\nDepth comes from matching reviewers to what actually changed, not from one\ngeneralist pass. After workflow step 2, group the changed files and symbols\nby the functional areas the graph already knows — the index's cluster\nlisting; `context` names each symbol's cluster — and give each touched area\nan expert lens: a reviewer charged with that domain's contracts, invariants,\nand failure modes, grounded in the repo's own material (architecture docs,\nagent rules, the domain's tests) before judging the diff. A lens verifies,\nnot just reads: when the changed code is a pure function reachable from the\nrepo's own toolchain — parsers, extractors, capture emitters, formatters —\nexecute it on the candidate failing shape (a scratch probe, deleted\nafterward) and cite the observed output. An empirical probe outranks source\nreading in the evidence hierarchy; role swaps, dead branches, and\nerror-recovery-dependent behavior repeatedly pass a reading and fail a\nten-line probe. The numbered\nworkflow runs exactly once; dispatch the lens passes after step 6, handing\neach lens the evidence already collected rather than letting lenses repeat\nthe `impact`, `context`, or taint calls. In GitNexus\nitself, for example: shared ingestion-pipeline changes get an ingestion\nexpert plus one language expert per changed language extractor; embeddings\nchanges an embeddings expert; LadybugDB\u002Fstorage changes a Ladybug expert.\n\nFour cross-cutting lenses run regardless of domain:\n\n- **Architectural fit** — the change lands where the architecture says the\n  concern lives, reuses existing seams, and adds no parallel structure.\n- **Language conformance** — the repo's own type\u002Flint\u002Ftest contract as\n  configured (tsconfig strictness, lint rules, test conventions); in a\n  strict TypeScript repo, for example: strictness intact, no `any`\u002F`as any`\n  escapes, module boundaries typed. Judge by the repo's contract, never a\n  universal style bar.\n- **Definition of Done** — changed behavior has tests, docs the change makes\n  stale are updated, and sync\u002Fdrift guards (shipped copies, manifests,\n  changelogs) still hold.\n- **Simplicity** — YAGNI and clear-code check: flag speculative abstraction,\n  unused knobs, and overengineering; the smallest diff that meets the\n  Definition of Done is the standard.\n\nScale effort to the surface: a single-domain change of a few files gets one\ncombined pass covering its domain lens plus the four cross-cutting checks;\na multi-domain change gets one lens per touched area — run as parallel\nsubagents where the harness supports them, each scoped to its own files\nplus the shared graph evidence, and as sequential passes otherwise. Never\nspawn a lens for a domain the diff does not touch. Merge lenses that ground\nin the same material — two lenses reading the same files pay twice for one\nread's coverage, so give one reviewer both charges. Where the harness\noffers model or effort tiers, run mechanical lenses (rename sweeps,\ndoc-consistency checks) on a cheaper tier and reserve the strongest engine\nfor adversarial judgment. Every lens reports\nthrough the Finding standard below; merge and dedup before the verdict,\ndropping anything without a concrete failing scenario.\n\n### Swarm lanes\n\nSix dispatchable lane definitions ship with this skill in `ci-personas\u002F` —\nread-only reviewers restricted to file reads plus the safe graph tools. Five are finder lanes: `ci-correctness-lens`, `ci-security-lens`,\n`ci-blast-radius-lens`, `ci-coverage-lens`, and `ci-adversarial-lens`\n(which assumes the change is broken and constructs reachable failure\nscenarios the pattern checks miss). They carry the verification\ndimensions of the numbered workflow across every touched domain; domain\ngrouping and the four cross-cutting checks above remain the\norchestrator's charge. The sixth, `ci-critic-lens`, is a gate, not a\nfinder — it audits the finished draft.\n\nWhen the harness supports subagents and these lanes are registered as\nagents (the CI review workflow installs them from its trusted control\ncheckout; a local harness may register them by copying `ci-personas\u002F*.md`\ninto `~\u002F.claude\u002Fagents\u002F` or the project's `.claude\u002Fagents\u002F`), run the\nexpert-lens pass as follows. First establish your own graph evidence —\nmake at least one substantive context call on a changed symbol yourself,\nbefore dispatching any lane, since lane calls never satisfy the evidence\nthis skill or its runner requires. Then dispatch all five finder lanes in\nparallel in a single message. Give each lane the diff, the changed-file\nmanifest, the exact base and head identifiers, the checkout paths, and the\nslice of changed files matching its charge.\n\nTreat every lane report as an unverified claim: re-anchor each finding to\nthe diff, the source, or your own graph queries before it enters the\nreview; dedup across lanes; drop anything without a concrete failing\nscenario. Lane tool calls never substitute for evidence this skill or its\nrunner requires from the orchestrating conversation itself.\n\nAfter composing the complete draft review, dispatch `ci-critic-lens` with\nthe full draft body plus the same context. On `DEFECTS`, repair the draft\nand re-dispatch the critic once; if defects remain after the second pass,\nfix what you accept, note the unresolved critic objections in the\ncoverage section, and proceed — the critic hardens the review; it never\nblocks it. This fail-open is deliberate: the critic is bounded to two\npasses so it cannot deadlock or wedge the run, and the review is still\ngated by the runner's own evidence and schema checks. (This is distinct\nfrom the separate `gitnexus-pr-swarm-review` skill, whose interactive\nroster treats its critic as a hard gate that must clear before emission;\nthis CI lane must always emit a review or a clean failure.) If subagent\ndispatch is unavailable or any lane fails, run that lane's charge inline —\nthe lanes structure the work; they never gate it.\n\n## Finding standard\n\nReport a finding only when the reviewed change introduces a concrete defect,\nregression, security issue, compatibility break, material coverage gap, or a\nmaintainability cost with a concrete carrying scenario (a dead knob, a\nduplicated contract, a drift-prone copy).\nEach finding must include:\n\n- severity and a precise `path:line` anchor;\n- the failing scenario or contract;\n- GitNexus evidence (dependent symbol\u002Fprocess) when applicable;\n- why existing code or tests do not mitigate it;\n- a concise remediation or missing test.\n\nDo not report style preferences, pre-existing issues, raw risk counts, or\nspeculation as defects. Do not infer safety from zero graph hits. Calibrate\noverall risk from consequence, reachability, reversibility, and test evidence,\nnot from the number of changed symbols alone.\n\n## Output\n\nLead with findings in severity order. If there are none, say so explicitly.\nThen provide:\n\n```markdown\n## Review: \u003Ctarget>\n\n### Findings\n\n- [HIGH|MEDIUM|LOW] `path:line` — \u003Cproblem, evidence, impact, remediation>\n\n### Change and blast-radius summary\n\n- Target\u002Fbase\u002Fhead\u002Fmerge-base and local states reviewed\n- Changed symbols and affected execution flows\n\n### Coverage and residual risk\n\n- Tests present, tests missing, graph\u002Fdiff limitations\n\n### Verdict\n\nAPPROVE | REQUEST CHANGES | NEEDS DISCUSSION\n```\n\nFor a branch or local review, use `READY`, `NOT READY`, or `NEEDS DISCUSSION`\ninstead of a PR approval action. Include the exact target SHAs so a later run\ncan tell whether the evidence is stale.\n","# GitNexus review\n\nReview the requested change surface without editing source, committing, pushing,\nposting, or resolving threads. A later explicit request may authorize those\nactions. Use GitNexus for structural evidence and source inspection for proof;\nneither substitutes for the other.\n\n## Resolve the target\n\nAccept these forms:\n\n| Input                                                  | Review surface                                                              |\n| ------------------------------------------------------ | --------------------------------------------------------------------------- |\n| PR URL, `owner\u002Frepo#42`, `#42`, or bare number         | GitHub PR                                                                   |\n| `base...head`                                          | Merge-base range                                                            |\n| `base..head`                                           | Exact two-dot range                                                         |\n| Branch, tag, or commit                                 | Ref against the repository default branch                                   |\n| `local`, `staged`, `unstaged`, or working-tree wording | Local changes                                                               |\n| No target                                              | Current branch's open PR; otherwise local changes; otherwise current branch |\n\nAn explicit target always wins. Interpret a bare number as a PR only in a\nGitHub repository with working `gh` authentication; otherwise ask for a ref or\nURL. If implicit mode finds both branch commits and local changes, review them\nas two labeled surfaces rather than silently dropping or blending either one.\n\nRecord the resolved target kind, repository root, default branch, base SHA,\nhead SHA, merge-base when applicable, and included local states. Resolve the\ndefault branch from remote metadata (`refs\u002Fremotes\u002F\u003Cremote>\u002FHEAD` or GitHub\nrepository metadata); use `main` or `master` only as an explicit fallback and\nsay when doing so.\n\n### PR\n\nUse `gh pr view`\u002F`gh api` to pin the PR number, repository, title, URL, base\nref, base SHA, head ref, and head SHA. Fetch those exact commits without\nswitching the user's branch. Compute `git merge-base \u003Cbase> \u003Chead>` and use\nthat SHA as the review base: GitHub PR diffs are merge-base diffs, while\n`detect_changes(scope: \"compare\")` is a two-dot comparison.\n\nUse the local `git diff \u003Cmerge-base> \u003Chead>` as the complete diff source of\ntruth; use GitHub metadata for PR facts and review state. For fork PRs, fetch\nthe pull ref or the contributor remote instead of assuming the head branch\nexists on `origin`.\n\n### Branch, ref, or range\n\nResolve every ref to a commit before reviewing. For a branch or `A...B`, use\nthe merge-base as the comparison base. For an explicit `A..B`, honor `A` as\nthe exact base. Do not compare a feature branch directly with a moving default\nbranch tip when merge-base semantics were intended.\n\n### Local changes\n\nInspect `git status --short`, the staged diff, the unstaged diff, and every\nuntracked file. Use `detect_changes` with `staged`, `unstaged`, or `all` as\nrequested. Untracked files are not guaranteed to appear in Git diff or graph\nmapping, so read them directly and list them in the review provenance.\n\n## Align the checkout and index\n\nThe graph and diff must describe the same head. Reuse an existing worktree only\nwhen it is at the exact target SHA. Otherwise create a temporary detached\nworktree for the PR\u002Fref head, review there, and remove only that temporary\nworktree afterward. Never switch or reset the user's current worktree.\n\nCheck GitNexus status in the target worktree. If stale, run\n`node .gitnexus\u002Frun.cjs analyze --index-only` before trusting graph results\n(temporary worktrees never carry the gitignored `run.cjs` — fall back to the\ninstalled `gitnexus` CLI, then `npx gitnexus`), and include `--pdg` in that\nsame refresh when the diff plausibly touches trust or data-flow boundaries,\nso the taint pass below doesn't pay a second full analyze. Taint and\ndependence evidence needs that PDG layer: when the workflow's taint pass\nfinds it missing, rebuild with `analyze --pdg --index-only` and record the\nrebuild in provenance. For local changes, refresh the index so new or\nmodified source is represented.\nIf an exact target checkout\u002Findex cannot be established, state the limitation\nand do not claim a complete graph-backed review.\n\n## Review workflow\n\n1. Read the full diff and changed-file list. Separate generated files,\n   dependency churn, tests, and behavior changes.\n2. Run `detect_changes` against the exact surface:\n   - PR\u002Fbranch\u002F`...`: `scope: \"compare\"`, `base_ref: \u003Cmerge-base SHA>`.\n   - Explicit `A..B`: `scope: \"compare\"`, `base_ref: \u003CA SHA>` from a worktree\n     at `B`.\n   - Local: `scope: \"staged\"`, `\"unstaged\"`, or `\"all\"`.\n     Pass `worktree` when the MCP server is attached elsewhere.\n3. Run upstream `impact` with `includeTests: true` for each behaviorally changed\n   symbol. Prioritize public contracts, shared types, control flow, persistence,\n   security boundaries, and error handling; skip mechanical\u002Fgenerated changes.\n4. Inspect every direct (`d=1`) dependent that is outside the diff. A dependent\n   outside the diff is a lead, not automatically a bug—verify the changed\n   contract and caller behavior in source.\n5. Use `context` on key or ambiguous symbols and inspect affected execution\n   flows. Read the surrounding implementation and tests at cited locations.\n6. **Taint and dependence pass.** For changed code on trust or data-flow\n   boundaries — external input, persistence, process execution, network,\n   auth — run `explain` on the changed files or symbols and judge its\n   source→sink taint findings against the diff: a flow the change\n   introduces, or a sanitizer\u002Fguard the change removes, is a finding; a\n   pre-existing flow is context, not a defect of this change. When the\n   change claims to guard or sanitize something, verify with `pdg_query`:\n   what controls the changed statement, and where its values flow. This\n   needs a `--pdg` index; if one cannot be built, state that the taint pass\n   was skipped rather than implying coverage.\n7. Check whether tests exercise the changed behavior, boundary conditions, and\n   affected flows. Run focused read-only validation when practical. When the\n   diff refreshes a committed baseline, fingerprint, or golden, re-run the\n   exact CI check command against the head instead of trusting the committed\n   value — a stale artifact is invisible in the diff and fails only in CI.\n8. Reconcile graph evidence with the raw diff. New files, dynamic dispatch,\n   configuration, reflection, and untracked content may require direct review\n   even when graph results are empty. Version and invalidation constants are\n   review surface: when the diff changes what gets emitted or persisted,\n   verify every schema\u002Fversion constant gating caches, incremental\n   writebacks, and fingerprint baselines was bumped or regenerated — in\n   GitNexus itself, for example: graph DDL needs no manual bump, because\n   `SCHEMA_FINGERPRINT` (`gitnexus\u002Fsrc\u002Fcore\u002Flbug\u002Fschema.ts`) is derived\n   from `NODE_SCHEMA_QUERIES` + `REL_SCHEMA_QUERIES` and moves on its own;\n   the check there is whether the diff changed any string in those arrays,\n   and, if it added a new DDL array, whether that array was folded into the\n   fingerprint. The hand-maintained ritual still applies where no\n   declarative artifact describes the invalidated set: the parse-store\n   `SCHEMA_BUMP` and both bench fingerprint sets still need an explicit\n   bump, re-checked against the base branch right before merge. Semantic\n   changes that leave the DDL untouched are outside the fingerprint; they\n   rely on the analyzer runner-identity receipt in the index metadata.\n\n## Expert lenses\n\nDepth comes from matching reviewers to what actually changed, not from one\ngeneralist pass. After workflow step 2, group the changed files and symbols\nby the functional areas the graph already knows — the index's cluster\nlisting; `context` names each symbol's cluster — and give each touched area\nan expert lens: a reviewer charged with that domain's contracts, invariants,\nand failure modes, grounded in the repo's own material (architecture docs,\nagent rules, the domain's tests) before judging the diff. A lens verifies,\nnot just reads: when the changed code is a pure function reachable from the\nrepo's own toolchain — parsers, extractors, capture emitters, formatters —\nexecute it on the candidate failing shape (a scratch probe, deleted\nafterward) and cite the observed output. An empirical probe outranks source\nreading in the evidence hierarchy; role swaps, dead branches, and\nerror-recovery-dependent behavior repeatedly pass a reading and fail a\nten-line probe. The numbered\nworkflow runs exactly once; dispatch the lens passes after step 6, handing\neach lens the evidence already collected rather than letting lenses repeat\nthe `impact`, `context`, or taint calls. In GitNexus\nitself, for example: shared ingestion-pipeline changes get an ingestion\nexpert plus one language expert per changed language extractor; embeddings\nchanges an embeddings expert; LadybugDB\u002Fstorage changes a Ladybug expert.\n\nFour cross-cutting lenses run regardless of domain:\n\n- **Architectural fit** — the change lands where the architecture says the\n  concern lives, reuses existing seams, and adds no parallel structure.\n- **Language conformance** — the repo's own type\u002Flint\u002Ftest contract as\n  configured (tsconfig strictness, lint rules, test conventions); in a\n  strict TypeScript repo, for example: strictness intact, no `any`\u002F`as any`\n  escapes, module boundaries typed. Judge by the repo's contract, never a\n  universal style bar.\n- **Definition of Done** — changed behavior has tests, docs the change makes\n  stale are updated, and sync\u002Fdrift guards (shipped copies, manifests,\n  changelogs) still hold.\n- **Simplicity** — YAGNI and clear-code check: flag speculative abstraction,\n  unused knobs, and overengineering; the smallest diff that meets the\n  Definition of Done is the standard.\n\nScale effort to the surface: a single-domain change of a few files gets one\ncombined pass covering its domain lens plus the four cross-cutting checks;\na multi-domain change gets one lens per touched area — run as parallel\nsubagents where the harness supports them, each scoped to its own files\nplus the shared graph evidence, and as sequential passes otherwise. Never\nspawn a lens for a domain the diff does not touch. Merge lenses that ground\nin the same material — two lenses reading the same files pay twice for one\nread's coverage, so give one reviewer both charges. Where the harness\noffers model or effort tiers, run mechanical lenses (rename sweeps,\ndoc-consistency checks) on a cheaper tier and reserve the strongest engine\nfor adversarial judgment. Every lens reports\nthrough the Finding standard below; merge and dedup before the verdict,\ndropping anything without a concrete failing scenario.\n\n### Swarm lanes\n\nSix dispatchable lane definitions ship with this skill in `ci-personas\u002F` —\nread-only reviewers restricted to file reads plus the safe graph tools. Five are finder lanes: `ci-correctness-lens`, `ci-security-lens`,\n`ci-blast-radius-lens`, `ci-coverage-lens`, and `ci-adversarial-lens`\n(which assumes the change is broken and constructs reachable failure\nscenarios the pattern checks miss). They carry the verification\ndimensions of the numbered workflow across every touched domain; domain\ngrouping and the four cross-cutting checks above remain the\norchestrator's charge. The sixth, `ci-critic-lens`, is a gate, not a\nfinder — it audits the finished draft.\n\nWhen the harness supports subagents and these lanes are registered as\nagents (the CI review workflow installs them from its trusted control\ncheckout; a local harness may register them by copying `ci-personas\u002F*.md`\ninto `~\u002F.claude\u002Fagents\u002F` or the project's `.claude\u002Fagents\u002F`), run the\nexpert-lens pass as follows. First establish your own graph evidence —\nmake at least one substantive context call on a changed symbol yourself,\nbefore dispatching any lane, since lane calls never satisfy the evidence\nthis skill or its runner requires. Then dispatch all five finder lanes in\nparallel in a single message. Give each lane the diff, the changed-file\nmanifest, the exact base and head identifiers, the checkout paths, and the\nslice of changed files matching its charge.\n\nTreat every lane report as an unverified claim: re-anchor each finding to\nthe diff, the source, or your own graph queries before it enters the\nreview; dedup across lanes; drop anything without a concrete failing\nscenario. Lane tool calls never substitute for evidence this skill or its\nrunner requires from the orchestrating conversation itself.\n\nAfter composing the complete draft review, dispatch `ci-critic-lens` with\nthe full draft body plus the same context. On `DEFECTS`, repair the draft\nand re-dispatch the critic once; if defects remain after the second pass,\nfix what you accept, note the unresolved critic objections in the\ncoverage section, and proceed — the critic hardens the review; it never\nblocks it. This fail-open is deliberate: the critic is bounded to two\npasses so it cannot deadlock or wedge the run, and the review is still\ngated by the runner's own evidence and schema checks. (This is distinct\nfrom the separate `gitnexus-pr-swarm-review` skill, whose interactive\nroster treats its critic as a hard gate that must clear before emission;\nthis CI lane must always emit a review or a clean failure.) If subagent\ndispatch is unavailable or any lane fails, run that lane's charge inline —\nthe lanes structure the work; they never gate it.\n\n## Finding standard\n\nReport a finding only when the reviewed change introduces a concrete defect,\nregression, security issue, compatibility break, material coverage gap, or a\nmaintainability cost with a concrete carrying scenario (a dead knob, a\nduplicated contract, a drift-prone copy).\nEach finding must include:\n\n- severity and a precise `path:line` anchor;\n- the failing scenario or contract;\n- GitNexus evidence (dependent symbol\u002Fprocess) when applicable;\n- why existing code or tests do not mitigate it;\n- a concise remediation or missing test.\n\nDo not report style preferences, pre-existing issues, raw risk counts, or\nspeculation as defects. Do not infer safety from zero graph hits. Calibrate\noverall risk from consequence, reachability, reversibility, and test evidence,\nnot from the number of changed symbols alone.\n\n## Output\n\nLead with findings in severity order. If there are none, say so explicitly.\nThen provide:\n\n```markdown\n## Review: \u003Ctarget>\n\n### Findings\n\n- [HIGH|MEDIUM|LOW] `path:line` — \u003Cproblem, evidence, impact, remediation>\n\n### Change and blast-radius summary\n\n- Target\u002Fbase\u002Fhead\u002Fmerge-base and local states reviewed\n- Changed symbols and affected execution flows\n\n### Coverage and residual risk\n\n- Tests present, tests missing, graph\u002Fdiff limitations\n\n### Verdict\n\nAPPROVE | REQUEST CHANGES | NEEDS DISCUSSION\n```\n\nFor a branch or local review, use `READY`, `NOT READY`, or `NEEDS DISCUSSION`\ninstead of a PR approval action. Include the exact target SHAs so a later run\ncan tell whether the evidence is stale.\n",{"name":5,"description":7},"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","07-后端与架构\u002Fgitnexus-review",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill gitnexus-review",{"tool":38,"command":39},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill gitnexus-review",{"tool":41,"command":42},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill gitnexus-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002FSKILL.md",true,[46,52,56,60,64,68],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"ci-personas\u002Fci-adversarial-lens.md","REFERENCE","参考文档",2421,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-adversarial-lens.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"ci-personas\u002Fci-blast-radius-lens.md",2241,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-blast-radius-lens.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"ci-personas\u002Fci-correctness-lens.md",2023,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-correctness-lens.md",{"path":61,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":62,"sourceHtmlUrl":63},"ci-personas\u002Fci-coverage-lens.md",2234,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-coverage-lens.md",{"path":65,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":66,"sourceHtmlUrl":67},"ci-personas\u002Fci-critic-lens.md",2338,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-critic-lens.md",{"path":69,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":70,"sourceHtmlUrl":71},"ci-personas\u002Fci-security-lens.md",2219,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fgitnexus-review\u002Fci-personas\u002Fci-security-lens.md",[73,80,85,92,99,106],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":19},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":81,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":19},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"publishedAt":19},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":19},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":19},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":107,"name":108,"nameZh":6,"description":109,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":110,"categories":111,"keywords":112,"publishedAt":19},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]