[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79DVJDB4EBECKBNN1ARR1G4":3,"skill-same:01M79DVJDB4EBECKBNN1ARR1G4":98},{"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":44,"commitSha":45,"skillDirectory":46,"forksCount":14,"repositoryHtmlUrl":47,"repositoryHomepageUrl":6,"ownerHtmlUrl":48,"ownerBlogUrl":6,"ownerDescription":49,"updatedAt":50,"install":51,"resources":63},"01M79DVJDB4EBECKBNN1ARR1G4","harness-review",null,"HAR: Multi-angle code, plan, scope review. Security\u002Fquality check. Trigger: review, code review, plan review, scope analysis. Do NOT load for: implementation, new features, bugfix, setup, release.","foden303\u002Fharness","foden303","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F77870743?v=4","01M79DK06BJVG7HTKT05FM221X","Foden Ngo","01M79DK071J1HMYV20FJSMG71X",0,"MIT","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: harness-review\ndescription: \"HAR: Multi-angle code, plan, scope review. Security\u002Fquality check. Trigger: review, code review, plan review, scope analysis. Do NOT load for: implementation, new features, bugfix, setup, release.\"\ndescription-en: \"HAR: Multi-angle code, plan, scope review. Security\u002Fquality check. Trigger: review, code review, plan review, scope analysis. Do NOT load for: implementation, new features, bugfix, setup, release.\"\nkind: workflow\npurpose: \"Review code, plans, scope, and evidence before acceptance\"\ntrigger: \"review, code review, plan review, scope analysis\"\nshape: evaluate\nrole: evaluator\npair: harness-work\nowner: harness-core\nsince: \"2026-05-05\"\nallowed-tools: [\"Read\", \"Grep\", \"Glob\", \"Bash\", \"Task\", \"Monitor\", \"AskUserQuestion\"]\nargument-hint: \"[code|plan|scope|--quick|--team-debate|--security|--ui-rubric]\"\ncontext: fork\neffort: high\nuser-invocable: true\n---\n\n# Harness Review\n\nHarness's unified review skill.\nThis `SKILL.md` is a thin dispatcher; the detailed quality criteria are read from `references\u002F`.\n\nif $ARGUMENTS == \"\":\n  → interpret it as \"review of the work so far\" and run Review target detection\n  → auto-start only when the review target can be uniquely determined\n  → if the review target is unclear or has multiple candidates, present options via AskUserQuestion and align understanding before starting\n\n\u003C!-- The 3 lines above are the AUTO-START CONTRACT. Following the skill-editing.md \"within the first 3 lines\" rule, do not push them down with a fence \u002F HTML comment -->\n\n### Output Contract (P35: countermeasure for the \"looks stuck\" UX)\n\nThe **last line** of the output at the skill's conclusion must always include the following literal:\n\n`↑ Claude will summarize this result. Press Enter to continue, or give a new instruction with a fresh prompt.`\n\nThis is an explicit instruction (patterns.md P35) for the UX problem where, when shown as a text response via `\u003Clocal-command-stdout>`, the user feels it has \"stopped.\"\n\n## Dispatcher Contract\n\nThis skill's responsibility is only the review verdict.\nIt does not commit \u002F push \u002F release by default.\n\n- review default read-only boundary: read-only by default. Does not auto-commit even on `APPROVE`\n- Do not push just to review: do not push merely for review purposes\n- When a commit is needed, delegate to an explicit user request, `harness-work`, or `harness-release`'s Work Commit Gate\n- Until an explicit opt-in like `--commit-on-approve` is designed, this skill's standalone default side effects are prohibited\n\n## Quick Reference\n\n| Command | Mode | Purpose |\n|---|---|---|\n| `\u002Fharness-review` | `code` | Auto-detect and review the work so far |\n| `\u002Fharness-review --quick` | `quick` | Lightly close out a small dirty change |\n| `\u002Fharness-review --team-debate` | `team-debate` | Force a TeamAgent Debate |\n| `\u002Fharness-review --security` | `security` | security-only review |\n| `\u002Fharness-review plan` | `plan` | review the plan in `Plans.md` |\n| `\u002Fharness-review scope` | `scope` | scope creep \u002F omission review |\n## Mode Decision\n\nDetermine the execution mode from the arguments and selectively load the needed `references\u002F`.\n\n| Input | mode | reference to read |\n|---|---|---|\n| no argument \u002F `code` | `code` | `references\u002Fcode-review.md`, `references\u002Fgovernance.md` |\n| `--quick` | `quick` | `references\u002Fcode-review.md`, `references\u002Fgovernance.md` |\n| `--team-debate` | `team-debate` | `references\u002Fteam-debate.md`, `references\u002Fgovernance.md` |\n| `--security` | `security` | `references\u002Fsecurity-profile.md`, `references\u002Fgovernance.md` |\n| `--ui-rubric` | `ui-rubric` | `references\u002Fui-rubric.md` |\n| `plan` | `plan` | `references\u002Fplan-review.md`, `references\u002Fgovernance.md` |\n| `scope` | `scope` | `references\u002Fscope-review.md`, `references\u002Fgovernance.md` |\n| `full` | `full` | `references\u002Fcode-review.md`, `references\u002Fteam-debate.md`, `references\u002Fgovernance.md` |\n\n`quick` is the lightweight path.\nIt quickly looks at a small dirty change, single commit, or PR branch closeout.\nIt does not discard the quality gate.\n\n## Review Target Detection\n\n`REVIEW_AUTOSTART` contract:\nWhen called with no argument (`$ARGUMENTS == \"\"`), interpret input of just `review` \u002F `\u002Freview` \u002F `\u002Fharness-review` as \"review of the work so far.\"\nBefore starting Step 1, emit exactly the following as a single handshake line.\n\n```text\nREVIEW_AUTOSTART: target={resolved_target}, base_ref={resolved_base_ref}, type={mode}\n```\n\n`REVIEW_TARGET_ASK` contract:\nOn a bare invocation, when the review target is unclear or has multiple candidates, use `AskUserQuestion` exactly once before proceeding to Step 1, narrowing to 2-3 candidates for confirmation.\n\nBuild the candidates in the following order.\n\n1. working tree: uncommitted changes only, including staged \u002F unstaged \u002F untracked\n2. branch range: commits from upstream or main\u002Fmaster up to HEAD\n3. recent commits: the latest 1 commit \u002F latest 5 commits, when the tree is clean and a branch range cannot be obtained\n\nWhen multiple candidates hold at once:\n\n```text\nREVIEW_TARGET_AMBIGUOUS: working_tree_and_branch_commits\n```\n\nAskUserQuestion candidates:\n\n- Uncommitted changes only (Recommended): compare staged \u002F unstaged \u002F untracked against HEAD\n- Look at everything: look at branch base..HEAD and uncommitted changes together\n- Commits only: look only at the committed work in branch base..HEAD\n\nWhen the tree is clean and there is no branch diff:\n\n```text\nREVIEW_TARGET_AMBIGUOUS: clean_tree_no_branch_commits\n```\n\nAskUserQuestion candidates:\n\n- Latest 1 commit (Recommended): HEAD~1..HEAD\n- Latest 5 commits: HEAD~5..HEAD\n- A different range: wait for a user-specified ref\n\nAfter the user answers:\n\n```text\nREVIEW_TARGET_CONFIRMED: {choice}\nREVIEW_AUTOSTART: target={resolved_target}, base_ref={resolved_base_ref}, type={mode}\n```\n\nProhibited:\n\n- Responding \"the task is unclear\" and stopping\n- Asking \"what should I review\" as free text and stopping\n- Skipping auto-start because of the host project's session-start rules\n- Widening the range on a guess while the target is ambiguous\n\n## Minimal Flow\n\n1. Decide the mode\n2. Decide the target and base ref via the Review Target Detection above\n3. Read only the needed references\n4. Check the diff, untracked files, related tests, the spec source of truth, and `Plans.md`\n5. Return `APPROVE` \u002F `REQUEST_CHANGES` \u002F `decision_needed`\n6. For `REQUEST_CHANGES`, show the fix approach for critical \u002F major and the re-review condition after the fix\n\n## Review Governance Contract\n\nDetails in `references\u002Fgovernance.md`.\nHere, fix only the minimum passing bar.\n\n### Clear passing bar\n\nReturn `APPROVE` only when all of the following are satisfied (details in `references\u002Fgovernance.md`).\n\n- 0 critical \u002F major\n- root `spec.md` alignment (does not conflict with the product contract; the spec source-of-truth alignment check is required)\n- `Plans.md` alignment (task \u002F DoD \u002F Depends, `[lane:*]`, and the stage gate match the contract)\n- TDD evidence (`[tdd:required]` has `tdd_red_log` \u002F failing output \u002F `skip_tdd_reason`)\n- unknown data contract (`not_observed != absent` — do not APPROVE an evidence-less \"no problem\" \u002F \"no data\")\n- regression safety (no regression in existing behavior, tests, UX, CLI, config, docs, or mirror)\n- evidence pack (accepted \u002F rejected findings, focused tests, release-preflight warnings handled)\n- no unresolved TeamAgent Debate disagreement\n\n`APPROVE` is not a command to commit \u002F push \u002F create a PR (read-only boundary).\n\n### TeamAgent Debate\n\nDetails in `references\u002Fteam-debate.md`.\nA TeamAgent Debate is a review pass that clashes differing views read-only.\n\n| Agent | Main question |\n|---|---|\n| Spec Agent | Look for contradictions between the spec source of truth and the implementation diff |\n| Plans Agent | Confirm the correspondence between `Plans.md`'s task \u002F DoD \u002F Depends and the diff |\n| Regression Agent | Look for regressions in existing behavior, tests, the distribution mirror, and CLI\u002Fskill UX |\n| Skeptic Agent | Look for major risks overlooked under the assumption of wanting to pass |\n\nEven when native TeamAgent is unavailable, do not skip this gate.\nReproduce the same 2-4 perspectives via an available reviewer subagent or an explicitly separated read-only manual-pass, and record `native` \u002F `manual-pass` \u002F `unavailable` in `team_agent_mode`.\n\n## Code Review Summary\n\nDetails in `references\u002Fcode-review.md`.\nA normal code review looks at the following.\n\n- Security\n- Performance\n- Quality\n- Accessibility\n- AI Residuals\n- Spec Alignment\n- Plans Alignment\n- Regression Safety\n- TDD compliance\n\nDetails of root `spec.md` alignment, Plans lane\u002Fstage, TDD evidence, the unknown data contract, and the evidence pack are in `references\u002Fgovernance.md` and `references\u002Fcode-review.md`.\n\nFor `AI Residuals`, prefer `scripts\u002Freview-ai-residuals.sh` and `scripts\u002Freview-weak-supervision-report.sh`.\nWhen also looking at untracked files, use `--include-untracked`.\n`mockData`, `dummy`, `fake`, `localhost`, `TODO`, `FIXME`, `it.skip`, `test.skip`, `expect(true).toBe(true)`, etc. are candidates; decide the severity from the diff context.\nThe finding stage prioritizes coverage. Keep even findings judged minor in `observations[]` \u002F `recommendations[]`, and gate only at the verdict stage (Opus 4.8 tends to trim reporting low-severity items. See Finding coverage in `references\u002Fcode-review.md`).\n\n## Quick Closeout Summary\n\nPrinciples of the lightweight path:\n\n- Fix the target selection first\n- Treat any reviewer findings as advisory, and decide adoption after confirming in the actual code\n- The final report includes the review command \u002F tests \u002F accepted findings \u002F rejected findings \u002F clean result\n- stop-on-clean: after a clean result, do not do extra review just for appearances\n- Do not treat a failed check as a success\n\n## Plan Review Summary\n\nDetails in `references\u002Fplan-review.md`.\nPlan Review looks at the DoD \u002F Depends \u002F Status in `Plans.md` and the implementation order.\nFor a task that needs a spec source of truth, if there is no `spec_path`, stop as `decision_needed`.\n\n## Scope Review Summary\n\nDetails in `references\u002Fscope-review.md`.\nScope Review looks at whether the boundary of the request \u002F diff \u002F tests \u002F docs has ballooned.\nIf a scope change is needed, do not proceed on a guess; return to `AskUserQuestion` or a plan update.\n\n## Security \u002F UI\n\n- Security: `references\u002Fsecurity-profile.md`\n- UI rubric: `references\u002Fui-rubric.md`\n- high-res vision flow: `references\u002Fvision-high-res-flow.md`\n\n`\u002Fultrareview` is not called by default within the Harness flow.\nThis is so it does not replace the Harness flow's connection to review-result.v1, the commit guard, and the sprint-contract.\nTreat `claude ultrareview [target] --json` only as a second-opinion from CI \u002F a script.\n\n## PR Host Boundary\n\nGitHub-first.\nTreat the review facts on the PR host as authoritative via GitHub, and the local diff as supporting evidence.\nHowever, do not push a local uncommitted review to GitHub.\n\n## Output Contract\n\nUser-facing prose follows the explicit session or project language.\nIf no language is configured, use English. Use Japanese only when\n`i18n.language: ja`, `CLAUDE_CODE_HARNESS_LANG=ja`, or an explicit session\ninstruction requests Japanese output.\nMachine-readable values stay English.\n\nStart with the result summary.\n\n~~~markdown\n## Review Result\n\n### {APPROVE | REQUEST_CHANGES | decision_needed} - {one-line conclusion}\n\nTarget: `{BASE_REF}..HEAD` or `{target}`\nVerification: {commands run}\n\nStrengths:\n- ...\n\nFindings:\n- [severity] file:line - issue and evidence\n\nNext Actions:\n- ...\n\nDetails:\n```json\n{\n  \"schema_version\": \"review-result.v1\",\n  \"verdict\": \"APPROVE | REQUEST_CHANGES\",\n  \"decision_needed\": {\n    \"required\": false,\n    \"ask_tool\": \"AskUserQuestion\"\n  },\n  \"accepted_findings\": [],\n  \"rejected_findings\": [],\n  \"acceptance_bar\": {\n    \"critical_major_zero\": true,\n    \"spec_alignment\": \"pass | fail | not_applicable\",\n    \"plans_alignment\": \"pass | fail | not_applicable\",\n    \"regression_safety\": \"pass | fail | not_applicable\",\n    \"verification_evidence\": \"pass | fail | not_applicable\"\n  },\n  \"team_debate\": {\n    \"required\": false,\n    \"mode\": \"native | manual-pass | unavailable\",\n    \"team_agent_mode\": \"native | manual-pass | unavailable\",\n    \"agents\": [],\n    \"disagreements\": []\n  },\n  \"critical_issues\": [],\n  \"major_issues\": [],\n  \"observations\": [],\n  \"recommendations\": []\n}\n```\n~~~\n\n## Environment Fallbacks\n\nWhen a native tool is unavailable, the passing bar, the spec source of truth, `Plans.md`, regressions, re-review after fixes, and the AskUserQuestion \u002F `decision_needed.v1` contracts stay the same.\n\n| Native tool | Fallback |\n|---|---|\n| TeamAgent Debate via the Task tool | reviewer subagent \u002F manual-pass |\n| AskUserQuestion | when unavailable, emit `decision_needed.v1` to stdout and do not proceed on a guess |\n| TaskList | read `Plans.md` directly |\n\n## Related Skills\n\n- `harness-work`: run the fix after `REQUEST_CHANGES`\n- `harness-plan`: update the plan \u002F scope \u002F spec\n- `harness-release`: commit \u002F release the reviewed work\n","# Harness Review\n\nHarness's unified review skill.\nThis `SKILL.md` is a thin dispatcher; the detailed quality criteria are read from `references\u002F`.\n\nif $ARGUMENTS == \"\":\n  → interpret it as \"review of the work so far\" and run Review target detection\n  → auto-start only when the review target can be uniquely determined\n  → if the review target is unclear or has multiple candidates, present options via AskUserQuestion and align understanding before starting\n\n\u003C!-- The 3 lines above are the AUTO-START CONTRACT. Following the skill-editing.md \"within the first 3 lines\" rule, do not push them down with a fence \u002F HTML comment -->\n\n### Output Contract (P35: countermeasure for the \"looks stuck\" UX)\n\nThe **last line** of the output at the skill's conclusion must always include the following literal:\n\n`↑ Claude will summarize this result. Press Enter to continue, or give a new instruction with a fresh prompt.`\n\nThis is an explicit instruction (patterns.md P35) for the UX problem where, when shown as a text response via `\u003Clocal-command-stdout>`, the user feels it has \"stopped.\"\n\n## Dispatcher Contract\n\nThis skill's responsibility is only the review verdict.\nIt does not commit \u002F push \u002F release by default.\n\n- review default read-only boundary: read-only by default. Does not auto-commit even on `APPROVE`\n- Do not push just to review: do not push merely for review purposes\n- When a commit is needed, delegate to an explicit user request, `harness-work`, or `harness-release`'s Work Commit Gate\n- Until an explicit opt-in like `--commit-on-approve` is designed, this skill's standalone default side effects are prohibited\n\n## Quick Reference\n\n| Command | Mode | Purpose |\n|---|---|---|\n| `\u002Fharness-review` | `code` | Auto-detect and review the work so far |\n| `\u002Fharness-review --quick` | `quick` | Lightly close out a small dirty change |\n| `\u002Fharness-review --team-debate` | `team-debate` | Force a TeamAgent Debate |\n| `\u002Fharness-review --security` | `security` | security-only review |\n| `\u002Fharness-review plan` | `plan` | review the plan in `Plans.md` |\n| `\u002Fharness-review scope` | `scope` | scope creep \u002F omission review |\n## Mode Decision\n\nDetermine the execution mode from the arguments and selectively load the needed `references\u002F`.\n\n| Input | mode | reference to read |\n|---|---|---|\n| no argument \u002F `code` | `code` | `references\u002Fcode-review.md`, `references\u002Fgovernance.md` |\n| `--quick` | `quick` | `references\u002Fcode-review.md`, `references\u002Fgovernance.md` |\n| `--team-debate` | `team-debate` | `references\u002Fteam-debate.md`, `references\u002Fgovernance.md` |\n| `--security` | `security` | `references\u002Fsecurity-profile.md`, `references\u002Fgovernance.md` |\n| `--ui-rubric` | `ui-rubric` | `references\u002Fui-rubric.md` |\n| `plan` | `plan` | `references\u002Fplan-review.md`, `references\u002Fgovernance.md` |\n| `scope` | `scope` | `references\u002Fscope-review.md`, `references\u002Fgovernance.md` |\n| `full` | `full` | `references\u002Fcode-review.md`, `references\u002Fteam-debate.md`, `references\u002Fgovernance.md` |\n\n`quick` is the lightweight path.\nIt quickly looks at a small dirty change, single commit, or PR branch closeout.\nIt does not discard the quality gate.\n\n## Review Target Detection\n\n`REVIEW_AUTOSTART` contract:\nWhen called with no argument (`$ARGUMENTS == \"\"`), interpret input of just `review` \u002F `\u002Freview` \u002F `\u002Fharness-review` as \"review of the work so far.\"\nBefore starting Step 1, emit exactly the following as a single handshake line.\n\n```text\nREVIEW_AUTOSTART: target={resolved_target}, base_ref={resolved_base_ref}, type={mode}\n```\n\n`REVIEW_TARGET_ASK` contract:\nOn a bare invocation, when the review target is unclear or has multiple candidates, use `AskUserQuestion` exactly once before proceeding to Step 1, narrowing to 2-3 candidates for confirmation.\n\nBuild the candidates in the following order.\n\n1. working tree: uncommitted changes only, including staged \u002F unstaged \u002F untracked\n2. branch range: commits from upstream or main\u002Fmaster up to HEAD\n3. recent commits: the latest 1 commit \u002F latest 5 commits, when the tree is clean and a branch range cannot be obtained\n\nWhen multiple candidates hold at once:\n\n```text\nREVIEW_TARGET_AMBIGUOUS: working_tree_and_branch_commits\n```\n\nAskUserQuestion candidates:\n\n- Uncommitted changes only (Recommended): compare staged \u002F unstaged \u002F untracked against HEAD\n- Look at everything: look at branch base..HEAD and uncommitted changes together\n- Commits only: look only at the committed work in branch base..HEAD\n\nWhen the tree is clean and there is no branch diff:\n\n```text\nREVIEW_TARGET_AMBIGUOUS: clean_tree_no_branch_commits\n```\n\nAskUserQuestion candidates:\n\n- Latest 1 commit (Recommended): HEAD~1..HEAD\n- Latest 5 commits: HEAD~5..HEAD\n- A different range: wait for a user-specified ref\n\nAfter the user answers:\n\n```text\nREVIEW_TARGET_CONFIRMED: {choice}\nREVIEW_AUTOSTART: target={resolved_target}, base_ref={resolved_base_ref}, type={mode}\n```\n\nProhibited:\n\n- Responding \"the task is unclear\" and stopping\n- Asking \"what should I review\" as free text and stopping\n- Skipping auto-start because of the host project's session-start rules\n- Widening the range on a guess while the target is ambiguous\n\n## Minimal Flow\n\n1. Decide the mode\n2. Decide the target and base ref via the Review Target Detection above\n3. Read only the needed references\n4. Check the diff, untracked files, related tests, the spec source of truth, and `Plans.md`\n5. Return `APPROVE` \u002F `REQUEST_CHANGES` \u002F `decision_needed`\n6. For `REQUEST_CHANGES`, show the fix approach for critical \u002F major and the re-review condition after the fix\n\n## Review Governance Contract\n\nDetails in `references\u002Fgovernance.md`.\nHere, fix only the minimum passing bar.\n\n### Clear passing bar\n\nReturn `APPROVE` only when all of the following are satisfied (details in `references\u002Fgovernance.md`).\n\n- 0 critical \u002F major\n- root `spec.md` alignment (does not conflict with the product contract; the spec source-of-truth alignment check is required)\n- `Plans.md` alignment (task \u002F DoD \u002F Depends, `[lane:*]`, and the stage gate match the contract)\n- TDD evidence (`[tdd:required]` has `tdd_red_log` \u002F failing output \u002F `skip_tdd_reason`)\n- unknown data contract (`not_observed != absent` — do not APPROVE an evidence-less \"no problem\" \u002F \"no data\")\n- regression safety (no regression in existing behavior, tests, UX, CLI, config, docs, or mirror)\n- evidence pack (accepted \u002F rejected findings, focused tests, release-preflight warnings handled)\n- no unresolved TeamAgent Debate disagreement\n\n`APPROVE` is not a command to commit \u002F push \u002F create a PR (read-only boundary).\n\n### TeamAgent Debate\n\nDetails in `references\u002Fteam-debate.md`.\nA TeamAgent Debate is a review pass that clashes differing views read-only.\n\n| Agent | Main question |\n|---|---|\n| Spec Agent | Look for contradictions between the spec source of truth and the implementation diff |\n| Plans Agent | Confirm the correspondence between `Plans.md`'s task \u002F DoD \u002F Depends and the diff |\n| Regression Agent | Look for regressions in existing behavior, tests, the distribution mirror, and CLI\u002Fskill UX |\n| Skeptic Agent | Look for major risks overlooked under the assumption of wanting to pass |\n\nEven when native TeamAgent is unavailable, do not skip this gate.\nReproduce the same 2-4 perspectives via an available reviewer subagent or an explicitly separated read-only manual-pass, and record `native` \u002F `manual-pass` \u002F `unavailable` in `team_agent_mode`.\n\n## Code Review Summary\n\nDetails in `references\u002Fcode-review.md`.\nA normal code review looks at the following.\n\n- Security\n- Performance\n- Quality\n- Accessibility\n- AI Residuals\n- Spec Alignment\n- Plans Alignment\n- Regression Safety\n- TDD compliance\n\nDetails of root `spec.md` alignment, Plans lane\u002Fstage, TDD evidence, the unknown data contract, and the evidence pack are in `references\u002Fgovernance.md` and `references\u002Fcode-review.md`.\n\nFor `AI Residuals`, prefer `scripts\u002Freview-ai-residuals.sh` and `scripts\u002Freview-weak-supervision-report.sh`.\nWhen also looking at untracked files, use `--include-untracked`.\n`mockData`, `dummy`, `fake`, `localhost`, `TODO`, `FIXME`, `it.skip`, `test.skip`, `expect(true).toBe(true)`, etc. are candidates; decide the severity from the diff context.\nThe finding stage prioritizes coverage. Keep even findings judged minor in `observations[]` \u002F `recommendations[]`, and gate only at the verdict stage (Opus 4.8 tends to trim reporting low-severity items. See Finding coverage in `references\u002Fcode-review.md`).\n\n## Quick Closeout Summary\n\nPrinciples of the lightweight path:\n\n- Fix the target selection first\n- Treat any reviewer findings as advisory, and decide adoption after confirming in the actual code\n- The final report includes the review command \u002F tests \u002F accepted findings \u002F rejected findings \u002F clean result\n- stop-on-clean: after a clean result, do not do extra review just for appearances\n- Do not treat a failed check as a success\n\n## Plan Review Summary\n\nDetails in `references\u002Fplan-review.md`.\nPlan Review looks at the DoD \u002F Depends \u002F Status in `Plans.md` and the implementation order.\nFor a task that needs a spec source of truth, if there is no `spec_path`, stop as `decision_needed`.\n\n## Scope Review Summary\n\nDetails in `references\u002Fscope-review.md`.\nScope Review looks at whether the boundary of the request \u002F diff \u002F tests \u002F docs has ballooned.\nIf a scope change is needed, do not proceed on a guess; return to `AskUserQuestion` or a plan update.\n\n## Security \u002F UI\n\n- Security: `references\u002Fsecurity-profile.md`\n- UI rubric: `references\u002Fui-rubric.md`\n- high-res vision flow: `references\u002Fvision-high-res-flow.md`\n\n`\u002Fultrareview` is not called by default within the Harness flow.\nThis is so it does not replace the Harness flow's connection to review-result.v1, the commit guard, and the sprint-contract.\nTreat `claude ultrareview [target] --json` only as a second-opinion from CI \u002F a script.\n\n## PR Host Boundary\n\nGitHub-first.\nTreat the review facts on the PR host as authoritative via GitHub, and the local diff as supporting evidence.\nHowever, do not push a local uncommitted review to GitHub.\n\n## Output Contract\n\nUser-facing prose follows the explicit session or project language.\nIf no language is configured, use English. Use Japanese only when\n`i18n.language: ja`, `CLAUDE_CODE_HARNESS_LANG=ja`, or an explicit session\ninstruction requests Japanese output.\nMachine-readable values stay English.\n\nStart with the result summary.\n\n~~~markdown\n## Review Result\n\n### {APPROVE | REQUEST_CHANGES | decision_needed} - {one-line conclusion}\n\nTarget: `{BASE_REF}..HEAD` or `{target}`\nVerification: {commands run}\n\nStrengths:\n- ...\n\nFindings:\n- [severity] file:line - issue and evidence\n\nNext Actions:\n- ...\n\nDetails:\n```json\n{\n  \"schema_version\": \"review-result.v1\",\n  \"verdict\": \"APPROVE | REQUEST_CHANGES\",\n  \"decision_needed\": {\n    \"required\": false,\n    \"ask_tool\": \"AskUserQuestion\"\n  },\n  \"accepted_findings\": [],\n  \"rejected_findings\": [],\n  \"acceptance_bar\": {\n    \"critical_major_zero\": true,\n    \"spec_alignment\": \"pass | fail | not_applicable\",\n    \"plans_alignment\": \"pass | fail | not_applicable\",\n    \"regression_safety\": \"pass | fail | not_applicable\",\n    \"verification_evidence\": \"pass | fail | not_applicable\"\n  },\n  \"team_debate\": {\n    \"required\": false,\n    \"mode\": \"native | manual-pass | unavailable\",\n    \"team_agent_mode\": \"native | manual-pass | unavailable\",\n    \"agents\": [],\n    \"disagreements\": []\n  },\n  \"critical_issues\": [],\n  \"major_issues\": [],\n  \"observations\": [],\n  \"recommendations\": []\n}\n```\n~~~\n\n## Environment Fallbacks\n\nWhen a native tool is unavailable, the passing bar, the spec source of truth, `Plans.md`, regressions, re-review after fixes, and the AskUserQuestion \u002F `decision_needed.v1` contracts stay the same.\n\n| Native tool | Fallback |\n|---|---|\n| TeamAgent Debate via the Task tool | reviewer subagent \u002F manual-pass |\n| AskUserQuestion | when unavailable, emit `decision_needed.v1` to stdout and do not proceed on a guess |\n| TaskList | read `Plans.md` directly |\n\n## Related Skills\n\n- `harness-work`: run the fix after `REQUEST_CHANGES`\n- `harness-plan`: update the plan \u002F scope \u002F spec\n- `harness-release`: commit \u002F release the reviewed work\n",{"kind":24,"name":5,"pair":25,"role":26,"owner":27,"shape":28,"since":29,"effort":30,"context":31,"purpose":32,"trigger":33,"description":7,"allowed-tools":34,"argument-hint":42,"description-en":7,"user-invocable":43},"workflow","harness-work","evaluator","harness-core","evaluate","2026-05-05","high","fork","Review code, plans, scope, and evidence before acceptance","review, code review, plan review, scope analysis",[35,36,37,38,39,40,41],"Read","Grep","Glob","Bash","Task","Monitor","AskUserQuestion","[code|plan|scope|--quick|--team-debate|--security|--ui-rubric]",true,"main","3517f6fc9c5159f32d585682386ed45e35c30ddf","skills\u002Fharness-review","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness","https:\u002F\u002Fgithub.com\u002Ffoden303","Software Engineer","2026-11-06 10:59:21",{"source":8,"skillName":5,"commands":52,"prompt":62,"zipAvailable":43},[53,56,59],{"tool":54,"command":55},"npx","npx skills add foden303\u002Fharness --skill harness-review",{"tool":57,"command":58},"bunx","bunx skills add foden303\u002Fharness --skill harness-review",{"tool":60,"command":61},"pnpm","pnpm dlx skills add foden303\u002Fharness --skill harness-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002FSKILL.md",[64,70,74,78,82,86,90,94],{"path":65,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":68,"sourceHtmlUrl":69},"references\u002Fcode-review.md","REFERENCE","参考文档",3855,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fcode-review.md",{"path":71,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":72,"sourceHtmlUrl":73},"references\u002Fgovernance.md",3039,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fgovernance.md",{"path":75,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":76,"sourceHtmlUrl":77},"references\u002Fplan-review.md",876,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fplan-review.md",{"path":79,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":80,"sourceHtmlUrl":81},"references\u002Fscope-review.md",1034,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fscope-review.md",{"path":83,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":84,"sourceHtmlUrl":85},"references\u002Fsecurity-profile.md",14653,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fsecurity-profile.md",{"path":87,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":88,"sourceHtmlUrl":89},"references\u002Fteam-debate.md",1935,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fteam-debate.md",{"path":91,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":92,"sourceHtmlUrl":93},"references\u002Fui-rubric.md",3417,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fui-rubric.md",{"path":95,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":96,"sourceHtmlUrl":97},"references\u002Fvision-high-res-flow.md",7419,"https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-review\u002Freferences\u002Fvision-high-res-flow.md",[99,105,111,118,125,131],{"skillNo":100,"name":101,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M79DWR27RWW11HNKY0MRECPP","escape-skill","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fzz-review-escape\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":19},"01M79DWPSR8ZK3V4DQK5EWVJ3K","zz-review-empty","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskills\u002Fzz-review-empty\u002FSKILL.md",[],[],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":19},"01M79DWFV6PYV0QEZ3SC32VMCM","memory","Manage SSOT, memory, and cross-tool memory search. Guardian of decisions.md and patterns.md. Use when user mentions memory, SSOT, decisions.md, patterns.md, merging, migration, SSOT promotion, sync memory, save learnings, memory search, harness-mem, past decisions, or record this. Do NOT load for: implementation work, reviews, ad-hoc notes, or in-session logging.","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fmemory\u002FSKILL.md",[],[],{"skillNo":119,"name":120,"nameZh":6,"description":121,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":122,"categories":123,"keywords":124,"publishedAt":19},"01M79DWB6MB7EQK2JT4A84NHJD","maintenance","File cleanup and archiving. Tidies up bloated Plans.md, session-log.md, old logs, and state files. Trigger: \u002Fmaintenance, cleanup, archive, organize, split session-log. Do NOT load for: implementation, review, release, new feature development.","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fmaintenance\u002FSKILL.md",[],[],{"skillNo":126,"name":25,"nameZh":6,"description":127,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":128,"categories":129,"keywords":130,"publishedAt":19},"01M79DW33AEPBCXFAMX8YH56PG","HAR: Execute Plans.md tasks from single task to full parallel team run. Trigger: implement, execute, do everything, breezing, team run, parallel. Do NOT load for: planning, review, release, setup.","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-work\u002FSKILL.md",[],[],{"skillNo":132,"name":133,"nameZh":6,"description":134,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":135,"categories":136,"keywords":137,"publishedAt":19},"01M79DVWV7PZC0E2GYC103QBYZ","harness-sync","HAR: Sync Plans.md with implementation. Drift detect, marker update, retrospective. Trigger: sync-status, where am I, check progress. --snapshot for snapshots. Do NOT load for: planning, implementation, review, release.","https:\u002F\u002Fgithub.com\u002Ffoden303\u002Fharness\u002Fblob\u002Fmain\u002Fskills\u002Fharness-sync\u002FSKILL.md",[],[]]