[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6H3T5SYNGMVE9AD5AATVGDM":3,"skill-same:01M6H3T5SYNGMVE9AD5AATVGDM":92},{"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":28,"publishedAt":49,"linkOnly":50,"markdownContent":51,"markdownBody":52,"frontMatter":53,"refName":61,"commitSha":62,"skillDirectory":63,"forksCount":64,"repositoryHtmlUrl":65,"repositoryHomepageUrl":65,"ownerHtmlUrl":66,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":67,"install":68,"resources":81},"01M6H3T5SYNGMVE9AD5AATVGDM","code-review",null,"Multi-agent code review with learning feedback loop — reviews PRs using parallel specialized agents (style, logic, security) with confidence scoring and false-positive reduction","sliamh11\u002FDeus","sliamh11","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F68326628?v=4","01M6H3PPFXQ6C98QV1RJA5W72S","Liam Steiner","01M6H3PPM81GJC6KYCNTY9A3AT",51,"MIT","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcode-review\u002FSKILL.md",[18,23],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000007","代码审查","SCENARIO","使用场景",{"categoryNo":24,"categoryName":25,"taxonomyType":26,"taxonomyTypeDesc":27},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48],"review","code review","code","reviews","with","multi","style","false","logic","scoring","confidence scoring","specialized","parallel","multi-agent","feedback","loop","learning","false-positive","pr","security","2026-11-10 10:12:30",false,"---\nname: code-review\ndescription: Multi-agent code review with learning feedback loop — reviews PRs using parallel specialized agents (style, logic, security) with confidence scoring and false-positive reduction\nversion: 1.0.0\ntriggers:\n  - code.?review\n  - review.?pr\n  - review.?code\n  - pr.?review\n  - deus.?review\n---\n\n# Multi-Agent Code Review\n\nRun a multi-agent code review on the current branch's PR. Three specialized agents (style, logic, security) review in parallel, findings are confidence-scored, and dismissed findings feed back into the evolution system as negative examples for future reviews.\n\n## Prerequisites\n\n- **Git** — must be in a git repository\n- **GitHub CLI** — `gh` authenticated\n- Current branch must have an open PR\n\n## Instructions\n\nWhen the user asks for a code review or triggers this skill:\n\n### Step 1: Validate environment\n\n```bash\ngit rev-parse --is-inside-work-tree  # Must be in a repo\ngh auth status                        # Must be authenticated\n```\n\nGet the current branch and find the open PR:\n\n```bash\nBRANCH=$(git branch --show-current)\ngh pr list --head \"$BRANCH\" --state open --json number,title,baseRefName\n```\n\nIf no PR exists, ask the user if they want to create one first. Do NOT proceed without a PR.\n\n### Step 2: Check eligibility\n\nSkip review if:\n- PR is a draft (`gh pr view \u003Cnumber> --json isDraft`)\n- PR has zero code changes (only docs\u002Fconfig)\n- This exact commit SHA was already reviewed (check `resources\u002Freview-history.log` if it exists)\n\n### Step 3: Gather context\n\nCollect all inputs the review agents will need:\n\n```bash\n# Get the unified diff\ngh pr diff \u003Cnumber>\n\n# Get file list and stats\ngh pr view \u003Cnumber> --json files,additions,deletions\n\n# Get PR description for intent context\ngh pr view \u003Cnumber> --json body\n```\n\nAlso load:\n1. **CLAUDE.md files** from the repo root AND from each modified directory (use Glob to find them)\n2. **Review criteria** from `resources\u002Freview-criteria.md` (repo-specific rules)\n3. **Dismissed findings** from `resources\u002Fdismissed-log.md` (negative examples — \"do NOT flag X when Y\")\n\n### Step 4: Parallel review — launch 3 agents\n\nLaunch three review agents in parallel using the Task tool. Each agent receives:\n- The PR diff\n- Relevant CLAUDE.md content\n- Review criteria for their domain\n- Negative examples from dismissed-log.md for their category\n- Instructions to output findings in a structured format\n\n**IMPORTANT:** Include the phrase \"in parallel\" in your orchestration to ensure Teams tools are available.\n\n#### Agent prompts\n\n**Style Agent (Haiku):**\n```\nYou are a code style reviewer. Analyze this PR diff for:\n- Naming inconsistencies (variables, functions, files)\n- Dead code or unused imports\n- Inconsistency with the codebase conventions described in CLAUDE.md\n- Code duplication within the diff\n- Missing or misleading comments\n\nDO NOT flag:\n- Anything a linter would catch (formatting, semicolons, trailing spaces)\n- Style preferences not documented in CLAUDE.md\n{dismissed_style_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the issue>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n**Logic Agent (Sonnet):**\n```\nYou are a code logic reviewer. Analyze this PR diff AND the full file context for:\n- Off-by-one errors\n- Null\u002Fundefined handling gaps\n- Race conditions or async issues\n- Missing error propagation\n- Incorrect boolean logic\n- Edge cases not handled\n- Type mismatches or unsafe casts\n\nRead the FULL file for each changed file to understand context beyond the diff.\n\nDO NOT flag:\n- Hypothetical issues that require specific runtime conditions unlikely in this codebase\n- Missing validation for internal-only code paths\n{dismissed_logic_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the issue and a concrete failure scenario>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n**Security Agent (Sonnet):**\n```\nYou are a security reviewer. Analyze this PR diff for OWASP Top 10 vulnerabilities:\n- SQL injection \u002F NoSQL injection\n- XSS (cross-site scripting)\n- Command injection\n- Path traversal\n- Hardcoded secrets, API keys, tokens\n- Insecure deserialization\n- Missing authentication\u002Fauthorization checks\n- Sensitive data exposure\n- SSRF (server-side request forgery)\n\nFocus on actual exploitable patterns, not theoretical risks.\n\nDO NOT flag:\n- Internal code paths with no user input\n- Environment variables used correctly\n- Test files or fixtures\n{dismissed_security_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the vulnerability and attack vector>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n### Step 5: Confidence scoring\n\nFor each finding from Step 4, launch a Haiku agent to score confidence (0-100):\n\n```\nGiven this code review finding and the actual code context, rate your confidence (0-100) that this is a genuine issue worth fixing, not a false positive.\n\nFinding: {finding}\nCode context: {surrounding code from the file}\n\nConsider:\n- Is this actually reachable in practice?\n- Does the surrounding code already handle this?\n- Is this a real bug or just a style preference?\n\nReply with ONLY a number 0-100.\n```\n\n**Drop findings scoring below 80.** This is the key false-positive filter.\n\n### Step 6: Deduplicate and present\n\n1. Deduplicate findings by file + line number (within 3 lines counts as same location)\n2. If multiple agents flagged the same location, merge into one finding with the highest severity\n3. Sort by severity (CRITICAL → HIGH → MEDIUM → LOW), then by file path\n\nPresent as a markdown table:\n\n```\n## Code Review: PR #\u003Cnumber> — \u003Ctitle>\n\n| # | Severity | File | Line | Issue | Suggestion |\n|---|----------|------|------|-------|------------|\n| 1 | ? CRITICAL | src\u002Fauth.ts | 42 | SQL injection in query builder | Use parameterized query |\n| 2 | ? HIGH | src\u002Fapi.ts | 156 | Unhandled null from getUserById | Add null check before access |\n```\n\n### Step 7: User action\n\nAsk the user how to proceed using AskUserQuestion:\n\n**Options:**\n- ? **Review each** — walk through findings one by one\n- ⚡ **Auto-fix all** — apply all fixes automatically\n- ? **Post to PR** — post findings as a PR comment without fixing\n- ❌ **Cancel**\n\n#### If \"Review each\":\nFor each finding:\n1. Show the finding with current code context\n2. Show the proposed fix as a diff\n3. Ask: ✅ Apply | ⏭️ Dismiss | ? Modify\n\n- **Apply**: Edit the file, commit: `git add \u003Cfile> && git commit -m \"fix: \u003Ctitle>\"`\n- **Dismiss**: Log to `resources\u002Fdismissed-log.md` with reason, AND call the evolution feedback loop (see Step 8)\n- **Modify**: Let the user adjust, then apply\n\n#### If \"Auto-fix all\":\nApply all fixes sequentially, commit each one.\n\n#### If \"Post to PR\":\nFormat findings and post via `gh pr comment \u003Cnumber> --body '\u003Cformatted findings>'`\n\n### Step 8: Feedback loop (on dismiss)\n\nWhen a finding is dismissed, two things happen:\n\n**A. Local persistent memory** — append to `resources\u002Fdismissed-log.md`:\n\n```markdown\n### [STYLE|LOGIC|SECURITY] \u003Ctitle>\n- **Dismissed:** \u003Cdate>\n- **File:** \u003Cpath>:\u003Cline>\n- **Reason:** \u003Cuser's reason>\n- **Rule:** Do NOT flag \u003Cspecific pattern> when \u003Cspecific context>\n```\n\n**B. Evolution reflexion** — create a forced reflection by building the JSON safely with Python:\n\n```bash\npython3 -c \"\nimport json, subprocess, sys\npayload = json.dumps({\n    'finding': sys.argv[1],\n    'reason': sys.argv[2],\n    'file': sys.argv[3],\n    'line': int(sys.argv[4]) if sys.argv[4] != 'null' else None,\n    'group_folder': sys.argv[5] if sys.argv[5] != 'null' else None\n})\nsubprocess.run([sys.executable, 'evolution\u002Fcli.py', 'dismiss_review_finding', payload])\n\" \"\u003Ctitle>\" \"\u003Cuser reason>\" \"\u003Cpath>\" \"\u003Cline or null>\" \"\u003Cgroup or null>\"\n```\n\nThis bypasses the judge and directly creates a negative reflection that will be retrieved in future reviews via `getReflections()`. Using `json.dumps` avoids shell injection from special characters in finding titles or reasons.\n\n### Step 9: Log review commit SHA\n\nAfter all findings are processed, append the reviewed commit SHA to `resources\u002Freview-history.log`:\n\n```\n\u003Ccommit-sha> \u003Cdate> \u003Cpr-number> \u003Cfindings-count> \u003Cfixed> \u003Cdismissed>\n```\n\n### Step 10: Push\n\nIf any fixes were applied, ask the user if they want to push:\n- Yes → `git push`\n- No → inform they can push later\n\n## Review Criteria Customization\n\nUsers can customize review rules by editing `resources\u002Freview-criteria.md`. The default is created on first run with sensible defaults. The file is loaded and injected into each agent's prompt.\n\n## How the Feedback Loop Works\n\n```\nUser dismisses finding\n    ↓\nresources\u002Fdismissed-log.md (persistent, per-repo)\n    ↓\nevolution\u002Fcli.py dismiss_review_finding\n    ↓\nsave_reflection(category=\"code_review\", content=\"Do NOT flag X when Y\")\n    ↓\nNext review → getReflections(query, tools: \"code-review\")\n    ↓\nInjected as negative examples into agent prompts\n    ↓\nFalse positive rate decreases over time\n```\n","# Multi-Agent Code Review\n\nRun a multi-agent code review on the current branch's PR. Three specialized agents (style, logic, security) review in parallel, findings are confidence-scored, and dismissed findings feed back into the evolution system as negative examples for future reviews.\n\n## Prerequisites\n\n- **Git** — must be in a git repository\n- **GitHub CLI** — `gh` authenticated\n- Current branch must have an open PR\n\n## Instructions\n\nWhen the user asks for a code review or triggers this skill:\n\n### Step 1: Validate environment\n\n```bash\ngit rev-parse --is-inside-work-tree  # Must be in a repo\ngh auth status                        # Must be authenticated\n```\n\nGet the current branch and find the open PR:\n\n```bash\nBRANCH=$(git branch --show-current)\ngh pr list --head \"$BRANCH\" --state open --json number,title,baseRefName\n```\n\nIf no PR exists, ask the user if they want to create one first. Do NOT proceed without a PR.\n\n### Step 2: Check eligibility\n\nSkip review if:\n- PR is a draft (`gh pr view \u003Cnumber> --json isDraft`)\n- PR has zero code changes (only docs\u002Fconfig)\n- This exact commit SHA was already reviewed (check `resources\u002Freview-history.log` if it exists)\n\n### Step 3: Gather context\n\nCollect all inputs the review agents will need:\n\n```bash\n# Get the unified diff\ngh pr diff \u003Cnumber>\n\n# Get file list and stats\ngh pr view \u003Cnumber> --json files,additions,deletions\n\n# Get PR description for intent context\ngh pr view \u003Cnumber> --json body\n```\n\nAlso load:\n1. **CLAUDE.md files** from the repo root AND from each modified directory (use Glob to find them)\n2. **Review criteria** from `resources\u002Freview-criteria.md` (repo-specific rules)\n3. **Dismissed findings** from `resources\u002Fdismissed-log.md` (negative examples — \"do NOT flag X when Y\")\n\n### Step 4: Parallel review — launch 3 agents\n\nLaunch three review agents in parallel using the Task tool. Each agent receives:\n- The PR diff\n- Relevant CLAUDE.md content\n- Review criteria for their domain\n- Negative examples from dismissed-log.md for their category\n- Instructions to output findings in a structured format\n\n**IMPORTANT:** Include the phrase \"in parallel\" in your orchestration to ensure Teams tools are available.\n\n#### Agent prompts\n\n**Style Agent (Haiku):**\n```\nYou are a code style reviewer. Analyze this PR diff for:\n- Naming inconsistencies (variables, functions, files)\n- Dead code or unused imports\n- Inconsistency with the codebase conventions described in CLAUDE.md\n- Code duplication within the diff\n- Missing or misleading comments\n\nDO NOT flag:\n- Anything a linter would catch (formatting, semicolons, trailing spaces)\n- Style preferences not documented in CLAUDE.md\n{dismissed_style_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the issue>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n**Logic Agent (Sonnet):**\n```\nYou are a code logic reviewer. Analyze this PR diff AND the full file context for:\n- Off-by-one errors\n- Null\u002Fundefined handling gaps\n- Race conditions or async issues\n- Missing error propagation\n- Incorrect boolean logic\n- Edge cases not handled\n- Type mismatches or unsafe casts\n\nRead the FULL file for each changed file to understand context beyond the diff.\n\nDO NOT flag:\n- Hypothetical issues that require specific runtime conditions unlikely in this codebase\n- Missing validation for internal-only code paths\n{dismissed_logic_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the issue and a concrete failure scenario>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n**Security Agent (Sonnet):**\n```\nYou are a security reviewer. Analyze this PR diff for OWASP Top 10 vulnerabilities:\n- SQL injection \u002F NoSQL injection\n- XSS (cross-site scripting)\n- Command injection\n- Path traversal\n- Hardcoded secrets, API keys, tokens\n- Insecure deserialization\n- Missing authentication\u002Fauthorization checks\n- Sensitive data exposure\n- SSRF (server-side request forgery)\n\nFocus on actual exploitable patterns, not theoretical risks.\n\nDO NOT flag:\n- Internal code paths with no user input\n- Environment variables used correctly\n- Test files or fixtures\n{dismissed_security_examples}\n\nFor each finding, output exactly:\nFILE: \u003Cpath>\nLINE: \u003Cnumber>\nSEVERITY: CRITICAL | HIGH | MEDIUM | LOW\nTITLE: \u003Cshort title>\nDETAIL: \u003C1-2 sentences explaining the vulnerability and attack vector>\nSUGGESTION: \u003Cconcrete fix>\n---\n```\n\n### Step 5: Confidence scoring\n\nFor each finding from Step 4, launch a Haiku agent to score confidence (0-100):\n\n```\nGiven this code review finding and the actual code context, rate your confidence (0-100) that this is a genuine issue worth fixing, not a false positive.\n\nFinding: {finding}\nCode context: {surrounding code from the file}\n\nConsider:\n- Is this actually reachable in practice?\n- Does the surrounding code already handle this?\n- Is this a real bug or just a style preference?\n\nReply with ONLY a number 0-100.\n```\n\n**Drop findings scoring below 80.** This is the key false-positive filter.\n\n### Step 6: Deduplicate and present\n\n1. Deduplicate findings by file + line number (within 3 lines counts as same location)\n2. If multiple agents flagged the same location, merge into one finding with the highest severity\n3. Sort by severity (CRITICAL → HIGH → MEDIUM → LOW), then by file path\n\nPresent as a markdown table:\n\n```\n## Code Review: PR #\u003Cnumber> — \u003Ctitle>\n\n| # | Severity | File | Line | Issue | Suggestion |\n|---|----------|------|------|-------|------------|\n| 1 | ? CRITICAL | src\u002Fauth.ts | 42 | SQL injection in query builder | Use parameterized query |\n| 2 | ? HIGH | src\u002Fapi.ts | 156 | Unhandled null from getUserById | Add null check before access |\n```\n\n### Step 7: User action\n\nAsk the user how to proceed using AskUserQuestion:\n\n**Options:**\n- ? **Review each** — walk through findings one by one\n- ⚡ **Auto-fix all** — apply all fixes automatically\n- ? **Post to PR** — post findings as a PR comment without fixing\n- ❌ **Cancel**\n\n#### If \"Review each\":\nFor each finding:\n1. Show the finding with current code context\n2. Show the proposed fix as a diff\n3. Ask: ✅ Apply | ⏭️ Dismiss | ? Modify\n\n- **Apply**: Edit the file, commit: `git add \u003Cfile> && git commit -m \"fix: \u003Ctitle>\"`\n- **Dismiss**: Log to `resources\u002Fdismissed-log.md` with reason, AND call the evolution feedback loop (see Step 8)\n- **Modify**: Let the user adjust, then apply\n\n#### If \"Auto-fix all\":\nApply all fixes sequentially, commit each one.\n\n#### If \"Post to PR\":\nFormat findings and post via `gh pr comment \u003Cnumber> --body '\u003Cformatted findings>'`\n\n### Step 8: Feedback loop (on dismiss)\n\nWhen a finding is dismissed, two things happen:\n\n**A. Local persistent memory** — append to `resources\u002Fdismissed-log.md`:\n\n```markdown\n### [STYLE|LOGIC|SECURITY] \u003Ctitle>\n- **Dismissed:** \u003Cdate>\n- **File:** \u003Cpath>:\u003Cline>\n- **Reason:** \u003Cuser's reason>\n- **Rule:** Do NOT flag \u003Cspecific pattern> when \u003Cspecific context>\n```\n\n**B. Evolution reflexion** — create a forced reflection by building the JSON safely with Python:\n\n```bash\npython3 -c \"\nimport json, subprocess, sys\npayload = json.dumps({\n    'finding': sys.argv[1],\n    'reason': sys.argv[2],\n    'file': sys.argv[3],\n    'line': int(sys.argv[4]) if sys.argv[4] != 'null' else None,\n    'group_folder': sys.argv[5] if sys.argv[5] != 'null' else None\n})\nsubprocess.run([sys.executable, 'evolution\u002Fcli.py', 'dismiss_review_finding', payload])\n\" \"\u003Ctitle>\" \"\u003Cuser reason>\" \"\u003Cpath>\" \"\u003Cline or null>\" \"\u003Cgroup or null>\"\n```\n\nThis bypasses the judge and directly creates a negative reflection that will be retrieved in future reviews via `getReflections()`. Using `json.dumps` avoids shell injection from special characters in finding titles or reasons.\n\n### Step 9: Log review commit SHA\n\nAfter all findings are processed, append the reviewed commit SHA to `resources\u002Freview-history.log`:\n\n```\n\u003Ccommit-sha> \u003Cdate> \u003Cpr-number> \u003Cfindings-count> \u003Cfixed> \u003Cdismissed>\n```\n\n### Step 10: Push\n\nIf any fixes were applied, ask the user if they want to push:\n- Yes → `git push`\n- No → inform they can push later\n\n## Review Criteria Customization\n\nUsers can customize review rules by editing `resources\u002Freview-criteria.md`. The default is created on first run with sensible defaults. The file is loaded and injected into each agent's prompt.\n\n## How the Feedback Loop Works\n\n```\nUser dismisses finding\n    ↓\nresources\u002Fdismissed-log.md (persistent, per-repo)\n    ↓\nevolution\u002Fcli.py dismiss_review_finding\n    ↓\nsave_reflection(category=\"code_review\", content=\"Do NOT flag X when Y\")\n    ↓\nNext review → getReflections(query, tools: \"code-review\")\n    ↓\nInjected as negative examples into agent prompts\n    ↓\nFalse positive rate decreases over time\n```\n",{"name":5,"version":54,"triggers":55,"description":7},"1.0.0",[56,57,58,59,60],"code.?review","review.?pr","review.?code","pr.?review","deus.?review","main","2f24c38a2ae42d3161492f0729975b286452f4b4",".claude\u002Fskills\u002Fcode-review",7,"https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus","https:\u002F\u002Fgithub.com\u002Fsliamh11","2026-11-10 03:41:56",{"source":8,"skillName":5,"commands":69,"prompt":79,"zipAvailable":80},[70,73,76],{"tool":71,"command":72},"npx","npx skills add sliamh11\u002FDeus --skill code-review",{"tool":74,"command":75},"bunx","bunx skills add sliamh11\u002FDeus --skill code-review",{"tool":77,"command":78},"pnpm","pnpm dlx skills add sliamh11\u002FDeus --skill code-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcode-review\u002FSKILL.md",true,[82,88],{"path":83,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":87},"resources\u002Fdismissed-log.md","REFERENCE","参考文档",422,"https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcode-review\u002Fresources\u002Fdismissed-log.md",{"path":89,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":90,"sourceHtmlUrl":91},"resources\u002Freview-criteria.md",1231,"https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcode-review\u002Fresources\u002Freview-criteria.md",[93,119,137,151,160,175],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":49},"01M6H3ZC87AJDM8566RP3CQ6BF","status","Quick read-only health check — session context, workspace mounts, tool availability, and task snapshot. Use when the user asks for system status or runs \u002Fstatus.","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fstatus\u002FSKILL.md",[],[95,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118],"health","snapshot","for","read-only","quick","availability","session context","task","health check","only","use","session","user","context","read","check","workspace","tool","system",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":128,"publishedAt":49},"01M6H3Z8SK38PVH2B0W7NBP6ZP","resume","Load context from permanent memory and recent session logs. Tiered loading — CLAUDE.md full, session logs frontmatter-only by default.","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fresume\u002FSKILL.md",[125],{"categoryNo":126,"categoryName":127,"taxonomyType":26,"taxonomyTypeDesc":27},"01JBSEEDCAT000000000000038","人力资源",[121,129,130,131,132,133,134,109,111,113,135,136],"memory","load","loading","full","session logs","default","logs","frontmatter",{"skillNo":138,"name":139,"nameZh":6,"description":140,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":141,"categories":142,"keywords":143,"publishedAt":49},"01M6H3Z56ZBPKX77THMWSSHZCW","preserve","Scan the current conversation and silently save anything worth permanent memory to CLAUDE.md. Keeps the file compact — structured key:value, no prose.","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fpreserve\u002FSKILL.md",[],[139,144,145,146,147,148,149,150,129],"structured","compact","scan","conversation","prose","save","file",{"skillNo":152,"name":153,"nameZh":6,"description":154,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":155,"categories":156,"keywords":157,"publishedAt":49},"01M6H3Z1AKK58CWWY762YD3X9A","compress","Save the current session to the vault. Writes a compact session log with a TL;DR frontmatter (always) and full details below (when useful).","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fcompress\u002FSKILL.md",[],[153,33,136,132,145,158,111,149,159],"log","vault",{"skillNo":161,"name":162,"nameZh":6,"description":163,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":164,"categories":165,"keywords":166,"publishedAt":49},"01M6H3YX6STE1ZJC06EKWMY0M6","capabilities","Show what this Deus instance can do — installed skills, available tools, and system info. Read-only. Use when the user asks what the bot can do, what's installed, or runs \u002Fcapabilities.","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fcapabilities\u002FSKILL.md",[],[162,167,168,118,103,169,170,171,172,173,174,109,110,112,114],"Tools","bot","can","system info","info","what the","do","show",{"skillNo":176,"name":177,"nameZh":6,"description":178,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":179,"categories":180,"keywords":184,"publishedAt":49},"01M6H3YSNXYAR7NC1TZPFCANZN","agent-browser","Browse the web for any task — research topics, read articles, interact with web apps, fill forms, take screenshots, extract data, and test web pages. Use whenever a browser would be useful, not just when the user explicitly asks.","https:\u002F\u002Fgithub.com\u002Fsliamh11\u002FDeus\u002Fblob\u002Fmain\u002Fcontainer\u002Fskills\u002Fagent-browser\u002FSKILL.md",[181],{"categoryNo":182,"categoryName":183,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[185,186,187,102,33,188,189,190,107,191,110,114,192,193,194,195,196,197,198,199],"agent browser","browser","take","apps","pages","forms","test web","just","Bash","browse","data","web","test","research","screenshots"]