[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79NCTFEY9JB1PC7JFVYBK7D":3,"skill-same:01M79NCTFEY9JB1PC7JFVYBK7D":58},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":13,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M79NCTFEY9JB1PC7JFVYBK7D","plugin-evaluation",null,"Use when designing systematic evaluation methodology for Codex agents and commands — constructing weighted rubrics, choosing between direct scoring and pairwise comparison, building LLM-as-judge pipelines, or mitigating position, verbosity, and self-enhancement bias. NOT for empirical benchmarks with timing\u002Ftoken metrics (use skill-tester) or for iterating on skill content during development (use skill-development).","AndreHahm\u002Fandres-cc-marketplace","AndreHahm","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F80748019?v=4","01M79N2VZJT9MAJ7K695C9VXQ4","01M79N2W06PFX8BSCQ592M8K8K",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fplugin-evaluation\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: plugin-evaluation\ndescription: >-\n  Use when designing systematic evaluation methodology for Codex agents and\n  commands — constructing weighted rubrics, choosing between direct scoring and\n  pairwise comparison, building LLM-as-judge pipelines, or mitigating position,\n  verbosity, and self-enhancement bias. NOT for empirical benchmarks with\n  timing\u002Ftoken metrics (use skill-tester) or for iterating on skill content\n  during development (use skill-development).\nallowed-tools: Read Glob Grep\n---\n\n# Evaluation Methods for Codex Agents\n\nEvaluation of agent systems requires outcome-focused approaches that account for non-determinism and multiple valid paths. Agents may reach goals through different routes — evaluation should judge whether they achieve right outcomes through reasonable processes, not match exact execution steps.\n\n## Quick Start\n\n1. **Identify what to evaluate** — a command's outputs, an agent's behavior, or a prompt variant\n2. **Design test cases** — cover at least two complexity levels: simple (single tool call) and complex (multi-step reasoning)\n3. **Choose method** — LLM-as-judge (scale + speed), human evaluation (catches edge cases), or end-state (artifact correctness)\n4. **Apply rubric** — instruction following (0.30), completeness (0.25), tool efficiency (0.20), reasoning (0.15), coherence (0.10)\n5. **Iterate** — compare before\u002Fafter scores on the **same** test set; check no dimension regressed\n\n## When to Use\n\n- Testing whether a command or skill produces consistently good outputs\n- Validating that context engineering changes improve (not regress) quality\n- Scoring agent outputs systematically with LLM-as-judge\n- Comparing prompt variants objectively before choosing one\n- Detecting or mitigating position bias, verbosity bias, or self-enhancement bias\n\n## When NOT to Use\n\n- Debugging why a single command invocation failed — use `Codex --debug` and troubleshooting instead\n- For one-off gut-check tests — evaluation is for systematic, repeatable quality measurement\n- For evaluating plugin structure or manifest correctness — use `plugin-development` or `plugin-validator` instead\n\n---\n\n## Core Concepts\n\nAgent evaluation captures multiple quality dimensions: factual accuracy, completeness, citation accuracy, source quality, and tool efficiency. LLM-as-judge provides scalable evaluation while human evaluation catches edge cases.\n\n**Performance Drivers: The 95% Finding**\n\nResearch on the BrowseComp evaluation found three factors explain 95% of performance variance:\n\n| Factor | Variance Explained | Implication |\n|--------|-------------------|-------------|\n| Token usage | 80% | More tokens = better performance |\n| Number of tool calls | ~10% | More exploration helps |\n| Model choice | ~5% | Better models multiply efficiency |\n\nEvaluate with realistic token constraints. Model upgrades beat token increases. Multi-agent architectures distribute work across separate context windows.\n\n## Evaluation Challenges\n\n**Non-Determinism and Multiple Valid Paths**\n\nAgents may take completely different valid paths to the same goal. Traditional step-by-step evaluation fails.\n\n**Solution**: Evaluate outcomes, not execution paths. Judge whether the agent achieves the right result through a reasonable process.\n\n**Context-Dependent Failures**\n\nFailures often emerge after extended interaction or at specific complexity levels.\n\n**Solution**: Cover a range of complexity levels. Test extended interactions, not just isolated queries.\n\n**Composite Quality Dimensions**\n\nAn agent might score high on accuracy but low in efficiency, or vice versa.\n\n**Solution**: Multi-dimensional rubrics with appropriate weighting per use case.\n\n## Evaluation Rubric Design\n\nDefault weights: instruction following (0.30), completeness (0.25), tool efficiency (0.20), reasoning (0.15), coherence (0.10). Convert assessments to 0.0–1.0 with weighting. Typical passing thresholds: 0.7 for general use, 0.85 for critical operations.\n\nFor per-criterion score bands and rubric generation patterns, see [`references\u002Fllm-judge-patterns.md`](references\u002Fllm-judge-patterns.md) (Pattern 1, Step 1).\n\n## Evaluation Methodologies\n\n### LLM-as-Judge\n\nScales to large test sets and provides consistent judgments. Design evaluation prompts that capture the dimensions of interest.\n\nFor the full evaluation prompt template, see [`references\u002Fllm-judge-patterns.md`](references\u002Fllm-judge-patterns.md) (Pattern 1, Step 3).\n\n**Chain-of-Thought Requirement**: Always require justification before the score — improves reliability by 15-25%.\n\n### Human Evaluation\n\nHuman evaluation catches what automation misses: hallucinated answers on unusual queries, subtle context misunderstandings, edge cases, qualitative issues with tone or approach. Sample systematically across complexity levels.\n\n### End-State Evaluation\n\nFor commands that produce artifacts, evaluate the final output rather than the process: does the code work? Is the configuration valid? Does the output meet requirements?\n\n## Test Set Design\n\n**Sample Selection**: Start small during development — early changes have large impact. Sample from real usage patterns. Add known edge cases. Ensure coverage across complexity levels.\n\n**Complexity Stratification**: simple (single tool call), medium (multiple tool calls), complex (many tool calls, ambiguity), very complex (extended interaction, deep reasoning).\n\n## Context Engineering Evaluation\n\n**Testing Prompt Variations**: Run current prompt as baseline, then modified prompt on same cases. Measure quality scores, token usage, efficiency. Identify which changes improved which dimensions.\n\n**Degradation Testing**: Run agents at different context sizes. Identify performance cliffs. Establish safe operating limits.\n\n## Evaluation Pitfalls\n\n**Scoring without justification** → Require evidence-based justification before every score.\n\n**Single-pass pairwise comparison** → Position bias corrupts results. Always swap positions and check consistency.\n\n**Overloaded criteria** → One criterion = one measurable aspect.\n\n**Missing edge case guidance** → Include edge cases in rubrics explicitly.\n\n**Ignoring confidence calibration** → Calibrate confidence to position consistency and evidence strength.\n\n## Decision Framework: Direct vs. Pairwise\n\n```\nIs there an objective ground truth?\n├── Yes → Direct Scoring\n│   └── Examples: factual accuracy, instruction following, format compliance\n└── No → Is it a preference or quality judgment?\n    ├── Yes → Pairwise Comparison\n    │   └── Examples: tone, style, persuasiveness, creativity\n    └── No → Reference-based evaluation\n        └── Examples: summarization, translation\n```\n\nFor pairwise: always run two passes with swapped positions. If passes disagree → TIE (position bias detected).\n\n## Guidelines\n\n1. Always require justification before scores — chain-of-thought improves reliability by 15-25%\n2. Always swap positions in pairwise comparison — single-pass is corrupted by position bias\n3. Match scale granularity to rubric specificity — don't use 1-10 without detailed levels\n4. Separate objective and subjective criteria — direct scoring for objective, pairwise for subjective\n5. Include confidence scores — calibrate to position consistency and evidence strength\n6. Define edge cases explicitly — ambiguous situations cause the most evaluation variance\n7. Use domain-specific rubrics — generic rubrics produce generic (less useful) evaluations\n8. Validate against human judgments — automated evaluation is only valuable if it correlates\n9. Monitor for systematic bias — track disagreement patterns by criterion and response type\n10. Design for iteration — evaluation systems improve with feedback loops\n\n## Example: Evaluating a Codex Command\n\n**Test Cases** for a `\u002Frefactor` command:\n1. Simple: Rename a variable across a single file\n2. Medium: Extract a function from existing code\n3. Complex: Refactor a class to use a new design pattern\n4. Very Complex: Restructure module dependencies\n\n**Evaluation Rubric**: Correctness (does it work?), Completeness (all instances updated?), Style (follows conventions?), Efficiency (unnecessary changes avoided?).\n\n**Iteration**: If evaluation reveals the command often misses instances, add the instruction: \"Search the entire codebase for all occurrences.\" Re-evaluate on same test cases. Compare completeness scores. Check correctness didn't regress.\n\n## Iterative Improvement Workflow\n\n1. **Identify weakness**: Use evaluation to find where agent struggles\n2. **Hypothesize cause**: Is it the prompt? The context? The examples?\n3. **Modify prompt**: Make targeted changes based on hypothesis\n4. **Re-evaluate**: Run same test cases with modified prompt\n5. **Compare**: Did the change improve the target dimension?\n6. **Check regression**: Did other dimensions suffer?\n7. **Iterate**: Repeat until quality meets threshold\n\n---\n\n## Testing & Validation\n\nAfter designing or running an evaluation:\n\n1. **Justification present** — every score has evidence-based justification written before it (not after)\n2. **Pairwise consistency** — if using pairwise: ran both position orders and checked for agreement\n3. **Rubric coverage** — all 5 criteria rated on every test case\n4. **Complexity coverage** — test cases span at least simple and complex levels\n5. **Baseline comparison** — before\u002Fafter improvement uses the same test set (not a new one)\n\n**Quality gates:**\n- [ ] Chain-of-thought justification required before every score\n- [ ] Pairwise comparisons run with swapped positions; disagreement → TIE\n- [ ] Rubric weights sum to 1.0 (0.30 + 0.25 + 0.20 + 0.15 + 0.10)\n- [ ] Test cases cover at least two complexity levels\n- [ ] Improvement measurement compares against the same baseline test set\n- [ ] At least one dimension validated against human judgment\n\n---\n\n## Reference Guide\n\n| Resource | Purpose |\n|---|---|\n| `references\u002Fbias-mitigation.md` | Position bias, length bias, self-enhancement, verbosity, authority bias — detection and mitigation |\n| `references\u002Fllm-judge-patterns.md` | Structured workflow, hierarchical, PoLL, confidence calibration, detailed prompt templates, meta-evaluation |\n| `references\u002Fmetric-selection.md` | Classification metrics (precision\u002Frecall\u002FF1), agreement (Cohen's κ), correlation (Spearman's ρ), decision tree |\n","# Evaluation Methods for Codex Agents\n\nEvaluation of agent systems requires outcome-focused approaches that account for non-determinism and multiple valid paths. Agents may reach goals through different routes — evaluation should judge whether they achieve right outcomes through reasonable processes, not match exact execution steps.\n\n## Quick Start\n\n1. **Identify what to evaluate** — a command's outputs, an agent's behavior, or a prompt variant\n2. **Design test cases** — cover at least two complexity levels: simple (single tool call) and complex (multi-step reasoning)\n3. **Choose method** — LLM-as-judge (scale + speed), human evaluation (catches edge cases), or end-state (artifact correctness)\n4. **Apply rubric** — instruction following (0.30), completeness (0.25), tool efficiency (0.20), reasoning (0.15), coherence (0.10)\n5. **Iterate** — compare before\u002Fafter scores on the **same** test set; check no dimension regressed\n\n## When to Use\n\n- Testing whether a command or skill produces consistently good outputs\n- Validating that context engineering changes improve (not regress) quality\n- Scoring agent outputs systematically with LLM-as-judge\n- Comparing prompt variants objectively before choosing one\n- Detecting or mitigating position bias, verbosity bias, or self-enhancement bias\n\n## When NOT to Use\n\n- Debugging why a single command invocation failed — use `Codex --debug` and troubleshooting instead\n- For one-off gut-check tests — evaluation is for systematic, repeatable quality measurement\n- For evaluating plugin structure or manifest correctness — use `plugin-development` or `plugin-validator` instead\n\n---\n\n## Core Concepts\n\nAgent evaluation captures multiple quality dimensions: factual accuracy, completeness, citation accuracy, source quality, and tool efficiency. LLM-as-judge provides scalable evaluation while human evaluation catches edge cases.\n\n**Performance Drivers: The 95% Finding**\n\nResearch on the BrowseComp evaluation found three factors explain 95% of performance variance:\n\n| Factor | Variance Explained | Implication |\n|--------|-------------------|-------------|\n| Token usage | 80% | More tokens = better performance |\n| Number of tool calls | ~10% | More exploration helps |\n| Model choice | ~5% | Better models multiply efficiency |\n\nEvaluate with realistic token constraints. Model upgrades beat token increases. Multi-agent architectures distribute work across separate context windows.\n\n## Evaluation Challenges\n\n**Non-Determinism and Multiple Valid Paths**\n\nAgents may take completely different valid paths to the same goal. Traditional step-by-step evaluation fails.\n\n**Solution**: Evaluate outcomes, not execution paths. Judge whether the agent achieves the right result through a reasonable process.\n\n**Context-Dependent Failures**\n\nFailures often emerge after extended interaction or at specific complexity levels.\n\n**Solution**: Cover a range of complexity levels. Test extended interactions, not just isolated queries.\n\n**Composite Quality Dimensions**\n\nAn agent might score high on accuracy but low in efficiency, or vice versa.\n\n**Solution**: Multi-dimensional rubrics with appropriate weighting per use case.\n\n## Evaluation Rubric Design\n\nDefault weights: instruction following (0.30), completeness (0.25), tool efficiency (0.20), reasoning (0.15), coherence (0.10). Convert assessments to 0.0–1.0 with weighting. Typical passing thresholds: 0.7 for general use, 0.85 for critical operations.\n\nFor per-criterion score bands and rubric generation patterns, see [`references\u002Fllm-judge-patterns.md`](references\u002Fllm-judge-patterns.md) (Pattern 1, Step 1).\n\n## Evaluation Methodologies\n\n### LLM-as-Judge\n\nScales to large test sets and provides consistent judgments. Design evaluation prompts that capture the dimensions of interest.\n\nFor the full evaluation prompt template, see [`references\u002Fllm-judge-patterns.md`](references\u002Fllm-judge-patterns.md) (Pattern 1, Step 3).\n\n**Chain-of-Thought Requirement**: Always require justification before the score — improves reliability by 15-25%.\n\n### Human Evaluation\n\nHuman evaluation catches what automation misses: hallucinated answers on unusual queries, subtle context misunderstandings, edge cases, qualitative issues with tone or approach. Sample systematically across complexity levels.\n\n### End-State Evaluation\n\nFor commands that produce artifacts, evaluate the final output rather than the process: does the code work? Is the configuration valid? Does the output meet requirements?\n\n## Test Set Design\n\n**Sample Selection**: Start small during development — early changes have large impact. Sample from real usage patterns. Add known edge cases. Ensure coverage across complexity levels.\n\n**Complexity Stratification**: simple (single tool call), medium (multiple tool calls), complex (many tool calls, ambiguity), very complex (extended interaction, deep reasoning).\n\n## Context Engineering Evaluation\n\n**Testing Prompt Variations**: Run current prompt as baseline, then modified prompt on same cases. Measure quality scores, token usage, efficiency. Identify which changes improved which dimensions.\n\n**Degradation Testing**: Run agents at different context sizes. Identify performance cliffs. Establish safe operating limits.\n\n## Evaluation Pitfalls\n\n**Scoring without justification** → Require evidence-based justification before every score.\n\n**Single-pass pairwise comparison** → Position bias corrupts results. Always swap positions and check consistency.\n\n**Overloaded criteria** → One criterion = one measurable aspect.\n\n**Missing edge case guidance** → Include edge cases in rubrics explicitly.\n\n**Ignoring confidence calibration** → Calibrate confidence to position consistency and evidence strength.\n\n## Decision Framework: Direct vs. Pairwise\n\n```\nIs there an objective ground truth?\n├── Yes → Direct Scoring\n│   └── Examples: factual accuracy, instruction following, format compliance\n└── No → Is it a preference or quality judgment?\n    ├── Yes → Pairwise Comparison\n    │   └── Examples: tone, style, persuasiveness, creativity\n    └── No → Reference-based evaluation\n        └── Examples: summarization, translation\n```\n\nFor pairwise: always run two passes with swapped positions. If passes disagree → TIE (position bias detected).\n\n## Guidelines\n\n1. Always require justification before scores — chain-of-thought improves reliability by 15-25%\n2. Always swap positions in pairwise comparison — single-pass is corrupted by position bias\n3. Match scale granularity to rubric specificity — don't use 1-10 without detailed levels\n4. Separate objective and subjective criteria — direct scoring for objective, pairwise for subjective\n5. Include confidence scores — calibrate to position consistency and evidence strength\n6. Define edge cases explicitly — ambiguous situations cause the most evaluation variance\n7. Use domain-specific rubrics — generic rubrics produce generic (less useful) evaluations\n8. Validate against human judgments — automated evaluation is only valuable if it correlates\n9. Monitor for systematic bias — track disagreement patterns by criterion and response type\n10. Design for iteration — evaluation systems improve with feedback loops\n\n## Example: Evaluating a Codex Command\n\n**Test Cases** for a `\u002Frefactor` command:\n1. Simple: Rename a variable across a single file\n2. Medium: Extract a function from existing code\n3. Complex: Refactor a class to use a new design pattern\n4. Very Complex: Restructure module dependencies\n\n**Evaluation Rubric**: Correctness (does it work?), Completeness (all instances updated?), Style (follows conventions?), Efficiency (unnecessary changes avoided?).\n\n**Iteration**: If evaluation reveals the command often misses instances, add the instruction: \"Search the entire codebase for all occurrences.\" Re-evaluate on same test cases. Compare completeness scores. Check correctness didn't regress.\n\n## Iterative Improvement Workflow\n\n1. **Identify weakness**: Use evaluation to find where agent struggles\n2. **Hypothesize cause**: Is it the prompt? The context? The examples?\n3. **Modify prompt**: Make targeted changes based on hypothesis\n4. **Re-evaluate**: Run same test cases with modified prompt\n5. **Compare**: Did the change improve the target dimension?\n6. **Check regression**: Did other dimensions suffer?\n7. **Iterate**: Repeat until quality meets threshold\n\n---\n\n## Testing & Validation\n\nAfter designing or running an evaluation:\n\n1. **Justification present** — every score has evidence-based justification written before it (not after)\n2. **Pairwise consistency** — if using pairwise: ran both position orders and checked for agreement\n3. **Rubric coverage** — all 5 criteria rated on every test case\n4. **Complexity coverage** — test cases span at least simple and complex levels\n5. **Baseline comparison** — before\u002Fafter improvement uses the same test set (not a new one)\n\n**Quality gates:**\n- [ ] Chain-of-thought justification required before every score\n- [ ] Pairwise comparisons run with swapped positions; disagreement → TIE\n- [ ] Rubric weights sum to 1.0 (0.30 + 0.25 + 0.20 + 0.15 + 0.10)\n- [ ] Test cases cover at least two complexity levels\n- [ ] Improvement measurement compares against the same baseline test set\n- [ ] At least one dimension validated against human judgment\n\n---\n\n## Reference Guide\n\n| Resource | Purpose |\n|---|---|\n| `references\u002Fbias-mitigation.md` | Position bias, length bias, self-enhancement, verbosity, authority bias — detection and mitigation |\n| `references\u002Fllm-judge-patterns.md` | Structured workflow, hierarchical, PoLL, confidence calibration, detailed prompt templates, meta-evaluation |\n| `references\u002Fmetric-selection.md` | Classification metrics (precision\u002Frecall\u002FF1), agreement (Cohen's κ), correlation (Spearman's ρ), decision tree |\n",{"name":5,"description":7,"allowed-tools":23},"Read Glob Grep","main","1bf12900cac6f28e346abd2912bdb78d306de723",".agents\u002Fskills\u002Fplugin-evaluation","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace","https:\u002F\u002Fgithub.com\u002FAndreHahm","2026-11-15 04:26:16",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add AndreHahm\u002Fandres-cc-marketplace --skill plugin-evaluation",{"tool":36,"command":37},"bunx","bunx skills add AndreHahm\u002Fandres-cc-marketplace --skill plugin-evaluation",{"tool":39,"command":40},"pnpm","pnpm dlx skills add AndreHahm\u002Fandres-cc-marketplace --skill plugin-evaluation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fplugin-evaluation\u002FSKILL.md",true,[44,50,54],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"references\u002Fbias-mitigation.md","REFERENCE","参考文档",8774,"https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fplugin-evaluation\u002Freferences\u002Fbias-mitigation.md",{"path":51,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Fllm-judge-patterns.md",24811,"https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fplugin-evaluation\u002Freferences\u002Fllm-judge-patterns.md",{"path":55,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":56,"sourceHtmlUrl":57},"references\u002Fmetric-selection.md",6165,"https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fplugin-evaluation\u002Freferences\u002Fmetric-selection.md",[59,66,73,80,87,94],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":18},"01M79R0K918Z0XDM74R0EY5H3P","demo","Fixture skill used by scripts.marketplace_ci sync tests.","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Ftests\u002Fmarketplace_ci\u002Ffixtures\u002Fplugins\u002Fsample-kit\u002Fskills\u002Fdemo\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":18},"01M79R08VMDJA2FPJKH3ZM309K","export-demo","Fixture skill used by scripts.marketplace_ci export tests.","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Ftests\u002Fmarketplace_ci\u002Ffixtures\u002Fclaude\u002Fskills\u002Fexport-demo\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M79R042FS581DS2JNW5AZ1MK","work-to-development","Validate an accepted Linear Issue's readiness for implementation (criteria, dependencies, duplicate\u002Fexisting artifacts) and request governed branch\u002Fworktree creation through git-kit:starting-work — recording work-started Git\u002FGitHub evidence and a deliberate Linear Started transition only after Git identity is read back and verified. Use when asked to start work on a Linear issue, begin development for an accepted issue, or prepare an accepted issue for implementation. Never creates a branch directly — always delegates to git-kit.","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fworkmanagement-kit\u002Fskills\u002Fwork-to-development\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M79QZZD4EHB82P72P4N92BKT","work-linking","Create and maintain stable cross-system links and authority labels between Notion and Linear records, and detect\u002Frepair drift between them. Use when asked to link a Notion record to a Linear record, check for drift between Notion and Linear, or repair a broken\u002Fstale link. Repairs only the bounded non-authoritative copy or link — never chooses by newest timestamp, and never creates a reverse-write loop against either system's own authority.","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fworkmanagement-kit\u002Fskills\u002Fwork-linking\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":18},"01M79QZVFA3GTVNZMJ51NESVSP","status-and-learning","Produce a dated, explicitly non-live Linear-to-Notion progress summary, or capture an outcome\u002Flearning after work completes. Use when asked to summarize progress to Notion, report portfolio status, or capture what was learned from a completed piece of work — not to disposition open follow-up items (see open-item-management for that). Never runs automatically or continuously — every summary is a deliberate, approved snapshot, not a live sync.","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fworkmanagement-kit\u002Fskills\u002Fstatus-and-learning\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":18},"01M79QZNSKSABK7X2PT93X8QT5","repository-gates","Discover, delegate, record, and invalidate this repository's actual Git\u002FGitHub lifecycle gates (pre-commit, pre-push, PR required checks, Codex delta review, review requirements, merge rights) and the repository-policy provider profile that maps governed operations to git-kit. Use when asked what gates apply to this repository, to discover repository policy before starting Wave 2 work, or to check whether prior gate evidence is still valid after new commits. Discovery and delegation only — never executes a gate's own internal logic, and never invents a gate the repository doesn't actually define (in particular, never assumes a \"Review Changes\" gate exists).","https:\u002F\u002Fgithub.com\u002FAndreHahm\u002Fandres-cc-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fworkmanagement-kit\u002Fskills\u002Frepository-gates\u002FSKILL.md",[],[]]