[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6EXY4HBS7FREWTEQCW2AJ6B":3,"skill-same:01M6EXY4HBS7FREWTEQCW2AJ6B":88},{"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":23,"publishedAt":48,"linkOnly":49,"markdownContent":50,"markdownBody":51,"frontMatter":52,"refName":54,"commitSha":55,"skillDirectory":56,"forksCount":57,"repositoryHtmlUrl":58,"repositoryHomepageUrl":59,"ownerHtmlUrl":60,"ownerBlogUrl":61,"ownerDescription":62,"updatedAt":63,"install":64,"resources":77},"01M6EXY4HBS7FREWTEQCW2AJ6B","embedding-anisotropy-check",null,"Use before reporting any cosine similarity computed over embeddings or transformer hidden states, whether the agent computed the vectors or loaded them. Measures the anisotropy of each side, centres by the population mean, and scores matched pairs against a permuted floor taken from the same comparison. Use when comparing one model's embeddings across precisions, runtimes or layers, when evaluating retrieval, probes or adapters, or before calling a similarity high, two conditions in agreement, or a small gap a null.","github\u002Fawesome-copilot","github","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9919?v=4","01M6AC83DS133TPXFSP1DHNCC6","GitHub","01M6AD3PBDYPK7NVFJNJ66Y8RS",41046,"MIT","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fembedding-anisotropy-check\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47],"check","embedding anisotropy check","embedding","retrieval","agreement","when","layers","high","with","states","comparison","runtimes","Transformer","python","similarity","gap","vectors","across","numpy","calling","use","model","reporting","embeddings","2026-11-10 10:12:30",false,"---\nname: embedding-anisotropy-check\ndescription: 'Use before reporting any cosine similarity computed over embeddings or transformer hidden states, whether the agent computed the vectors or loaded them. Measures the anisotropy of each side, centres by the population mean, and scores matched pairs against a permuted floor taken from the same comparison. Use when comparing one model''s embeddings across precisions, runtimes or layers, when evaluating retrieval, probes or adapters, or before calling a similarity high, two conditions in agreement, or a small gap a null.'\nlicense: MIT\ncompatibility: Python 3.9+ with numpy\n---\n\n# Embedding anisotropy check\n\nRaw cosine between embeddings is rarely centred on zero. Most sentence encoders and most hidden-state layers place\nevery vector near one shared direction, so two unrelated inputs can score 0.6, 0.9 or 0.99. A matched pair at 0.9 is\nthen not evidence of anything by itself, and a gap of 0.01 between two conditions can be the whole signal or none of\nit. This skill makes every such comparison carry its own floor.\n\n## How an agent reaches this skill\n\nThe agent does not vectorise anything here. The skill applies at the point where the agent is about to report a cosine\nnumber, whoever produced the vectors:\n\n- **The agent wrote the evaluation.** A request such as \"write a retrieval eval for our index\" or \"check that the 4-bit\n  export of this encoder still matches bf16\" ends in a script that prints a mean cosine. Before the agent says\n  \"0.91, the export is fine\", it runs the same model's unrelated pairs through the script and reports the floor beside\n  the matched figure.\n- **The vectors already exist in the workspace.** An `embeddings.npy`, a probe or adapter evaluated on saved hidden\n  states, a vector-store dump. Asked \"are these two layers encoding the same thing\" or \"did the fine-tune move the\n  representations\", the agent loads the arrays and runs the check before answering.\n- **A claim is being written.** The agent is drafting a README, a PR description or a results table that calls a\n  similarity high, says two conditions agree, or calls a small difference a null. The number in that sentence is the\n  trigger.\n\n## When to use this skill\n\n- The same inputs under one model at two precisions (bf16 against 4-bit), in two runtimes, or at two nearby layers,\n  which share one coordinate space.\n- Query and passage, image and caption, or any other paired evaluation inside one embedding space.\n- Any sentence claiming that a similarity is high, that two conditions agree, or that a small difference is a null.\n\nNot for two independently trained models as they stand. A rotation of one model's space keeps its geometry and changes\nevery cosine against the other model, and centring does not align the two bases. Fit an explicit map first, for\nexample orthogonal Procrustes on a train split of paired rows, and compare the mapped rows; or compare each model's\nown within-space similarities instead.\n\n## Procedure\n\n1. Save the compared rows as two `.npy` arrays of shape `(n, d)`, row `i` of A paired with row `i` of B.\n2. Choose where the means come from. The best source is a train split from the same domain, model and modality as the\n   compared rows. A mean fitted on another domain mis-centres: it can leave part of the shared direction in place,\n   remove too much of it, or add a direction of its own, and what it leaves then looks like an effect of whatever\n   differs between the domains.\n3. Run the bundled script:\n\n   ```bash\n   python scripts\u002Fanisotropy_check.py A.npy B.npy --fit-a trainA.npy --fit-b trainB.npy\n   ```\n\n   Pass `--shared` when A and B come from one model and one population, so that both sides are centred by one mean.\n   Without `--fit-a` and `--fit-b` the means are fitted on the compared rows themselves, and the output's `means`\n   field says, for each side, which rows its mean came from.\n4. Report, side by side: the anisotropy of each side, matched against floor before centring, matched against floor\n   after centring, and `partner_ranked_first` against its chance rate `1\u002Fn`.\n\n## Reading the output\n\n- `anisotropy` is the mean cosine between different rows of one side. For large `n` it is close to the squared length\n  of the mean of the unit rows, so it measures how much the rows share one common direction, which is what centring\n  removes. It does not detect every departure from isotropy: rows spread along one axis in both directions score near\n  0 and still give raw cosines of +1 and -1. A value near 0 says there is little within-side mean direction to remove,\n  not that raw cosine can be read without its floor. Above about 0.3, rows within that side cluster strongly around a\n  mean direction (the script adds a warning); check `raw.floor` to see how much of that direction is shared across\n  sides, and report both the raw and centred matched-versus-floor readings (centring changes the scoring function and\n  does not replace evaluating an index that searches by raw cosine).\n- `raw.floor` is what an unrelated pair scores in this comparison. A matched cosine means something only through its\n  distance from this floor.\n- `centred.gap_over_floor_sd` and `centred.cohens_d` say whether matched pairs separate from unrelated ones at all.\n  When unrelated pairs all score the same, the floor has no spread and `gap_over_floor_sd` is `null`: the gap cannot\n  be expressed in floor standard deviations, and the plain `gap` is the figure to read.\n- `partner_ranked_first` is the retrieval view of the same pairs. Centring can move a ranking either way, so report it\n  both ways rather than assuming it transfers.\n\n## Rules that go with the numbers\n\n- Average the signed difference across seeds or items, then take its absolute value. Averaging absolute differences\n  is biased upward, and the bias is greatest when the gap is small relative to its noise.\n- Give a gap with its spread and the smallest effect the test could have detected. An underpowered comparison is not\n  a refutation, and the report should say which of the two it is.\n- Put the floor from the same comparison beside every result. A floor copied from another run, model or domain does\n  not correct this one.\n\n## Example\n\nbge-small-en-v1.5 on the first 100 documents of `mteb\u002Fscifact`'s corpus, each split into its first and second half at\nthe middle sentence boundary, with the halves of one abstract as the matched pair and one mean shared by both sides\n(`--shared`). `scripts\u002Fexample_scifact_halves.py` rebuilds it: it pins the corpus at revision `cf10ab6` and the model\nat `5c38ec7`, states the sentence split, and needs `datasets` and `sentence-transformers` besides numpy. With datasets\n5.0.1, sentence-transformers 6.0.1, torch 2.13.0 and numpy 2.5.1 it gives (rounded to four decimal places):\n\n| | matched | floor (unrelated halves) | gap over floor sd |\n| --- | ---: | ---: | ---: |\n| raw cosine | 0.8636 | 0.6024 | 4.38 |\n| centred cosine | 0.6487 | -0.0131 | 5.87 |\n\nThe anisotropy is 0.5956 and 0.6148, and here the two sides share one space, so before centring an unrelated half\nalready scores 70% of what the matching half scores. Both spaces separate matched from unrelated halves (4.38 floor sd\nraw and 5.87 centred) and put the partner first for most abstracts (97 and 98 of 100), while centring removes the\n0.6024 baseline offset.\n","# Embedding anisotropy check\n\nRaw cosine between embeddings is rarely centred on zero. Most sentence encoders and most hidden-state layers place\nevery vector near one shared direction, so two unrelated inputs can score 0.6, 0.9 or 0.99. A matched pair at 0.9 is\nthen not evidence of anything by itself, and a gap of 0.01 between two conditions can be the whole signal or none of\nit. This skill makes every such comparison carry its own floor.\n\n## How an agent reaches this skill\n\nThe agent does not vectorise anything here. The skill applies at the point where the agent is about to report a cosine\nnumber, whoever produced the vectors:\n\n- **The agent wrote the evaluation.** A request such as \"write a retrieval eval for our index\" or \"check that the 4-bit\n  export of this encoder still matches bf16\" ends in a script that prints a mean cosine. Before the agent says\n  \"0.91, the export is fine\", it runs the same model's unrelated pairs through the script and reports the floor beside\n  the matched figure.\n- **The vectors already exist in the workspace.** An `embeddings.npy`, a probe or adapter evaluated on saved hidden\n  states, a vector-store dump. Asked \"are these two layers encoding the same thing\" or \"did the fine-tune move the\n  representations\", the agent loads the arrays and runs the check before answering.\n- **A claim is being written.** The agent is drafting a README, a PR description or a results table that calls a\n  similarity high, says two conditions agree, or calls a small difference a null. The number in that sentence is the\n  trigger.\n\n## When to use this skill\n\n- The same inputs under one model at two precisions (bf16 against 4-bit), in two runtimes, or at two nearby layers,\n  which share one coordinate space.\n- Query and passage, image and caption, or any other paired evaluation inside one embedding space.\n- Any sentence claiming that a similarity is high, that two conditions agree, or that a small difference is a null.\n\nNot for two independently trained models as they stand. A rotation of one model's space keeps its geometry and changes\nevery cosine against the other model, and centring does not align the two bases. Fit an explicit map first, for\nexample orthogonal Procrustes on a train split of paired rows, and compare the mapped rows; or compare each model's\nown within-space similarities instead.\n\n## Procedure\n\n1. Save the compared rows as two `.npy` arrays of shape `(n, d)`, row `i` of A paired with row `i` of B.\n2. Choose where the means come from. The best source is a train split from the same domain, model and modality as the\n   compared rows. A mean fitted on another domain mis-centres: it can leave part of the shared direction in place,\n   remove too much of it, or add a direction of its own, and what it leaves then looks like an effect of whatever\n   differs between the domains.\n3. Run the bundled script:\n\n   ```bash\n   python scripts\u002Fanisotropy_check.py A.npy B.npy --fit-a trainA.npy --fit-b trainB.npy\n   ```\n\n   Pass `--shared` when A and B come from one model and one population, so that both sides are centred by one mean.\n   Without `--fit-a` and `--fit-b` the means are fitted on the compared rows themselves, and the output's `means`\n   field says, for each side, which rows its mean came from.\n4. Report, side by side: the anisotropy of each side, matched against floor before centring, matched against floor\n   after centring, and `partner_ranked_first` against its chance rate `1\u002Fn`.\n\n## Reading the output\n\n- `anisotropy` is the mean cosine between different rows of one side. For large `n` it is close to the squared length\n  of the mean of the unit rows, so it measures how much the rows share one common direction, which is what centring\n  removes. It does not detect every departure from isotropy: rows spread along one axis in both directions score near\n  0 and still give raw cosines of +1 and -1. A value near 0 says there is little within-side mean direction to remove,\n  not that raw cosine can be read without its floor. Above about 0.3, rows within that side cluster strongly around a\n  mean direction (the script adds a warning); check `raw.floor` to see how much of that direction is shared across\n  sides, and report both the raw and centred matched-versus-floor readings (centring changes the scoring function and\n  does not replace evaluating an index that searches by raw cosine).\n- `raw.floor` is what an unrelated pair scores in this comparison. A matched cosine means something only through its\n  distance from this floor.\n- `centred.gap_over_floor_sd` and `centred.cohens_d` say whether matched pairs separate from unrelated ones at all.\n  When unrelated pairs all score the same, the floor has no spread and `gap_over_floor_sd` is `null`: the gap cannot\n  be expressed in floor standard deviations, and the plain `gap` is the figure to read.\n- `partner_ranked_first` is the retrieval view of the same pairs. Centring can move a ranking either way, so report it\n  both ways rather than assuming it transfers.\n\n## Rules that go with the numbers\n\n- Average the signed difference across seeds or items, then take its absolute value. Averaging absolute differences\n  is biased upward, and the bias is greatest when the gap is small relative to its noise.\n- Give a gap with its spread and the smallest effect the test could have detected. An underpowered comparison is not\n  a refutation, and the report should say which of the two it is.\n- Put the floor from the same comparison beside every result. A floor copied from another run, model or domain does\n  not correct this one.\n\n## Example\n\nbge-small-en-v1.5 on the first 100 documents of `mteb\u002Fscifact`'s corpus, each split into its first and second half at\nthe middle sentence boundary, with the halves of one abstract as the matched pair and one mean shared by both sides\n(`--shared`). `scripts\u002Fexample_scifact_halves.py` rebuilds it: it pins the corpus at revision `cf10ab6` and the model\nat `5c38ec7`, states the sentence split, and needs `datasets` and `sentence-transformers` besides numpy. With datasets\n5.0.1, sentence-transformers 6.0.1, torch 2.13.0 and numpy 2.5.1 it gives (rounded to four decimal places):\n\n| | matched | floor (unrelated halves) | gap over floor sd |\n| --- | ---: | ---: | ---: |\n| raw cosine | 0.8636 | 0.6024 | 4.38 |\n| centred cosine | 0.6487 | -0.0131 | 5.87 |\n\nThe anisotropy is 0.5956 and 0.6148, and here the two sides share one space, so before centring an unrelated half\nalready scores 70% of what the matching half scores. Both spaces separate matched from unrelated halves (4.38 floor sd\nraw and 5.87 centred) and put the partner first for most abstracts (97 and 98 of 100), while centring removes the\n0.6024 baseline offset.\n",{"name":5,"license":15,"description":7,"compatibility":53},"Python 3.9+ with numpy","main","61fa1fa8fcc63d6dbd2f07a75aeb8e214df5acfb","skills\u002Fembedding-anisotropy-check",5340,"https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot","https:\u002F\u002Fawesome-copilot.github.com\u002F","https:\u002F\u002Fgithub.com\u002Fgithub","https:\u002F\u002Fgithub.com\u002Fabout","How people build software.","2026-11-11 13:41:32",{"source":8,"skillName":5,"commands":65,"prompt":75,"zipAvailable":76},[66,69,72],{"tool":67,"command":68},"npx","npx skills add github\u002Fawesome-copilot --skill embedding-anisotropy-check",{"tool":70,"command":71},"bunx","bunx skills add github\u002Fawesome-copilot --skill embedding-anisotropy-check",{"tool":73,"command":74},"pnpm","pnpm dlx skills add github\u002Fawesome-copilot --skill embedding-anisotropy-check","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fembedding-anisotropy-check\u002FSKILL.md",true,[78,84],{"path":79,"resourceType":80,"resourceTypeDesc":81,"sizeBytes":82,"sourceHtmlUrl":83},"scripts\u002Fanisotropy_check.py","SCRIPT","脚本",7626,"https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fembedding-anisotropy-check\u002Fscripts\u002Fanisotropy_check.py",{"path":85,"resourceType":80,"resourceTypeDesc":81,"sizeBytes":86,"sourceHtmlUrl":87},"scripts\u002Fexample_scifact_halves.py",2494,"https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fembedding-anisotropy-check\u002Fscripts\u002Fexample_scifact_halves.py",[89,117,144,174,208,225],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":99,"publishedAt":48},"01M70KXZNX9ES38N2A8M58ZZMZ","supabase-rls-policies","Write, review, and test Supabase Row Level Security (RLS) policies so that one user cannot read or change another user's rows. Use when creating or changing tables, policies, views, or security definer functions in a Supabase project, when reviewing a migration that touches RLS, or when asked to prove user isolation with pgTAP tests that sign in as two users.","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fsupabase-rls-policies\u002FSKILL.md",[95,98],{"categoryNo":96,"categoryName":97,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[100,101,102,103,104,29,105,106,107,108,109,110,111,112,113,114,44,115,116,32],"supabase","rls","supabase rls","supabase rls policies","project","review","change","read","write","security","isolation","tables","as","tests","prove","users","functions",{"skillNo":118,"name":119,"nameZh":6,"description":120,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":121,"categories":122,"keywords":124,"publishedAt":48},"01M70KWA5YWXDV6V8DZ19E3T2Y","skill-distribution-audit","Audit a skill collection before release or investigate installer omissions, archived\u002Ftemplate leakage, duplicate identities, and manifest drift. Reconcile the intended public source set, distribution manifest, and actual discovery output for a specified consumer version. Includes a deterministic set-comparison helper.","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fskill-distribution-audit\u002FSKILL.md",[123],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[125,126,127,128,129,130,131,132,34,133,134,135,136,137,138,139,140,141,142,143],"distribution","skill distribution audit","audit","SET","deterministic","installer","for","discovery","helper","manifest","version","collection","drift","output","investigate","duplicate","template","release","reconcile",{"skillNo":145,"name":146,"nameZh":6,"description":147,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":148,"sourceHtmlUrl":149,"categories":150,"keywords":156,"publishedAt":48},"01M70KCJPFVW7PPQ49CC5R8JGK","logicapps-custom-code-net10-migration","Migrate Azure Logic Apps Standard local-function custom-code projects from .NET 8 to .NET 10. Use for legacy Logic Apps custom-code workspace upgrades, project\u002Fpackage configuration, custom-worker settings, publication, and integrated VS Code debugging. Not for ordinary .NET applications, standalone Azure Functions, Consumption workflows, or cloud deployment.","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Flogicapps-custom-code-net10-migration\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Flogicapps-custom-code-net10-migration\u002FSKILL.md",[151],{"categoryNo":152,"categoryName":153,"taxonomyType":154,"taxonomyTypeDesc":155},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[157,158,159,160,161,162,131,104,163,164,165,166,167,168,169,44,170,171,172,173],"custom","code","logicapps custom code net10 migration","migration","projects","standalone","deployment","applications","local","azure","worker","upgrades","package","apps","Legacy","debugging","workflows",{"skillNo":175,"name":176,"nameZh":6,"description":177,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":178,"sourceHtmlUrl":179,"categories":180,"keywords":185,"publishedAt":48},"01M6AEBN37RQMXJ3298PEY7PG1","winmd-api-search","Find and explore Windows desktop APIs. Use when building features that need platform capabilities — camera, file access, notifications, UI controls, AI\u002FML, sensors, networking, etc. Discovers the right API for a task and retrieves full type details (methods, properties, events, enumeration values).","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fwinmd-api-search\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fwinmd-api-search\u002FSKILL.md",[181,184],{"categoryNo":182,"categoryName":183,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000010","后端与接口",{"categoryNo":152,"categoryName":153,"taxonomyType":154,"taxonomyTypeDesc":155},[186,187,188,189,190,191,192,193,194,195,196,197,198,199,200,29,201,202,203,204,205,131,206,207],"winmd api search","search","api","windows","desktop","file access","enumeration","ui","ml","file","networking","capabilities","notifications","explore","access","features","full","camera","events","platform","methods","controls",{"skillNo":209,"name":210,"nameZh":6,"description":211,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":212,"licenseFileUrl":213,"sourceHtmlUrl":214,"categories":215,"keywords":221,"publishedAt":48},"01M6AE571R3SC1ZGE202B3NB5S","resemble-detect","Deepfake detection and media safety — detect AI-generated audio, images, video, and text, trace synthesis sources, apply watermarks, verify speaker identity, and analyze media intelligence using Resemble AI","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fresemble-detect\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fresemble-detect\u002FSKILL.md",[216,217,220],{"categoryNo":182,"categoryName":183,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":218,"categoryName":219,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000016","图像与多媒体",{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[222,128,195,223,224],"resemble detect","https","identity",{"skillNo":226,"name":227,"nameZh":6,"description":228,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":212,"licenseFileUrl":229,"sourceHtmlUrl":230,"categories":231,"keywords":236,"publishedAt":48},"01M6AE30BSK2HCY4TQQ22Z8JRD","quality-playbook","Run a complete quality engineering audit on any codebase. Derives behavioral requirements from the code, generates spec-traced functional tests, runs a three-pass code review with regression tests, executes a multi-model spec audit (Council of Three), and produces a consolidated bug report with TDD-verified patches. Finds the 35% of real defects that structural code review alone cannot catch. Works with any language. Trigger on 'quality playbook', 'spec audit', 'Council of Three', 'fitness-to-purpose', or 'coverage theater'.","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fquality-playbook\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fgithub\u002Fawesome-copilot\u002Fblob\u002Fmain\u002Fskills\u002Fquality-playbook\u002FSKILL.md",[232,235],{"categoryNo":233,"categoryName":234,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000007","代码审查",{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[237,238,239,240,158,105,9,127,241,242,223,243,244,245,246,247,248,249,250,251,252,253,113,254,45,255,256],"playbook","quality playbook","quality","code review","tdd","run","report","spec","bug report","council","engineering","functional","spec audit","bug","verified","multi-model","three","coverage","regression","structural"]