[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78Y19RTQ0EAM8HHJM2X2HG3":3,"skill-same:01M78Y19RTQ0EAM8HHJM2X2HG3":75},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M78Y19RTQ0EAM8HHJM2X2HG3","cpp-header-api",null,"Operate hnswlib's C++11 header-only API for HNSW and brute-force indexes, filters, persistence, deletion\u002Freplacement, stop conditions, multivector search, and bounded concurrency.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: cpp-header-api\ndescription: \"Operate hnswlib's C++11 header-only API for HNSW and brute-force\n  indexes, filters, persistence, deletion\u002Freplacement, stop conditions,\n  multivector search, and bounded concurrency.\"\nmetadata:\n  disco-role: operating\ndisable-model-invocation: true\nlicense: Apache 2.0\n---\n\n# C++ header API\n\nUse this sub-skill when the task is about the C++ or header-only interface: including\n`hnswlib\u002Fhnswlib.h`, constructing `HierarchicalNSW`, `searchKnn`,\n`searchKnnCloserFirst`, `BaseFilterFunctor`, `markDelete`, `addPoint`,\n`saveIndex`, CMake, epsilon search, multivector search, or C++ concurrency.\nThis skill covers the 0.9.0 header API and the installed CPU-native extension\nfacts; it does not expand the Python binding surface.\n\n## Operating contract\n\n- Require an include root containing the public `hnswlib\u002F` headers. Do not assume\n  a particular checkout layout, build directory, virtual environment, or dataset.\n- Treat the library as header-only. A C++11 compiler compiles the implementation\n  into the client; there is no hnswlib binary or CUDA API to link against.\n- Keep the `SpaceInterface` object alive until every index using it has been\n  destroyed. The index retains its distance function and parameter pointer.\n- Use `float` with `L2Space` or `InnerProductSpace` for the normal vector API.\n  Read [api-reference.md](references\u002Fapi-reference.md) before selecting an\n  optional stop-condition or multivector signature.\n\n## Standard workflow\n\n1. **Normalize the request.** Identify dimension, metric, capacity, whether labels\n   can be deleted\u002Freplaced, query ordering, filter requirements, and whether the\n   request is optional epsilon or multivector search.\n2. **Check the build boundary.** Include `\u003Chnswlib\u002Fhnswlib.h>`, compile as C++11,\n   add the include root, and use `-pthread` (or the platform equivalent) for code\n   using the library's mutexes\u002Fthreads. Use [workflows.md](references\u002Fworkflows.md)\n   for direct compiler and CMake patterns.\n3. **Create the space before the index.** Construct `L2Space(dim)` for squared\n   Euclidean distance or `InnerProductSpace(dim)` for `1 - inner product`.\n   Construct `HierarchicalNSW\u003Cfloat>` with capacity, `M`, `ef_construction`,\n   seed, and replacement policy. Prefer stack ownership or an RAII owner.\n4. **Populate deliberately.** Pass a contiguous vector pointer and a stable,\n   unique external label to `addPoint`. A repeated label updates its existing\n   item. Increase capacity with `resizeIndex` before inserting, or use the\n   persistence-based capacity extension described in the references.\n5. **Query with the intended ordering.** `searchKnn` returns a priority queue\n   whose `top()` is the farthest member of the retained result set. Drain and\n   reverse it, or call `searchKnnCloserFirst` for a vector ordered closest first.\n   A filter is a live pointer to a functor that accepts external labels.\n6. **Apply mutation preconditions.** `markDelete` only marks; it does not remove\n   graph connections. Replacement requires construction\u002Fload with\n   `allow_replace_deleted=true` and `addPoint(..., true)`. Do not call\n   `unmarkDelete` as a recovery mechanism while replacement is enabled.\n7. **Persist and reload safely.** Call `saveIndex(file)`, destroy the index if\n   desired, then construct the load form with the same compatible space. Set\n   `ef` again after loading: it is not saved and loading initializes it to 10.\n8. **Verify a small invariant.** Assert result count\u002Forder, filter acceptance,\n   and a post-load query. Run the bundled smoke wrapper with an explicit include\n   root; do not use a large stress or benchmark as the basic API gate.\n9. **Escalate optional APIs.** For epsilon or multivector work, use the exact\n   declarations and data layout in [api-reference.md](references\u002Fapi-reference.md)\n   and the recipes in [workflows.md](references\u002Fworkflows.md). Do not infer a\n   Python equivalent or invent a convenience overload.\n10. **Bound concurrency.** Concurrent read-only searches are the normal safe\n    pattern; concurrent insertion\u002Fupdate calls are demonstrated and internally\n    label-serialized. Coordinate structural operations, persistence, resizing,\n    destruction, and mixed read\u002Fwrite phases externally. See\n    [concurrency.md](references\u002Fconcurrency.md).\n\n## API routing map\n\n- Metric and memory layout: [api-reference.md](references\u002Fapi-reference.md).\n- Construction, query, filter, persistence, and mutation recipes:\n  [workflows.md](references\u002Fworkflows.md).\n- Threads, compiler flags, OpenMP, native SIMD flags, and ownership:\n  [concurrency.md](references\u002Fconcurrency.md).\n- Failure diagnosis, including low `ef`, capacity, paths, deletion policy, and\n  benchmark exclusions: [troubleshooting.md](references\u002Ftroubleshooting.md).\n- Executable smoke check: `scripts\u002Frun_cpp_smoke.sh --include-dir \u003Cinclude-root>`.\n\n## Hard guardrails\n\n- Do not claim CUDA support, GPU kernels, a compiled library ABI, or a required\n  OpenMP dependency. The exposed C++ API is CPU-native and header-only.\n- Do not treat a priority-queue pop sequence as closest-first. The queue's\n  further-first behavior is intentional; the vector helper reverses it.\n- Do not pass `replace_deleted=true` unless the index was created or loaded with\n  replacement enabled. Do not use a filter to make an unavailable `k` appear\n  available: filtering can legitimately return fewer than `k` items.\n- Do not let a temporary query buffer, metric space, filter, stop condition, or\n  data backing an in-flight operation go out of scope.\n- Do not use the smoke result to make recall, latency, SIMD portability, or\n  multi-thread stress claims. Those require a separately chosen workload and\n  compatible hardware.\n\n## Completion checklist\n\nBefore handing off a C++ solution, confirm:\n\n- the include root and compiler standard are explicit;\n- the space outlives the index and matches the data dimension\u002Fmetric;\n- `M`, `ef_construction`, capacity, and query `ef` have an intentional value;\n- queue versus closer-first semantics are documented at the call site;\n- filters inspect external labels and tolerate fewer than `k` matches;\n- deletion\u002Freplacement flags match the requested lifecycle;\n- persistence reloads with a compatible space and resets `ef` explicitly;\n- optional stop-condition signatures were copied from the bundled reference;\n- no operation is concurrently resizing, saving, loading, or destroying the index;\n- the smoke program compiles and passes with assertions.\n","# C++ header API\n\nUse this sub-skill when the task is about the C++ or header-only interface: including\n`hnswlib\u002Fhnswlib.h`, constructing `HierarchicalNSW`, `searchKnn`,\n`searchKnnCloserFirst`, `BaseFilterFunctor`, `markDelete`, `addPoint`,\n`saveIndex`, CMake, epsilon search, multivector search, or C++ concurrency.\nThis skill covers the 0.9.0 header API and the installed CPU-native extension\nfacts; it does not expand the Python binding surface.\n\n## Operating contract\n\n- Require an include root containing the public `hnswlib\u002F` headers. Do not assume\n  a particular checkout layout, build directory, virtual environment, or dataset.\n- Treat the library as header-only. A C++11 compiler compiles the implementation\n  into the client; there is no hnswlib binary or CUDA API to link against.\n- Keep the `SpaceInterface` object alive until every index using it has been\n  destroyed. The index retains its distance function and parameter pointer.\n- Use `float` with `L2Space` or `InnerProductSpace` for the normal vector API.\n  Read [api-reference.md](references\u002Fapi-reference.md) before selecting an\n  optional stop-condition or multivector signature.\n\n## Standard workflow\n\n1. **Normalize the request.** Identify dimension, metric, capacity, whether labels\n   can be deleted\u002Freplaced, query ordering, filter requirements, and whether the\n   request is optional epsilon or multivector search.\n2. **Check the build boundary.** Include `\u003Chnswlib\u002Fhnswlib.h>`, compile as C++11,\n   add the include root, and use `-pthread` (or the platform equivalent) for code\n   using the library's mutexes\u002Fthreads. Use [workflows.md](references\u002Fworkflows.md)\n   for direct compiler and CMake patterns.\n3. **Create the space before the index.** Construct `L2Space(dim)` for squared\n   Euclidean distance or `InnerProductSpace(dim)` for `1 - inner product`.\n   Construct `HierarchicalNSW\u003Cfloat>` with capacity, `M`, `ef_construction`,\n   seed, and replacement policy. Prefer stack ownership or an RAII owner.\n4. **Populate deliberately.** Pass a contiguous vector pointer and a stable,\n   unique external label to `addPoint`. A repeated label updates its existing\n   item. Increase capacity with `resizeIndex` before inserting, or use the\n   persistence-based capacity extension described in the references.\n5. **Query with the intended ordering.** `searchKnn` returns a priority queue\n   whose `top()` is the farthest member of the retained result set. Drain and\n   reverse it, or call `searchKnnCloserFirst` for a vector ordered closest first.\n   A filter is a live pointer to a functor that accepts external labels.\n6. **Apply mutation preconditions.** `markDelete` only marks; it does not remove\n   graph connections. Replacement requires construction\u002Fload with\n   `allow_replace_deleted=true` and `addPoint(..., true)`. Do not call\n   `unmarkDelete` as a recovery mechanism while replacement is enabled.\n7. **Persist and reload safely.** Call `saveIndex(file)`, destroy the index if\n   desired, then construct the load form with the same compatible space. Set\n   `ef` again after loading: it is not saved and loading initializes it to 10.\n8. **Verify a small invariant.** Assert result count\u002Forder, filter acceptance,\n   and a post-load query. Run the bundled smoke wrapper with an explicit include\n   root; do not use a large stress or benchmark as the basic API gate.\n9. **Escalate optional APIs.** For epsilon or multivector work, use the exact\n   declarations and data layout in [api-reference.md](references\u002Fapi-reference.md)\n   and the recipes in [workflows.md](references\u002Fworkflows.md). Do not infer a\n   Python equivalent or invent a convenience overload.\n10. **Bound concurrency.** Concurrent read-only searches are the normal safe\n    pattern; concurrent insertion\u002Fupdate calls are demonstrated and internally\n    label-serialized. Coordinate structural operations, persistence, resizing,\n    destruction, and mixed read\u002Fwrite phases externally. See\n    [concurrency.md](references\u002Fconcurrency.md).\n\n## API routing map\n\n- Metric and memory layout: [api-reference.md](references\u002Fapi-reference.md).\n- Construction, query, filter, persistence, and mutation recipes:\n  [workflows.md](references\u002Fworkflows.md).\n- Threads, compiler flags, OpenMP, native SIMD flags, and ownership:\n  [concurrency.md](references\u002Fconcurrency.md).\n- Failure diagnosis, including low `ef`, capacity, paths, deletion policy, and\n  benchmark exclusions: [troubleshooting.md](references\u002Ftroubleshooting.md).\n- Executable smoke check: `scripts\u002Frun_cpp_smoke.sh --include-dir \u003Cinclude-root>`.\n\n## Hard guardrails\n\n- Do not claim CUDA support, GPU kernels, a compiled library ABI, or a required\n  OpenMP dependency. The exposed C++ API is CPU-native and header-only.\n- Do not treat a priority-queue pop sequence as closest-first. The queue's\n  further-first behavior is intentional; the vector helper reverses it.\n- Do not pass `replace_deleted=true` unless the index was created or loaded with\n  replacement enabled. Do not use a filter to make an unavailable `k` appear\n  available: filtering can legitimately return fewer than `k` items.\n- Do not let a temporary query buffer, metric space, filter, stop condition, or\n  data backing an in-flight operation go out of scope.\n- Do not use the smoke result to make recall, latency, SIMD portability, or\n  multi-thread stress claims. Those require a separately chosen workload and\n  compatible hardware.\n\n## Completion checklist\n\nBefore handing off a C++ solution, confirm:\n\n- the include root and compiler standard are explicit;\n- the space outlives the index and matches the data dimension\u002Fmetric;\n- `M`, `ef_construction`, capacity, and query `ef` have an intentional value;\n- queue versus closer-first semantics are documented at the call site;\n- filters inspect external labels and tolerate fewer than `k` matches;\n- deletion\u002Freplacement flags match the requested lifecycle;\n- persistence reloads with a compatible space and resets `ef` explicitly;\n- optional stop-condition signatures were copied from the bundled reference;\n- no operation is concurrently resizing, saving, loading, or destroying the index;\n- the smoke program compiles and passes with assertions.\n",{"name":5,"license":23,"metadata":24,"description":7,"disable-model-invocation":26},"Apache 2.0",{"disco-role":25},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":26},[36,39,42],{"tool":37,"command":38},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill cpp-header-api",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill cpp-header-api",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill cpp-header-api","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002FSKILL.md",[47,53,57,61,65,71],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fapi-reference.md","REFERENCE","参考文档",11433,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Freferences\u002Fapi-reference.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fconcurrency.md",6258,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Freferences\u002Fconcurrency.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",8987,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Fworkflows.md",8774,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Freferences\u002Fworkflows.md",{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Fcpp_smoke.cpp","SCRIPT","脚本",3489,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Fscripts\u002Fcpp_smoke.cpp",{"path":72,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Frun_cpp_smoke.sh",1835,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fhnswlib\u002Fsub-skills\u002Fcpp-header-api\u002Fscripts\u002Frun_cpp_smoke.sh",[76,83,90,97,104,111],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":18},"01M799NYSH1R3RVAHSM2D88Y84","passnet-triton-opt","Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block\u002Fgrid\u002Fwarp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":18},"01M799NS5APEC0N1FZ7XZVW6BN","passnet-solve","END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-solve\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":18},"01M799NNY9BKSM21V5R7QVTBJX","passnet-skill","PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under .\u002Fpass_dir\u002F, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-skill\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":18},"01M799NHH6NP7Z1RK77KGXAN6S","passnet-pattern-fusion","Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()\u002Freplacement_args()\u002Freplacement_func() files, when a pass \"failed to match\", or when deciding how to fuse multiple kernels\u002Fops.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-pattern-fusion\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":18},"01M799ND2J4WNC2PGAZ0A06NDF","passnet-orchestrate","Round\u002Fbatch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep\u002Frevert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-orchestrate\u002FSKILL.md",[],[],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":18},"01M799N8SBTMY02ZDG2AZF5ZWE","passnet-feedback","Fast, deterministic feedback for PassNet work: pre-flight pattern\u002Fmatch verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-feedback\u002FSKILL.md",[],[]]