[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M792QEFGJS0D6KDWFMRQDBGN":3,"skill-same:01M792QEFGJS0D6KDWFMRQDBGN":67},{"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},"01M792QEFGJS0D6KDWFMRQDBGN","benchmarks",null,"Load PhySO's shipped Feynman and Class benchmark problems, generate bounded samples, inspect symbolic metadata, and compare candidate SymPy expressions.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"MIT","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: benchmarks\ndescription: \"Load PhySO's shipped Feynman and Class benchmark problems,\n  generate bounded samples, inspect symbolic metadata, and compare candidate\n  SymPy expressions.\"\nread_when: \"Use when a task needs benchmark-problem data or symbolic-equivalence\n  checks; do not use it to launch the maintainer benchmark suites.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# PhySO benchmark problems\n\n## Role\n\nThis sub-skill is the bounded, read-only benchmark-data route for the installed\n`physo` package. It teaches a Researcher to select a Feynman or Class problem,\ninspect its target and metadata, generate a small in-memory sample, obtain the\nprefix form, and apply the package's symbolic comparison convention.\n\nIt is not a general benchmarking or model-training skill. Keep generated arrays\nsmall unless the caller explicitly owns a separate experiment workflow.\n\n## Boundaries\n\n**Included**\n\n- `FeynmanProblem` for the 120 Feynman entries (bulk indices 0--99 and bonus\n  indices 100--119), including equation-name lookup.\n- `ClassProblem` for the 8 Class entries, including realization-specific `K`\n  constants and equation-name lookup.\n- Formula, variable, target, range, unit, SymPy-assumption, and prefix metadata.\n- `generate_data_points`, `target_function`, `get_sympy`, and non-blocking\n  `show_sample` usage.\n- `FeynmanProblem.compare_expression` and\n  `physo.benchmark.utils.symbolic_utils.compare_expression`.\n- The small loader check in `scripts\u002Fsmoke_benchmark_loaders.py`.\n\n**Excluded**\n\n- Running SR or Class SR (`physo.SR` and `physo.ClassSR`); route those requests\n  to [`sr`](..\u002Fsr\u002FSKILL.md) or [`class-sr`](..\u002Fclass-sr\u002FSKILL.md).\n- Full benchmark reproduction, noisy benchmark campaigns, result analysis,\n  cluster\u002FHPC jobfiles, and long-running runners.\n- Generic data-science benchmark design or numerical-only model scoring.\n- Pareto-front and training-log handling; route those questions to\n  [`toolkit`](..\u002Ftoolkit\u002FSKILL.md) or the root skill.\n\nThe maintainer scripts `benchmarking\u002FFeynmanBenchmark\u002Ffeynman_run.py` and\n`benchmarking\u002FClassBenchmark\u002Fclassbench_run.py` are reference-only. The\n`benchmarking\u002F*_make_run_file.py` files and `benchmarking\u002Futils.py` jobfile\nhelper are deliberately outside this runtime skill.\n\n## Verified entry points\n\nThe following imports and package facts are verified for PhySO 1.2.0:\n\n```python\nimport physo.benchmark.FeynmanDataset.FeynmanProblem as Feyn\nimport physo.benchmark.ClassDataset.ClassProblem as Cls\nfrom physo.benchmark.utils import symbolic_utils as su\n\nfeyn_pb = Feyn.FeynmanProblem(i_eq=0)\nclass_pb = Cls.ClassProblem(i_eq=0)\n```\n\nThe verified inspection\u002Fruntime baseline is CPU-only Python 3.12 with NumPy,\nSymPy, pandas, Matplotlib, scikit-learn, and a CPU PyTorch build. No CUDA claim\nis made here. Import-time warnings about unavailable system LaTeX are expected\nand do not by themselves invalidate loading or comparison.\n\n## Workflow\n\n1. **Validate the selector before constructing the object.** Use an integer in\n   `0 \u003C= i_eq \u003C Feyn.N_EQS` (120) or `0 \u003C= i_eq \u003C Cls.N_EQS` (8), or use the\n   exact `eq_name` from the dataset. A name can be inspected with\n   `Feyn.EQS_FEYNMAN_DF[\"Name\"]` or `Cls.EQS_CLASS_DF[\"Name\"]`. If both selector\n   arguments are supplied, the current constructors take the index branch.\n2. **Load and inspect metadata.** Start with `print(pb)`, then inspect\n   `formula_sympy`, `formula_sympy_eval`, `formula_latex`, `X_names`,\n   `X_lows`, `X_highs`, `X_units`, `y_name`, and `y_units`. Class problems also\n   expose `K_names`, `K_lows`, `K_highs`, and `K_units`; `K` means one row of\n   realization-specific constants, not a Class SR class constant.\n3. **Generate a bounded sample.** Set `np.random.seed(...)` when reproducible\n   values matter and pass an explicit small `n_samples`. Feynman returns\n   `X.shape == (pb.n_vars, n_samples)` and `y.shape == (n_samples,)`. Class\n   returns `multi_X.shape == (n_realizations, pb.n_vars, n_samples)` and\n   `multi_y.shape == (n_realizations, n_samples)`; with `return_K=True`, it\n   also returns `K.shape == (n_realizations, pb.n_spe)`.\n4. **Check the target directly.** For Feynman use `pb.target_function(X)`.\n   For a Class realization use `pb.target_function(multi_X[i], K[i])`. Compare\n   shapes before calling `np.allclose`; do not transpose the package's\n   `(variables, samples)` convention.\n5. **Inspect the expression representation.**\n   `pb.get_prefix_expression()` returns a dictionary with `tokens_str`,\n   `arities`, and `tokens`, traversed in SymPy pre-order. For a Class problem,\n   the prefix form still contains symbolic `k*` tokens. Use\n   `pb.get_sympy(K_vals=...)` to substitute one or more concrete `K` rows.\n6. **Compare symbolically.** Parse a candidate with the problem's assumption-\n   carrying symbol dictionary, then call the problem wrapper for Feynman or\n   `su.compare_expression` for Class. Follow the exact examples in\n   [symbolic-equivalence.md](references\u002Fsymbolic-equivalence.md), and retain\n   the returned `report` instead of relying only on the Boolean.\n7. **Stop at inspection.** If the request asks for all equations, noise sweeps,\n   SR training, job submission, or result reproduction, hand off to the\n   appropriate root\u002FSR\u002FClass SR workflow and keep this route as the loader and\n   equivalence utility only.\n\n## Input\u002Foutput contracts\n\n| Operation | Inputs | Outputs and invariants |\n| --- | --- | --- |\n| `FeynmanProblem(i_eq=...)` or `eq_name=...` | Valid index or exact name | One problem; `X` is variables-first; target `y` is one-dimensional. |\n| `ClassProblem(i_eq=...)` or `eq_name=...` | Valid index or exact name | One problem; `K` has one row per realization and one column per `K_names`. |\n| `generate_data_points` | Small positive `n_samples`; Class also `n_realizations` | NumPy float arrays sampled within declared input and constant ranges. |\n| `target_function` | Feynman `X`, or Class `X` plus one `K` row | Target values for the supplied shape; it does not validate a transposed data layout for you. |\n| `get_prefix_expression` | No arguments | Dict keys `tokens_str`, `arities`, `tokens` with equal lengths. |\n| `compare_expression` | SymPy expressions with matching assumption-aware symbols | `(is_equivalent: bool, report: dict)`; report includes symbolic error\u002Ffraction and solution flags. |\n\n## Safe helper and native checks\n\nRun the bundled helper from an environment where `physo` is installed:\n\n```bash\npython skills\u002Fdisco\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Fscripts\u002Fsmoke_benchmark_loaders.py\n```\n\nIt loads Feynman problem 0 and Class problem 0, generates only a few points,\nchecks target evaluation and array shapes, and prints stable metadata. It never\nruns a benchmark trainer or creates result directories.\n\nThe corresponding native candidates are:\n\n- `python -m unittest physo.benchmark.FeynmanDataset.tests.FeynmanProblem_UnitTest -q`\n- `python -m unittest physo.benchmark.ClassDataset.tests.ClassProblem_UnitTest -q`\n\nThese tests iterate over all shipped equations with small samples; they are\nCPU-safe correctness checks, not full SR benchmark reproduction.\n\n## Troubleshooting\n\nSee [problem-loaders.md](references\u002Fproblem-loaders.md) for selector, shape,\nmetadata, and plotting details; [symbolic-equivalence.md](references\u002Fsymbolic-equivalence.md)\nfor assumptions, rounding, and trigonometric comparison; and\n[troubleshooting.md](references\u002Ftroubleshooting.md) for recovery and the\nmaintainer-runner boundary.\n\nCommon first responses:\n\n- Normalize an index before construction. Positive out-of-range and unknown\n  names currently surface as pandas `IndexError`; negative `.iloc` indices can\n  accidentally select the last row, so reject them in caller code.\n- Use the exact shape contracts above. Feynman is `(n_vars, n_samples)`, not\n  `(n_samples, n_vars)`; Class adds a realization axis.\n- Parse candidates with `pb.sympy_X_symbols_dict` (or the documented original-\n  name dictionary), not fresh symbols with different assumptions.\n- Treat `compare_expression` as the benchmark's symbolic convention, not as a\n  numerical error metric. Inspect its report, especially when rounding or\n  trigonometric handling is enabled.\n- Use `show_sample(..., do_show=False)` in headless work. Missing Matplotlib is\n  an installation problem because these benchmark modules import it; missing\n  system LaTeX only affects optional display and is normally warning-level.\n\n## Related routes\n\n- Root routing: [`..\u002F..\u002FSKILL.md`](..\u002F..\u002FSKILL.md)\n- Single-dataset SR: [`..\u002Fsr\u002FSKILL.md`](..\u002Fsr\u002FSKILL.md)\n- Class SR data-shape and training guidance: [`..\u002Fclass-sr\u002FSKILL.md`](..\u002Fclass-sr\u002FSKILL.md)\n- Low-level expression and result inspection: [`..\u002Ftoolkit\u002FSKILL.md`](..\u002Ftoolkit\u002FSKILL.md)\n","# PhySO benchmark problems\n\n## Role\n\nThis sub-skill is the bounded, read-only benchmark-data route for the installed\n`physo` package. It teaches a Researcher to select a Feynman or Class problem,\ninspect its target and metadata, generate a small in-memory sample, obtain the\nprefix form, and apply the package's symbolic comparison convention.\n\nIt is not a general benchmarking or model-training skill. Keep generated arrays\nsmall unless the caller explicitly owns a separate experiment workflow.\n\n## Boundaries\n\n**Included**\n\n- `FeynmanProblem` for the 120 Feynman entries (bulk indices 0--99 and bonus\n  indices 100--119), including equation-name lookup.\n- `ClassProblem` for the 8 Class entries, including realization-specific `K`\n  constants and equation-name lookup.\n- Formula, variable, target, range, unit, SymPy-assumption, and prefix metadata.\n- `generate_data_points`, `target_function`, `get_sympy`, and non-blocking\n  `show_sample` usage.\n- `FeynmanProblem.compare_expression` and\n  `physo.benchmark.utils.symbolic_utils.compare_expression`.\n- The small loader check in `scripts\u002Fsmoke_benchmark_loaders.py`.\n\n**Excluded**\n\n- Running SR or Class SR (`physo.SR` and `physo.ClassSR`); route those requests\n  to [`sr`](..\u002Fsr\u002FSKILL.md) or [`class-sr`](..\u002Fclass-sr\u002FSKILL.md).\n- Full benchmark reproduction, noisy benchmark campaigns, result analysis,\n  cluster\u002FHPC jobfiles, and long-running runners.\n- Generic data-science benchmark design or numerical-only model scoring.\n- Pareto-front and training-log handling; route those questions to\n  [`toolkit`](..\u002Ftoolkit\u002FSKILL.md) or the root skill.\n\nThe maintainer scripts `benchmarking\u002FFeynmanBenchmark\u002Ffeynman_run.py` and\n`benchmarking\u002FClassBenchmark\u002Fclassbench_run.py` are reference-only. The\n`benchmarking\u002F*_make_run_file.py` files and `benchmarking\u002Futils.py` jobfile\nhelper are deliberately outside this runtime skill.\n\n## Verified entry points\n\nThe following imports and package facts are verified for PhySO 1.2.0:\n\n```python\nimport physo.benchmark.FeynmanDataset.FeynmanProblem as Feyn\nimport physo.benchmark.ClassDataset.ClassProblem as Cls\nfrom physo.benchmark.utils import symbolic_utils as su\n\nfeyn_pb = Feyn.FeynmanProblem(i_eq=0)\nclass_pb = Cls.ClassProblem(i_eq=0)\n```\n\nThe verified inspection\u002Fruntime baseline is CPU-only Python 3.12 with NumPy,\nSymPy, pandas, Matplotlib, scikit-learn, and a CPU PyTorch build. No CUDA claim\nis made here. Import-time warnings about unavailable system LaTeX are expected\nand do not by themselves invalidate loading or comparison.\n\n## Workflow\n\n1. **Validate the selector before constructing the object.** Use an integer in\n   `0 \u003C= i_eq \u003C Feyn.N_EQS` (120) or `0 \u003C= i_eq \u003C Cls.N_EQS` (8), or use the\n   exact `eq_name` from the dataset. A name can be inspected with\n   `Feyn.EQS_FEYNMAN_DF[\"Name\"]` or `Cls.EQS_CLASS_DF[\"Name\"]`. If both selector\n   arguments are supplied, the current constructors take the index branch.\n2. **Load and inspect metadata.** Start with `print(pb)`, then inspect\n   `formula_sympy`, `formula_sympy_eval`, `formula_latex`, `X_names`,\n   `X_lows`, `X_highs`, `X_units`, `y_name`, and `y_units`. Class problems also\n   expose `K_names`, `K_lows`, `K_highs`, and `K_units`; `K` means one row of\n   realization-specific constants, not a Class SR class constant.\n3. **Generate a bounded sample.** Set `np.random.seed(...)` when reproducible\n   values matter and pass an explicit small `n_samples`. Feynman returns\n   `X.shape == (pb.n_vars, n_samples)` and `y.shape == (n_samples,)`. Class\n   returns `multi_X.shape == (n_realizations, pb.n_vars, n_samples)` and\n   `multi_y.shape == (n_realizations, n_samples)`; with `return_K=True`, it\n   also returns `K.shape == (n_realizations, pb.n_spe)`.\n4. **Check the target directly.** For Feynman use `pb.target_function(X)`.\n   For a Class realization use `pb.target_function(multi_X[i], K[i])`. Compare\n   shapes before calling `np.allclose`; do not transpose the package's\n   `(variables, samples)` convention.\n5. **Inspect the expression representation.**\n   `pb.get_prefix_expression()` returns a dictionary with `tokens_str`,\n   `arities`, and `tokens`, traversed in SymPy pre-order. For a Class problem,\n   the prefix form still contains symbolic `k*` tokens. Use\n   `pb.get_sympy(K_vals=...)` to substitute one or more concrete `K` rows.\n6. **Compare symbolically.** Parse a candidate with the problem's assumption-\n   carrying symbol dictionary, then call the problem wrapper for Feynman or\n   `su.compare_expression` for Class. Follow the exact examples in\n   [symbolic-equivalence.md](references\u002Fsymbolic-equivalence.md), and retain\n   the returned `report` instead of relying only on the Boolean.\n7. **Stop at inspection.** If the request asks for all equations, noise sweeps,\n   SR training, job submission, or result reproduction, hand off to the\n   appropriate root\u002FSR\u002FClass SR workflow and keep this route as the loader and\n   equivalence utility only.\n\n## Input\u002Foutput contracts\n\n| Operation | Inputs | Outputs and invariants |\n| --- | --- | --- |\n| `FeynmanProblem(i_eq=...)` or `eq_name=...` | Valid index or exact name | One problem; `X` is variables-first; target `y` is one-dimensional. |\n| `ClassProblem(i_eq=...)` or `eq_name=...` | Valid index or exact name | One problem; `K` has one row per realization and one column per `K_names`. |\n| `generate_data_points` | Small positive `n_samples`; Class also `n_realizations` | NumPy float arrays sampled within declared input and constant ranges. |\n| `target_function` | Feynman `X`, or Class `X` plus one `K` row | Target values for the supplied shape; it does not validate a transposed data layout for you. |\n| `get_prefix_expression` | No arguments | Dict keys `tokens_str`, `arities`, `tokens` with equal lengths. |\n| `compare_expression` | SymPy expressions with matching assumption-aware symbols | `(is_equivalent: bool, report: dict)`; report includes symbolic error\u002Ffraction and solution flags. |\n\n## Safe helper and native checks\n\nRun the bundled helper from an environment where `physo` is installed:\n\n```bash\npython skills\u002Fdisco\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Fscripts\u002Fsmoke_benchmark_loaders.py\n```\n\nIt loads Feynman problem 0 and Class problem 0, generates only a few points,\nchecks target evaluation and array shapes, and prints stable metadata. It never\nruns a benchmark trainer or creates result directories.\n\nThe corresponding native candidates are:\n\n- `python -m unittest physo.benchmark.FeynmanDataset.tests.FeynmanProblem_UnitTest -q`\n- `python -m unittest physo.benchmark.ClassDataset.tests.ClassProblem_UnitTest -q`\n\nThese tests iterate over all shipped equations with small samples; they are\nCPU-safe correctness checks, not full SR benchmark reproduction.\n\n## Troubleshooting\n\nSee [problem-loaders.md](references\u002Fproblem-loaders.md) for selector, shape,\nmetadata, and plotting details; [symbolic-equivalence.md](references\u002Fsymbolic-equivalence.md)\nfor assumptions, rounding, and trigonometric comparison; and\n[troubleshooting.md](references\u002Ftroubleshooting.md) for recovery and the\nmaintainer-runner boundary.\n\nCommon first responses:\n\n- Normalize an index before construction. Positive out-of-range and unknown\n  names currently surface as pandas `IndexError`; negative `.iloc` indices can\n  accidentally select the last row, so reject them in caller code.\n- Use the exact shape contracts above. Feynman is `(n_vars, n_samples)`, not\n  `(n_samples, n_vars)`; Class adds a realization axis.\n- Parse candidates with `pb.sympy_X_symbols_dict` (or the documented original-\n  name dictionary), not fresh symbols with different assumptions.\n- Treat `compare_expression` as the benchmark's symbolic convention, not as a\n  numerical error metric. Inspect its report, especially when rounding or\n  trigonometric handling is enabled.\n- Use `show_sample(..., do_show=False)` in headless work. Missing Matplotlib is\n  an installation problem because these benchmark modules import it; missing\n  system LaTeX only affects optional display and is normally warning-level.\n\n## Related routes\n\n- Root routing: [`..\u002F..\u002FSKILL.md`](..\u002F..\u002FSKILL.md)\n- Single-dataset SR: [`..\u002Fsr\u002FSKILL.md`](..\u002Fsr\u002FSKILL.md)\n- Class SR data-shape and training guidance: [`..\u002Fclass-sr\u002FSKILL.md`](..\u002Fclass-sr\u002FSKILL.md)\n- Low-level expression and result inspection: [`..\u002Ftoolkit\u002FSKILL.md`](..\u002Ftoolkit\u002FSKILL.md)\n",{"name":5,"license":14,"metadata":23,"read_when":25,"description":7,"disable-model-invocation":26},{"disco-role":24},"operating","Use when a task needs benchmark-problem data or symbolic-equivalence checks; do not use it to launch the maintainer benchmark suites.",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks",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 benchmarks",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill benchmarks",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill benchmarks","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002FSKILL.md",[47,53,57,61],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fproblem-loaders.md","REFERENCE","参考文档",5953,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Freferences\u002Fproblem-loaders.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fsymbolic-equivalence.md",5887,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Freferences\u002Fsymbolic-equivalence.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",6447,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":63,"resourceTypeDesc":64,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Fsmoke_benchmark_loaders.py","SCRIPT","脚本",2981,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fphyso\u002Fsub-skills\u002Fbenchmarks\u002Fscripts\u002Fsmoke_benchmark_loaders.py",[68,76,83,90,97,104],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"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.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"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":72,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"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":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"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":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"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":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"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":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"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",[],[]]