[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M794ATAPTE6VABVKVDR9BASR":3,"skill-same:01M794ATAPTE6VABVKVDR9BASR":66},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":32,"install":33,"resources":45},"01M794ATAPTE6VABVKVDR9BASR","isaac-gym-simulation",null,"Operate the Isaac Gym Preview 4 simulation, vectorized locomotion environments, robot assets, terrain, state tensors, effort control, and interactive MPC launcher for this project.","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\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: isaac-gym-simulation\ndescription: \"Operate the Isaac Gym Preview 4 simulation, vectorized locomotion\n  environments, robot assets, terrain, state tensors, effort control, and\n  interactive MPC launcher for this project.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Isaac Gym simulation\n\nUse this sub-skill when the task involves Isaac Gym setup, PhysX simulation,\nURDF loading, env\u002Factor creation, terrain, state or force data, vectorized RL\nstepping, or the interactive `RL_MPC_Locomotion` launcher. It is a procedural,\ndata-oriented guide: handles create the simulation, while NumPy or PyTorch\narrays\u002Ftensors carry state and commands.\n\n## Required-backend gate\n\n**RL and simulation execution is currently blocked.** The closed-source Isaac\nGym Preview 4 package was unavailable in the construction environment. The\npartial handoff records PyTorch 1.10\u002FCUDA 11.3, an A100 allocation, repository\nimports, `rsl_rl`, and `mpc_osqp` as passing checks; those results do **not**\nprove that Isaac Gym is installed or usable. From this sub-skill's directory, run the bundled diagnostic first:\n\n```bash\npython scripts\u002Fcheck_isaacgym.py\n```\n\nDo not claim that an RL rollout, viewer run, terrain run, or Isaac Gym import\npassed until this gate is cleared with the vendor package and a compatible\nGPU\u002Fruntime. The diagnostic never creates a viewer or simulation.\n\n## Route quickly\n\n- Need the exact lifecycle, API calls, tensor shapes, or viewer loop: read\n  [references\u002Fsimulation-api.md](references\u002Fsimulation-api.md).\n- Need robot selection, URDF\u002Fmesh caveats, ground, or terrain generation: read\n  [references\u002Fassets-and-terrains.md](references\u002Fassets-and-terrains.md).\n- Need a failure diagnosis or a safe stop condition: read\n  [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\n- Need the PPO\u002FHydra task, checkpoint, observation, or training contract: route\n  to [rl-training](..\u002Frl-training\u002FSKILL.md).\n- Need MPC state conversion, controller modes, gait, or solver behavior: route\n  to [mpc-control](..\u002Fmpc-control\u002FSKILL.md).\n\n## Setup contract\n\nUse Python 3.8 with PyTorch 1.10.0 and CUDA 11.3, Isaac Gym Preview 4, NumPy,\nHydra\u002FOmegaConf, and the pinned `rsl_rl` integration. A compatible NVIDIA\nGPU\u002Fdriver is required for the configured PhysX GPU pipeline; the upgrade notes\ncall out driver 470 or newer. The project normally uses `cuda:0` for both\nsimulation and RL, `physics_engine: physx`, Z-up gravity `[0, 0, -9.81]`,\n`dt: 0.01`, and two substeps. Keep the setup sequence and backend verdict\nexplicit; CUDA-visible PyTorch alone is insufficient.\n\n## Canonical procedural order\n\n1. Acquire a gym handle, create `SimParams`, and set the control timestep,\n   substeps, up-axis, gravity, and PhysX options.\n2. Create the PhysX simulation with compute and graphics device ids. Stop on a\n   null simulation handle.\n3. Add exactly the intended ground plane or triangle terrain before creating\n   environments. Do not silently combine terrain conventions from the\n   interactive and RL paths.\n4. Select one supported URDF, configure `AssetOptions`, load it, inspect DOF\n   and rigid-body names\u002Fcounts, and configure effort properties.\n5. Create the environment grid and one actor per environment. Cache env and\n   actor handles and resolve the `trunk`, foot, thigh, and hip body handles.\n6. Prepare the simulation before acquiring\u002Fwrapping GPU state tensors. Refresh\n   every tensor immediately before reading it.\n7. Create a viewer only when rendering is requested. In each loop apply\n   controls, simulate, fetch results as required by the pipeline, read state,\n   update graphics, and synchronize frame time.\n8. On every exit path stop the gamepad thread, destroy the viewer if created,\n   then destroy the simulation. Use `try\u002Ffinally` in new integrations.\n\n## Control and data rules\n\n- Effort mode is the project default. Set each DOF `driveMode` to\n  `DOF_MODE_EFFORT` and stiffness\u002Fdamping to zero, then apply a complete effort\n  vector every physics frame. Missing a frame does not preserve a safe command.\n- The direct launcher uses structured NumPy state (`pos`, `vel`; body pose,\n  quaternion, linear and angular velocity) and applies per-actor efforts. The\n  RL tasks use `gymtorch` wrappers and batched tensors; do not mix the two\n  contracts without an explicit CPU\u002FGPU conversion.\n- With one actor per env, root state is `(num_envs, 13)` in the order position\n  3, quaternion 4 `(x,y,z,w)`, linear velocity 3, angular velocity 3. DOF state\n  is flattened as `(num_envs * num_dof, 2)` and reshaped to positions and\n  velocities `(num_envs, num_dof)`. Net contacts are `(num_envs, num_bodies,\n  3)`.\n- RL actions are `(num_envs, 12)` and clipped to `[-1, 1]`; observations are\n  `(num_envs, 48)`, commands are `(num_envs, 3)`, rewards\u002Fresets are per-env.\n  The vector task returns `(obs, privileged_obs, rewards, resets, extras)`.\n- The RL observation is base position, body-frame linear\u002Fangular velocity,\n  scaled velocity command, offset DOF position, scaled DOF velocity, and the\n  previous action. The standard simulator must refresh DOF, root, and contact\n  tensors before computing it.\n\n## Interactive launcher contract\n\nThe supported robots are `Aliengo`, `A1`, and `Go1`. The launcher accepts\n`--robot`, `--mode {Fsm,Min,Policy}`, `--num-envs`, `--render-fps`,\n`--disable-gamepad`, and `--checkpoint`. `Policy` consumes a checkpoint; if it\nis omitted, the policy runner resolves the latest project run. `Fsm` is the\nnormal hierarchical controller and `Min` is the minimal MPC runner.\n\nWithout `--disable-gamepad`, an Xbox-like device is required. The left stick\ncommands forward\u002Flateral velocity, the right stick commands yaw, LB cycles\nTrot\u002FWalk\u002FBound, and RB cycles Locomotion\u002FRecovery Stand. LB+RB requests an\nemergency stop; press the left stick to release it. For unattended or\nheadless automation, disable the gamepad and do not create a viewer. The\ninteractive launcher still renders and therefore remains backend-blocked here.\n\n## Completion criteria\n\nA simulation task is complete only when the backend diagnostic passes, asset\npaths and names resolve, the simulation is stepped in the documented order,\nstate refreshes and tensor shapes are checked, and cleanup is observed. If the\nbackend is absent, provide the exact blocked prerequisite and stop; do not\nsubstitute a CPU-only import or a successful MPC build as Isaac Gym evidence.\n","# Isaac Gym simulation\n\nUse this sub-skill when the task involves Isaac Gym setup, PhysX simulation,\nURDF loading, env\u002Factor creation, terrain, state or force data, vectorized RL\nstepping, or the interactive `RL_MPC_Locomotion` launcher. It is a procedural,\ndata-oriented guide: handles create the simulation, while NumPy or PyTorch\narrays\u002Ftensors carry state and commands.\n\n## Required-backend gate\n\n**RL and simulation execution is currently blocked.** The closed-source Isaac\nGym Preview 4 package was unavailable in the construction environment. The\npartial handoff records PyTorch 1.10\u002FCUDA 11.3, an A100 allocation, repository\nimports, `rsl_rl`, and `mpc_osqp` as passing checks; those results do **not**\nprove that Isaac Gym is installed or usable. From this sub-skill's directory, run the bundled diagnostic first:\n\n```bash\npython scripts\u002Fcheck_isaacgym.py\n```\n\nDo not claim that an RL rollout, viewer run, terrain run, or Isaac Gym import\npassed until this gate is cleared with the vendor package and a compatible\nGPU\u002Fruntime. The diagnostic never creates a viewer or simulation.\n\n## Route quickly\n\n- Need the exact lifecycle, API calls, tensor shapes, or viewer loop: read\n  [references\u002Fsimulation-api.md](references\u002Fsimulation-api.md).\n- Need robot selection, URDF\u002Fmesh caveats, ground, or terrain generation: read\n  [references\u002Fassets-and-terrains.md](references\u002Fassets-and-terrains.md).\n- Need a failure diagnosis or a safe stop condition: read\n  [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\n- Need the PPO\u002FHydra task, checkpoint, observation, or training contract: route\n  to [rl-training](..\u002Frl-training\u002FSKILL.md).\n- Need MPC state conversion, controller modes, gait, or solver behavior: route\n  to [mpc-control](..\u002Fmpc-control\u002FSKILL.md).\n\n## Setup contract\n\nUse Python 3.8 with PyTorch 1.10.0 and CUDA 11.3, Isaac Gym Preview 4, NumPy,\nHydra\u002FOmegaConf, and the pinned `rsl_rl` integration. A compatible NVIDIA\nGPU\u002Fdriver is required for the configured PhysX GPU pipeline; the upgrade notes\ncall out driver 470 or newer. The project normally uses `cuda:0` for both\nsimulation and RL, `physics_engine: physx`, Z-up gravity `[0, 0, -9.81]`,\n`dt: 0.01`, and two substeps. Keep the setup sequence and backend verdict\nexplicit; CUDA-visible PyTorch alone is insufficient.\n\n## Canonical procedural order\n\n1. Acquire a gym handle, create `SimParams`, and set the control timestep,\n   substeps, up-axis, gravity, and PhysX options.\n2. Create the PhysX simulation with compute and graphics device ids. Stop on a\n   null simulation handle.\n3. Add exactly the intended ground plane or triangle terrain before creating\n   environments. Do not silently combine terrain conventions from the\n   interactive and RL paths.\n4. Select one supported URDF, configure `AssetOptions`, load it, inspect DOF\n   and rigid-body names\u002Fcounts, and configure effort properties.\n5. Create the environment grid and one actor per environment. Cache env and\n   actor handles and resolve the `trunk`, foot, thigh, and hip body handles.\n6. Prepare the simulation before acquiring\u002Fwrapping GPU state tensors. Refresh\n   every tensor immediately before reading it.\n7. Create a viewer only when rendering is requested. In each loop apply\n   controls, simulate, fetch results as required by the pipeline, read state,\n   update graphics, and synchronize frame time.\n8. On every exit path stop the gamepad thread, destroy the viewer if created,\n   then destroy the simulation. Use `try\u002Ffinally` in new integrations.\n\n## Control and data rules\n\n- Effort mode is the project default. Set each DOF `driveMode` to\n  `DOF_MODE_EFFORT` and stiffness\u002Fdamping to zero, then apply a complete effort\n  vector every physics frame. Missing a frame does not preserve a safe command.\n- The direct launcher uses structured NumPy state (`pos`, `vel`; body pose,\n  quaternion, linear and angular velocity) and applies per-actor efforts. The\n  RL tasks use `gymtorch` wrappers and batched tensors; do not mix the two\n  contracts without an explicit CPU\u002FGPU conversion.\n- With one actor per env, root state is `(num_envs, 13)` in the order position\n  3, quaternion 4 `(x,y,z,w)`, linear velocity 3, angular velocity 3. DOF state\n  is flattened as `(num_envs * num_dof, 2)` and reshaped to positions and\n  velocities `(num_envs, num_dof)`. Net contacts are `(num_envs, num_bodies,\n  3)`.\n- RL actions are `(num_envs, 12)` and clipped to `[-1, 1]`; observations are\n  `(num_envs, 48)`, commands are `(num_envs, 3)`, rewards\u002Fresets are per-env.\n  The vector task returns `(obs, privileged_obs, rewards, resets, extras)`.\n- The RL observation is base position, body-frame linear\u002Fangular velocity,\n  scaled velocity command, offset DOF position, scaled DOF velocity, and the\n  previous action. The standard simulator must refresh DOF, root, and contact\n  tensors before computing it.\n\n## Interactive launcher contract\n\nThe supported robots are `Aliengo`, `A1`, and `Go1`. The launcher accepts\n`--robot`, `--mode {Fsm,Min,Policy}`, `--num-envs`, `--render-fps`,\n`--disable-gamepad`, and `--checkpoint`. `Policy` consumes a checkpoint; if it\nis omitted, the policy runner resolves the latest project run. `Fsm` is the\nnormal hierarchical controller and `Min` is the minimal MPC runner.\n\nWithout `--disable-gamepad`, an Xbox-like device is required. The left stick\ncommands forward\u002Flateral velocity, the right stick commands yaw, LB cycles\nTrot\u002FWalk\u002FBound, and RB cycles Locomotion\u002FRecovery Stand. LB+RB requests an\nemergency stop; press the left stick to release it. For unattended or\nheadless automation, disable the gamepad and do not create a viewer. The\ninteractive launcher still renders and therefore remains backend-blocked here.\n\n## Completion criteria\n\nA simulation task is complete only when the backend diagnostic passes, asset\npaths and names resolve, the simulation is stepped in the documented order,\nstate refreshes and tensor shapes are checked, and cleanup is observed. If the\nbackend is absent, provide the exact blocked prerequisite and stop; do not\nsubstitute a CPU-only import or a successful MPC build as Isaac Gym evidence.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":25},[35,38,41],{"tool":36,"command":37},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill isaac-gym-simulation",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill isaac-gym-simulation",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill isaac-gym-simulation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002FSKILL.md",[46,52,56,60],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fassets-and-terrains.md","REFERENCE","参考文档",7185,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002Freferences\u002Fassets-and-terrains.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fsimulation-api.md",10413,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002Freferences\u002Fsimulation-api.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Ftroubleshooting.md",9115,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002Freferences\u002Ftroubleshooting.md",{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"scripts\u002Fcheck_isaacgym.py","SCRIPT","脚本",4383,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Frl-mpc-locomotion\u002Fsub-skills\u002Fisaac-gym-simulation\u002Fscripts\u002Fcheck_isaacgym.py",[67,75,82,89,96,103],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"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":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"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":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"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":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"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":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"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":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":109,"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",[],[]]