[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78WHTFKDEBCRZDWB3G4FSBP":3,"skill-same:01M78WHTFKDEBCRZDWB3G4FSBP":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},"01M78WHTFKDEBCRZDWB3G4FSBP","policy-evaluation",null,"Evaluate a trained TD3 actor for the Velodyne navigation environment with checkpoint, contract, simulator-readiness, and bounded-run gates.","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\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: policy-evaluation\ndescription: \"Evaluate a trained TD3 actor for the Velodyne navigation\n  environment with checkpoint, contract, simulator-readiness, and bounded-run\n  gates.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Policy evaluation\n\nUse this skill when the task is to evaluate an already-trained TD3 **actor** for the\nVelodyne navigation environment. It is an evaluation and artifact-safety guide, not a\ntraining or ROS setup guide.\n\n## Operating contract\n\n- **Input:** a model directory, a checkpoint base name (normally `TD3_velodyne`), and\n  an evaluation request with explicit episode and step bounds.\n- **Actor artifact:** `\u003Cbase>_actor.pth`, a PyTorch `state_dict` for the actor below.\n- **Optional companion:** `\u003Cbase>_critic.pth` is produced by training, but the test\n  policy needs only the actor. Treat a missing critic as an incomplete TD3 pair, not as\n  an actor failure.\n- **Output:** bounded episode records and aggregate metrics, plus the exact checkpoint,\n  device, state\u002Faction contract, and termination policy used.\n- **Hard safety rule:** never load an artifact merely to inspect its name. Run the\n  bundled `scripts\u002Fcheck_policy_artifacts.py` first; it does not deserialize tensors\n  unless `--load-state-dict` is explicitly requested.\n\n## Policy and environment contract\n\nThe evaluator's actor is `24 -> 800 -> 600 -> 2`: `Linear(24,800)`, ReLU,\n`Linear(800,600)`, ReLU, `Linear(600,2)`, and `tanh`. The 24-element observation is:\n\n1. 20 Velodyne distance bins (`environment_dim=20`), followed by\n2. distance to goal, relative heading, previous\u002Fcommanded linear velocity, and\n   previous\u002Fcommanded angular velocity (`robot_dim=4`).\n\nThe actor returns two normalized values in `[-1, 1]`. Before the environment step, map\nonly the first value to linear velocity with `(a[0] + 1) \u002F 2`; pass the second unchanged\nas angular velocity. Do not feed the environment's `[0, 1]`\u002F`[-1, 1]` command back into\nthe actor as a replacement for the normalized action. Assert finite numeric state of\nlength 24 and finite action of length 2 at every step.\n\nThe reference evaluator uses seed `0`, waits about five seconds after environment\nconstruction, and caps an episode at **500 transitions**. A simulator `done` must end an\nepisode earlier; the cap must also end it when the simulator fails to terminate. Reset\nbefore the next episode and reset the per-episode step counter. Preserve the distinction\nbetween a goal\u002Fcollision termination and a synthetic time-limit termination in metrics.\n\n## Checkpoint gates\n\n1. Resolve the requested model directory and base name without guessing another run.\n2. Run the bundled artifact check. Confirm the actor exists, is a regular file, and has\n   the exact expected name. Record critic presence separately.\n3. For a new or untrusted checkpoint, opt into the bounded `--load-state-dict` check.\n   It uses CPU mapping, `weights_only=True` when supported, a file-size limit, and shape\n   checks; it must not fall back to unrestricted pickle loading. A missing, corrupt,\n   non-tensor, extra-key, or shape-incompatible actor is a hard stop.\n4. If the artifact check reports only a critic, stop: a critic cannot produce actions.\n   If the critic is present but incompatible, stop the full-pair evaluation or label the\n   actor-only run explicitly; never silently substitute or rename it.\n5. Instantiate the actor on CPU unless CUDA availability and the requested device have\n   been independently checked. The reference `torch.load` call does not set\n   `map_location`, so a CUDA-origin checkpoint can fail on a CPU-only host; use the\n   safe CPU-mapped validation\u002Fload path and record any device remapping. Load the actor\n   state dict into the same architecture, call `eval()`, and use inference\u002Fno-grad\n   execution. Record the actual device.\n\nSee `references\u002Fcheckpoint-format.md` for key and shape expectations.\n\n## Simulator readiness gate\n\nA real run requires an operational ROS Noetic\u002FGazebo stack, a running ROS master,\nthe Velodyne and odometry topics, the navigation launch asset, and the services used by\nreset, pause, and unpause. The workspace must already be built and sourced by the run\nowner. Check readiness before constructing the environment; do not claim a simulator\nresult from a Python-only import or from a host without ROS\u002FGazebo\u002Fmessage modules.\n\nThe reference environment starts `roscore` and Gazebo in its constructor. The reference\ntest module constructs that environment at import time and then enters an unbounded\n`while True` loop. Therefore do **not** import that module for a smoke test, and do not\nstart an unconstrained production evaluation. Use a controlled runner that loads the\nactor and applies the same distilled contract, with an explicit episode count and a\nwall-clock\u002Fstep watchdog. A safe dry run can stop after artifact and architecture checks\nwithout touching ROS.\n\n## Bounded evaluation workflow\n\n1. **Plan:** choose `N >= 1` episodes, `max_steps \u003C= 500` per episode, a wall-clock\n   deadline, and a cleanup\u002Fabort action. State whether this is actor-only or a complete\n   TD3 artifact evaluation.\n2. **Preflight:** run `check_policy_artifacts.py --model-dir ... --name ...`; add\n   `--load-state-dict` only after the file-size and provenance decision. Verify simulator\n   readiness separately and record failures as blocked, not as poor policy scores.\n3. **Load:** build the exact actor, load only the validated actor state dict, switch to\n   evaluation mode, and initialize deterministic seeds if reproducibility is desired.\n4. **Run:** for each episode, reset the environment, validate the initial 24-value\n   state, call the actor without exploration noise, convert the action, step the\n   simulator, and record reward, collision\u002Fgoal flag, and step count. Stop on `done`,\n   500 steps, watchdog expiry, or operator abort.\n5. **Aggregate:** report per-episode return, length, termination reason, goal count,\n   collision count, time-limit count, invalid-state\u002Faction count, and timeout\u002Fabort\n   status. Do not call a truncated run a completed evaluation.\n6. **Clean up:** stop the controlled runner and simulator processes according to the\n   host's approved procedure. Preserve the checkpoint path\u002Fbase name and run bounds in\n   the result record.\n\nThe original program's continuous loop is useful evidence of intended repeated testing,\nnot an acceptable default for automation. Never omit the bound because the simulator\nusually terminates episodes.\n\n## Failure routing\n\n- Artifact or shape failure: fix\u002Fselect a compatible actor; do not enter ROS.\n- Missing ROS\u002FGazebo, topics, services, or launch asset: mark **environment blocked**;\n  do not report zero reward or policy failure.\n- State\u002Faction contract violation: abort the episode and mark the result invalid.\n- Watchdog or operator stop: preserve partial records and label the run incomplete.\n- A successful artifact check alone proves checkpoint compatibility only; it proves no\n  navigation quality and no simulator behavior.\n\nFor predictable cases and recovery actions, use `references\u002Ftroubleshooting.md`. The\nbundled script is the only practical helper in this skill and is intentionally independent\nof the original checkout.\n\n## Evidence boundary\n\nThis contract was distilled from `TD3\u002Ftest_velodyne_td3.py`, the actor\u002Fsave\u002Fload portions\nof `TD3\u002Ftrain_velodyne_td3.py`, `TD3\u002Fvelodyne_env.py`, and the README's test procedure.\nThose paths are provenance only. Runtime use should rely on this skill and its bundled\nscript rather than opening or executing source-checkout files.\n","# Policy evaluation\n\nUse this skill when the task is to evaluate an already-trained TD3 **actor** for the\nVelodyne navigation environment. It is an evaluation and artifact-safety guide, not a\ntraining or ROS setup guide.\n\n## Operating contract\n\n- **Input:** a model directory, a checkpoint base name (normally `TD3_velodyne`), and\n  an evaluation request with explicit episode and step bounds.\n- **Actor artifact:** `\u003Cbase>_actor.pth`, a PyTorch `state_dict` for the actor below.\n- **Optional companion:** `\u003Cbase>_critic.pth` is produced by training, but the test\n  policy needs only the actor. Treat a missing critic as an incomplete TD3 pair, not as\n  an actor failure.\n- **Output:** bounded episode records and aggregate metrics, plus the exact checkpoint,\n  device, state\u002Faction contract, and termination policy used.\n- **Hard safety rule:** never load an artifact merely to inspect its name. Run the\n  bundled `scripts\u002Fcheck_policy_artifacts.py` first; it does not deserialize tensors\n  unless `--load-state-dict` is explicitly requested.\n\n## Policy and environment contract\n\nThe evaluator's actor is `24 -> 800 -> 600 -> 2`: `Linear(24,800)`, ReLU,\n`Linear(800,600)`, ReLU, `Linear(600,2)`, and `tanh`. The 24-element observation is:\n\n1. 20 Velodyne distance bins (`environment_dim=20`), followed by\n2. distance to goal, relative heading, previous\u002Fcommanded linear velocity, and\n   previous\u002Fcommanded angular velocity (`robot_dim=4`).\n\nThe actor returns two normalized values in `[-1, 1]`. Before the environment step, map\nonly the first value to linear velocity with `(a[0] + 1) \u002F 2`; pass the second unchanged\nas angular velocity. Do not feed the environment's `[0, 1]`\u002F`[-1, 1]` command back into\nthe actor as a replacement for the normalized action. Assert finite numeric state of\nlength 24 and finite action of length 2 at every step.\n\nThe reference evaluator uses seed `0`, waits about five seconds after environment\nconstruction, and caps an episode at **500 transitions**. A simulator `done` must end an\nepisode earlier; the cap must also end it when the simulator fails to terminate. Reset\nbefore the next episode and reset the per-episode step counter. Preserve the distinction\nbetween a goal\u002Fcollision termination and a synthetic time-limit termination in metrics.\n\n## Checkpoint gates\n\n1. Resolve the requested model directory and base name without guessing another run.\n2. Run the bundled artifact check. Confirm the actor exists, is a regular file, and has\n   the exact expected name. Record critic presence separately.\n3. For a new or untrusted checkpoint, opt into the bounded `--load-state-dict` check.\n   It uses CPU mapping, `weights_only=True` when supported, a file-size limit, and shape\n   checks; it must not fall back to unrestricted pickle loading. A missing, corrupt,\n   non-tensor, extra-key, or shape-incompatible actor is a hard stop.\n4. If the artifact check reports only a critic, stop: a critic cannot produce actions.\n   If the critic is present but incompatible, stop the full-pair evaluation or label the\n   actor-only run explicitly; never silently substitute or rename it.\n5. Instantiate the actor on CPU unless CUDA availability and the requested device have\n   been independently checked. The reference `torch.load` call does not set\n   `map_location`, so a CUDA-origin checkpoint can fail on a CPU-only host; use the\n   safe CPU-mapped validation\u002Fload path and record any device remapping. Load the actor\n   state dict into the same architecture, call `eval()`, and use inference\u002Fno-grad\n   execution. Record the actual device.\n\nSee `references\u002Fcheckpoint-format.md` for key and shape expectations.\n\n## Simulator readiness gate\n\nA real run requires an operational ROS Noetic\u002FGazebo stack, a running ROS master,\nthe Velodyne and odometry topics, the navigation launch asset, and the services used by\nreset, pause, and unpause. The workspace must already be built and sourced by the run\nowner. Check readiness before constructing the environment; do not claim a simulator\nresult from a Python-only import or from a host without ROS\u002FGazebo\u002Fmessage modules.\n\nThe reference environment starts `roscore` and Gazebo in its constructor. The reference\ntest module constructs that environment at import time and then enters an unbounded\n`while True` loop. Therefore do **not** import that module for a smoke test, and do not\nstart an unconstrained production evaluation. Use a controlled runner that loads the\nactor and applies the same distilled contract, with an explicit episode count and a\nwall-clock\u002Fstep watchdog. A safe dry run can stop after artifact and architecture checks\nwithout touching ROS.\n\n## Bounded evaluation workflow\n\n1. **Plan:** choose `N >= 1` episodes, `max_steps \u003C= 500` per episode, a wall-clock\n   deadline, and a cleanup\u002Fabort action. State whether this is actor-only or a complete\n   TD3 artifact evaluation.\n2. **Preflight:** run `check_policy_artifacts.py --model-dir ... --name ...`; add\n   `--load-state-dict` only after the file-size and provenance decision. Verify simulator\n   readiness separately and record failures as blocked, not as poor policy scores.\n3. **Load:** build the exact actor, load only the validated actor state dict, switch to\n   evaluation mode, and initialize deterministic seeds if reproducibility is desired.\n4. **Run:** for each episode, reset the environment, validate the initial 24-value\n   state, call the actor without exploration noise, convert the action, step the\n   simulator, and record reward, collision\u002Fgoal flag, and step count. Stop on `done`,\n   500 steps, watchdog expiry, or operator abort.\n5. **Aggregate:** report per-episode return, length, termination reason, goal count,\n   collision count, time-limit count, invalid-state\u002Faction count, and timeout\u002Fabort\n   status. Do not call a truncated run a completed evaluation.\n6. **Clean up:** stop the controlled runner and simulator processes according to the\n   host's approved procedure. Preserve the checkpoint path\u002Fbase name and run bounds in\n   the result record.\n\nThe original program's continuous loop is useful evidence of intended repeated testing,\nnot an acceptable default for automation. Never omit the bound because the simulator\nusually terminates episodes.\n\n## Failure routing\n\n- Artifact or shape failure: fix\u002Fselect a compatible actor; do not enter ROS.\n- Missing ROS\u002FGazebo, topics, services, or launch asset: mark **environment blocked**;\n  do not report zero reward or policy failure.\n- State\u002Faction contract violation: abort the episode and mark the result invalid.\n- Watchdog or operator stop: preserve partial records and label the run incomplete.\n- A successful artifact check alone proves checkpoint compatibility only; it proves no\n  navigation quality and no simulator behavior.\n\nFor predictable cases and recovery actions, use `references\u002Ftroubleshooting.md`. The\nbundled script is the only practical helper in this skill and is intentionally independent\nof the original checkout.\n\n## Evidence boundary\n\nThis contract was distilled from `TD3\u002Ftest_velodyne_td3.py`, the actor\u002Fsave\u002Fload portions\nof `TD3\u002Ftrain_velodyne_td3.py`, `TD3\u002Fvelodyne_env.py`, and the README's test procedure.\nThose paths are provenance only. Runtime use should rely on this skill and its bundled\nscript rather than opening or executing source-checkout files.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation",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 policy-evaluation",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill policy-evaluation",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill policy-evaluation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002FSKILL.md",[46,52,56,60],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fcheckpoint-format.md","REFERENCE","参考文档",3425,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002Freferences\u002Fcheckpoint-format.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftroubleshooting.md",5172,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002Freferences\u002Ftroubleshooting.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fworkflows.md",4528,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002Freferences\u002Fworkflows.md",{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"scripts\u002Fcheck_policy_artifacts.py","SCRIPT","脚本",14560,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdrl-robot-navigation\u002Fsub-skills\u002Fpolicy-evaluation\u002Fscripts\u002Fcheck_policy_artifacts.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",[],[]]