[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FBYP9WCQPDYB1FX9PFNAYM":3,"skill-same:01M7FBYP9WCQPDYB1FX9PFNAYM":67},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":28,"commitSha":29,"skillDirectory":30,"forksCount":31,"repositoryHtmlUrl":32,"repositoryHomepageUrl":6,"ownerHtmlUrl":33,"ownerBlogUrl":6,"ownerDescription":34,"updatedAt":35,"install":36,"resources":48},"01M7FBYP9WCQPDYB1FX9PFNAYM","training-policy",null,"Teaches safe Rex-Gym legacy PPO training setup and packaged policy playback, including CLI flags, controller-policy mapping, configs, checkpoints, agent counts, and bounded preflight diagnosis.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: training-policy\ndescription: \"Teaches safe Rex-Gym legacy PPO training setup and packaged policy\n  playback, including CLI flags, controller-policy mapping, configs,\n  checkpoints, agent counts, and bounded preflight diagnosis.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# Rex-Gym training and policy operation\n\nUse this route for Rex-Gym PPO training, `rex-gym train`, pretrained `policy`\nplayback, checkpoint or policy-catalog questions, agent counts, log directories,\nopen-loop\u002FIK selection, or TensorFlow graph setup. It covers the legacy\nPython 3.7\u002FTensorFlow 1.x-style runtime and safe preparation; it does not claim\nthat a policy is good, that it transfers to hardware, or that a GUI is\navailable.\n\n## Scope and routing\n\n- Read [CLI reference](references\u002Fcli-reference.md) for the exact Click\n  commands, choices, defaults, and argument mapping.\n- Read [training workflows](references\u002Ftraining-workflows.md) before preparing\n  a log directory or considering a PPO launch.\n- Read [policy playback](references\u002Fpolicy-playback.md) for policy ids,\n  package-data mapping, and the finite playback loop.\n- Read [troubleshooting](references\u002Ftroubleshooting.md) for dependency,\n  checkpoint, controller, and runtime failures.\n- Run the safe [policy catalog inspector](scripts\u002Finspect_policy_catalog.py)\n  from any working directory before using packaged policies. It only reads\n  distribution metadata and small config text; it never trains, opens a GUI,\n  restores a graph, or copies checkpoint data.\n- Route task spaces, terrain semantics, robot\u002Flow-level models, and controller\n  behavior to [simulation environments](..\u002Fsimulation-environments\u002FSKILL.md)\n  and [locomotion modeling](..\u002Flocomotion-modeling\u002FSKILL.md). This route only\n  explains how those choices enter the training\u002Fplayback CLI.\n\nDo not put a large checkpoint into a skill or a new log tree merely to inspect\nit. Treat a missing package asset, missing display, incompatible dependency, or\nuncertain controller\u002Ftask pairing as a preflight stop rather than improvising.\n\n## Install and bounded entry checks\n\nThe package README targets Python 3.7 and the legacy dependency set. The\npublished package command is:\n\n```bash\npython3.7 -m pip install rex_gym\nrex-gym --help\nrex-gym train --help\nrex-gym policy --help\n```\n\n`rex_gym` and `rex-gym` are the underscore\u002Fhyphen spellings of the package\ndistribution; use the spelling accepted by the package index in the chosen\nPython 3.7 environment. TensorFlow 1.15.5 and TensorFlow Probability 0.8 are\nlegacy dependencies. In the verified setup, TensorFlow\u002FTFP imports require\nprotobuf 3.20.3. Do not upgrade this stack casually to modern TensorFlow.\n\nFirst perform bounded checks only:\n\nFrom the generated skill directory, run the bundled helper with a path inside this tree:\n\n```bash\npython sub-skills\u002Ftraining-policy\u002Fscripts\u002Finspect_policy_catalog.py\npython sub-skills\u002Ftraining-policy\u002Fscripts\u002Finspect_policy_catalog.py --policy walk_ik --config-summary\n```\n\nThe second form is an asset\u002Fconfig check, not a playback test. Confirm the\npackage, the requested policy id, all three TensorFlow checkpoint sidecars,\nand a readable config before considering a launch. Also check that the parent\nof the proposed log directory exists and is writable, and choose a positive\nagent count appropriate to the selected config.\n\n## Select the workflow\n\n### Train with PPO\n\nTraining is an explicit, potentially long-running side effect. Ask for a\nbounded run plan and authorization before launching; the ordinary command is:\n\n```bash\nrex-gym train --env walk --log-dir .\u002Frex-logs\n```\n\nFor a rendered one-agent playground session, the README form is:\n\n```bash\nrex-gym train --playground True --env walk --log-dir .\u002Frex-logs\n```\n\nAdd controller, terrain, robot mark, environment arguments, and flags only\nafter validating them against the sibling environment\u002Fmodel routes. A complete\nshape is:\n\n```bash\nrex-gym train --env walk --log-dir .\u002Frex-logs --agents-number 25 \\\n  --inverse-kinematics --terrain plane --mark base \\\n  --arg target_position 1.0 --flag on_rack True\n```\n\n`--log-dir` is required. `--playground True` sets one rendered\u002Fdebug\nenvironment; ordinary training uses the selected agent count or the config\ndefault. The trainer appends a timestamp and `\u003Cenv>_\u003Csignal>` to the log path,\nsaves `config.yaml`, builds the PPO graph, and writes summaries\u002Fcheckpoints.\nThere is no safe dry-run flag in the CLI: use the inspector and `--help`, not a\ntraining command, for bounded verification.\n\n### Play a packaged policy\n\nThe command has no log-dir or agent-count flag:\n\n```bash\nrex-gym policy --env walk --inverse-kinematics --terrain plane --mark base\n```\n\nThe player resolves the signal, loads package `config.yaml`, creates the\nrendered environment, restores the mapped TensorFlow checkpoint, and steps\nuntil the environment returns `done`. This is a GUI\u002Fruntime operation, so do\nnot call it as a smoke test. A display, time, and enough storage for logs or\nruntime caches may be required; a terminal policy check is not a quality\nclaim.\n\n## Mapping facts to preserve\n\n- Environment choices are `gallop`, `walk`, `turn`, `standup`, `go`, and\n  `poses`. With neither controller flag, defaults are respectively `ik`,\n  `ik`, `ol`, `ol`, `ik`, and `ik`.\n- Supported packaged policy ids are `gallop_ol`, `gallop_ik`, `walk_ik`,\n  `walk_ol`, `standup_ol`, `turn_ik`, `turn_ol`, and `poses_ik`. The exact\n  folder\u002Fcheckpoint mapping is in [policy playback](references\u002Fpolicy-playback.md).\n- `go` is present in the environment\u002Fdefault mapping, but `go_ik` is not a\n  supported policy-catalog entry. Do not infer a pretrained Go policy from the\n  environment choice; stop when the catalog\u002Fconfig\u002Fcheckpoint is absent.\n- If both controller flags are supplied, the parser chooses open loop first;\n  it does not enforce mutual exclusion. Use exactly one or neither.\n\n## Graph and agent safety notes\n\n`Trainer` creates `LimitDuration`, `RangeNormalize`, `ClipAction`, and\n`ConvertTo32Bit` wrappers. Training graph construction is placed under\n`\u002Fcpu:0`; the source sets `use_gpu=False`, permits soft placement, and enables\nGPU memory growth in the TensorFlow 1.x session. Non-Windows training can use\nexternal-process environment wrappers; playground mode stays one agent.\n\nThe current code default is `num_agents = 25` and `update_every = 25` (an older\nREADME snippet says 20). `--agents-number` overrides the number used to build\nthe batch, but the source warning checks the config's own divisibility and does\nnot robustly validate every override. Prefer a positive count dividing the\nconfig update interval, and stop on batch-shape or resource errors instead of\nblindly increasing it.\n\nFor failures, use [troubleshooting](references\u002Ftroubleshooting.md), then\nrepeat the catalog check. Do not promise a full training result, policy\nquality, hardware transfer, or GUI playback when the relevant runtime is not\navailable.\n","# Rex-Gym training and policy operation\n\nUse this route for Rex-Gym PPO training, `rex-gym train`, pretrained `policy`\nplayback, checkpoint or policy-catalog questions, agent counts, log directories,\nopen-loop\u002FIK selection, or TensorFlow graph setup. It covers the legacy\nPython 3.7\u002FTensorFlow 1.x-style runtime and safe preparation; it does not claim\nthat a policy is good, that it transfers to hardware, or that a GUI is\navailable.\n\n## Scope and routing\n\n- Read [CLI reference](references\u002Fcli-reference.md) for the exact Click\n  commands, choices, defaults, and argument mapping.\n- Read [training workflows](references\u002Ftraining-workflows.md) before preparing\n  a log directory or considering a PPO launch.\n- Read [policy playback](references\u002Fpolicy-playback.md) for policy ids,\n  package-data mapping, and the finite playback loop.\n- Read [troubleshooting](references\u002Ftroubleshooting.md) for dependency,\n  checkpoint, controller, and runtime failures.\n- Run the safe [policy catalog inspector](scripts\u002Finspect_policy_catalog.py)\n  from any working directory before using packaged policies. It only reads\n  distribution metadata and small config text; it never trains, opens a GUI,\n  restores a graph, or copies checkpoint data.\n- Route task spaces, terrain semantics, robot\u002Flow-level models, and controller\n  behavior to [simulation environments](..\u002Fsimulation-environments\u002FSKILL.md)\n  and [locomotion modeling](..\u002Flocomotion-modeling\u002FSKILL.md). This route only\n  explains how those choices enter the training\u002Fplayback CLI.\n\nDo not put a large checkpoint into a skill or a new log tree merely to inspect\nit. Treat a missing package asset, missing display, incompatible dependency, or\nuncertain controller\u002Ftask pairing as a preflight stop rather than improvising.\n\n## Install and bounded entry checks\n\nThe package README targets Python 3.7 and the legacy dependency set. The\npublished package command is:\n\n```bash\npython3.7 -m pip install rex_gym\nrex-gym --help\nrex-gym train --help\nrex-gym policy --help\n```\n\n`rex_gym` and `rex-gym` are the underscore\u002Fhyphen spellings of the package\ndistribution; use the spelling accepted by the package index in the chosen\nPython 3.7 environment. TensorFlow 1.15.5 and TensorFlow Probability 0.8 are\nlegacy dependencies. In the verified setup, TensorFlow\u002FTFP imports require\nprotobuf 3.20.3. Do not upgrade this stack casually to modern TensorFlow.\n\nFirst perform bounded checks only:\n\nFrom the generated skill directory, run the bundled helper with a path inside this tree:\n\n```bash\npython sub-skills\u002Ftraining-policy\u002Fscripts\u002Finspect_policy_catalog.py\npython sub-skills\u002Ftraining-policy\u002Fscripts\u002Finspect_policy_catalog.py --policy walk_ik --config-summary\n```\n\nThe second form is an asset\u002Fconfig check, not a playback test. Confirm the\npackage, the requested policy id, all three TensorFlow checkpoint sidecars,\nand a readable config before considering a launch. Also check that the parent\nof the proposed log directory exists and is writable, and choose a positive\nagent count appropriate to the selected config.\n\n## Select the workflow\n\n### Train with PPO\n\nTraining is an explicit, potentially long-running side effect. Ask for a\nbounded run plan and authorization before launching; the ordinary command is:\n\n```bash\nrex-gym train --env walk --log-dir .\u002Frex-logs\n```\n\nFor a rendered one-agent playground session, the README form is:\n\n```bash\nrex-gym train --playground True --env walk --log-dir .\u002Frex-logs\n```\n\nAdd controller, terrain, robot mark, environment arguments, and flags only\nafter validating them against the sibling environment\u002Fmodel routes. A complete\nshape is:\n\n```bash\nrex-gym train --env walk --log-dir .\u002Frex-logs --agents-number 25 \\\n  --inverse-kinematics --terrain plane --mark base \\\n  --arg target_position 1.0 --flag on_rack True\n```\n\n`--log-dir` is required. `--playground True` sets one rendered\u002Fdebug\nenvironment; ordinary training uses the selected agent count or the config\ndefault. The trainer appends a timestamp and `\u003Cenv>_\u003Csignal>` to the log path,\nsaves `config.yaml`, builds the PPO graph, and writes summaries\u002Fcheckpoints.\nThere is no safe dry-run flag in the CLI: use the inspector and `--help`, not a\ntraining command, for bounded verification.\n\n### Play a packaged policy\n\nThe command has no log-dir or agent-count flag:\n\n```bash\nrex-gym policy --env walk --inverse-kinematics --terrain plane --mark base\n```\n\nThe player resolves the signal, loads package `config.yaml`, creates the\nrendered environment, restores the mapped TensorFlow checkpoint, and steps\nuntil the environment returns `done`. This is a GUI\u002Fruntime operation, so do\nnot call it as a smoke test. A display, time, and enough storage for logs or\nruntime caches may be required; a terminal policy check is not a quality\nclaim.\n\n## Mapping facts to preserve\n\n- Environment choices are `gallop`, `walk`, `turn`, `standup`, `go`, and\n  `poses`. With neither controller flag, defaults are respectively `ik`,\n  `ik`, `ol`, `ol`, `ik`, and `ik`.\n- Supported packaged policy ids are `gallop_ol`, `gallop_ik`, `walk_ik`,\n  `walk_ol`, `standup_ol`, `turn_ik`, `turn_ol`, and `poses_ik`. The exact\n  folder\u002Fcheckpoint mapping is in [policy playback](references\u002Fpolicy-playback.md).\n- `go` is present in the environment\u002Fdefault mapping, but `go_ik` is not a\n  supported policy-catalog entry. Do not infer a pretrained Go policy from the\n  environment choice; stop when the catalog\u002Fconfig\u002Fcheckpoint is absent.\n- If both controller flags are supplied, the parser chooses open loop first;\n  it does not enforce mutual exclusion. Use exactly one or neither.\n\n## Graph and agent safety notes\n\n`Trainer` creates `LimitDuration`, `RangeNormalize`, `ClipAction`, and\n`ConvertTo32Bit` wrappers. Training graph construction is placed under\n`\u002Fcpu:0`; the source sets `use_gpu=False`, permits soft placement, and enables\nGPU memory growth in the TensorFlow 1.x session. Non-Windows training can use\nexternal-process environment wrappers; playground mode stays one agent.\n\nThe current code default is `num_agents = 25` and `update_every = 25` (an older\nREADME snippet says 20). `--agents-number` overrides the number used to build\nthe batch, but the source warning checks the config's own divisibility and does\nnot robustly validate every override. Prefer a positive count dividing the\nconfig update interval, and stop on batch-shape or resource errors instead of\nblindly increasing it.\n\nFor failures, use [troubleshooting](references\u002Ftroubleshooting.md), then\nrepeat the catalog check. Do not promise a full training result, policy\nquality, hardware transfer, or GUI playback when the relevant runtime is not\navailable.\n",{"name":5,"license":24,"metadata":25,"description":7,"disable-model-invocation":27},"Apache 2.0",{"disco-role":26},"operating",true,"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":37,"prompt":47,"zipAvailable":27},[38,41,44],{"tool":39,"command":40},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill training-policy",{"tool":42,"command":43},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill training-policy",{"tool":45,"command":46},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill training-policy","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002FSKILL.md",[49,55,59,63],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fcli-reference.md","REFERENCE","参考文档",3602,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002Freferences\u002Fcli-reference.md",{"path":56,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Fpolicy-playback.md",3150,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002Freferences\u002Fpolicy-playback.md",{"path":60,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Ftraining-workflows.md",5069,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002Freferences\u002Ftraining-workflows.md",{"path":64,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Ftroubleshooting.md",4870,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F06-前端与设计\u002Frex-gym\u002Fsub-skills\u002Ftraining-policy\u002Freferences\u002Ftroubleshooting.md",[68,76,81,88,95,102],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":77,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":19},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":19},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":72,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":19},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]