[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7F56XF2PPWXC9BQFKBCFQG6":3,"skill-same:01M7F56XF2PPWXC9BQFKBCFQG6":62},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":33,"updatedAt":34,"install":35,"resources":47},"01M7F56XF2PPWXC9BQFKBCFQG6","jax-agent-training",null,"Build and debug public single-agent skrl 2.1.0 reinforcement-learning workflows with JAX, Flax, Optax, JAX models, memories, resources, agents, trainers, checkpoints, and evaluation.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: jax-agent-training\ndescription: \"Build and debug public single-agent skrl 2.1.0\n  reinforcement-learning workflows with JAX, Flax, Optax, JAX models, memories,\n  resources, agents, trainers, checkpoints, and evaluation.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# JAX single-agent training\n\nUse this branch when the requested workflow is a **single-agent** skrl 2.1.0\nworkflow backed by JAX\u002FFlax\u002FOptax: A2C, CEM, DDPG, DDQN, DQN, PPO, RPO, SAC,\nor TD3. It covers model definition, parameter\u002Fstate initialization, JAX\nresources and trainer composition, device and PRNG choices, and the boundary\nbetween training, checkpointing, and evaluation.\n\n## Route before building\n\n1. Select JAX installation and backend policy in\n   [framework selection](..\u002F..\u002Freferences\u002Fframework-selection.md); do not infer\n   CUDA support from a CPU import. JAX's `jaxlib` build and the available\n   `jax.devices()` determine the usable accelerator.\n2. For Gymnasium\u002FGym, vectorization, simulator loaders, or `wrap_env`, use\n   [environment integration](..\u002Fenvironment-integration\u002FSKILL.md). This skill\n   assumes the environment is already wrapped and exposes the relevant spaces\n   and device.\n3. For IPPO\u002FMAPPO, simultaneous scopes, or Runner\u002FYAML component mapping, use\n   [multi-agent and Runner](..\u002Fmulti-agent-and-runner\u002FSKILL.md).\n4. Read [algorithm and component selection](references\u002Falgorithm-and-component-selection.md)\n   to select model keys, mixins, memory type, and the algorithm configuration.\n5. Follow [the workflow\u002FAPI contract](references\u002Fworkflow-and-api.md), then use\n   [JAX troubleshooting](references\u002Ftroubleshooting.md) when a construction,\n   initialization, device, or checkpoint check fails.\n\n## Minimal construction order\n\nFor a normal single-agent build, keep this order:\n\n1. Install the desired JAX\u002FJAXLIB variant first, then install `skrl[jax]`. The\n   package extra supplies JAX, JAXLIB, Flax, and Optax requirements, but the\n   public installation guidance warns that installing JAX first avoids\n   accidentally retaining its CPU build.\n2. Wrap the environment and take `observation_space`, optional `state_space`,\n   `action_space`, `num_envs`, and `device` from that wrapper.\n3. Define Flax models as `Mixin, Model` classes. Call `Model.__init__` before\n   the mixin constructor; give every custom constructor argument a default.\n   Implement Flax `__call__(inputs, role)` and return `(output, {})` or\n   `(output, extra_outputs)`.\n4. Call `model.init_state_dict(role=\u003Cagent-model-key>)` for **every** model\n   before constructing or using the agent. If needed, pass explicit sampled\n   `inputs` and a JAX PRNG `key`; otherwise initialization samples each\n   declared space using the model device and `config.jax.key`.\n5. Create `RandomMemory` for rollouts or replay data when the algorithm needs\n   memory. Instantiate the algorithm's `*_CFG`, set experiment\u002Fresource fields,\n   and construct the JAX agent with the exact model dictionary keys.\n6. Construct `SequentialTrainer` for the ordinary loop or `StepTrainer` when\n   the caller owns one-step train\u002Feval iteration. Call the trainer only after\n   the component\u002Finit smoke passes; this branch's bundled script deliberately\n   does not train.\n7. During evaluation, use the agent checkpoint rather than only a raw model\n   when optimizers or preprocessors matter. Set logging\u002Fcheckpoint intervals\n   to zero for a no-output evaluation smoke.\n\n## Bundled safe check\n\nFrom the generated skill root, run:\n\n```bash\npython sub-skills\u002Fjax-agent-training\u002Fscripts\u002Fjax_ppo_components.py --help\npython sub-skills\u002Fjax-agent-training\u002Fscripts\u002Fjax_ppo_components.py\n```\n\nThe helper constructs a small CPU Pendulum PPO stack, initializes Flax state,\ncreates the JAX memory and PPO agent, initializes a sequential trainer, and\nperforms one model action\u002Fvalue-shape check. It does **not** call `train()` or\n`eval()`, download anything, import a source checkout, create checkpoints, or\nwrite persistent output.\n\n## Scope boundary\n\nThis branch does not own generic framework routing, environment installation or\nwrapping, Torch\u002FWarp APIs, IPPO\u002FMAPPO, Runner YAML, or external simulator\nexecution. Keep those decisions in the linked sibling\u002Froot skills. See the\n[root troubleshooting guide](..\u002F..\u002Freferences\u002Ftroubleshooting.md) for\ncross-framework failures and the sibling skills for their own framework or\nmulti-agent contracts.\n\n## Evidence basis\n\nClaims in this branch are distilled from the public skrl 2.1.0 JAX package APIs,\nJAX model\u002Fagent\u002Fresource documentation, the maintained Gymnasium Pendulum PPO\nconstruction example, JAX configuration tests, JAX model-instantiator tests,\nand the installed public signatures. The exact evidence map and unresolved\nbackend limits are recorded in the branch handoff outside this runtime tree.\n","# JAX single-agent training\n\nUse this branch when the requested workflow is a **single-agent** skrl 2.1.0\nworkflow backed by JAX\u002FFlax\u002FOptax: A2C, CEM, DDPG, DDQN, DQN, PPO, RPO, SAC,\nor TD3. It covers model definition, parameter\u002Fstate initialization, JAX\nresources and trainer composition, device and PRNG choices, and the boundary\nbetween training, checkpointing, and evaluation.\n\n## Route before building\n\n1. Select JAX installation and backend policy in\n   [framework selection](..\u002F..\u002Freferences\u002Fframework-selection.md); do not infer\n   CUDA support from a CPU import. JAX's `jaxlib` build and the available\n   `jax.devices()` determine the usable accelerator.\n2. For Gymnasium\u002FGym, vectorization, simulator loaders, or `wrap_env`, use\n   [environment integration](..\u002Fenvironment-integration\u002FSKILL.md). This skill\n   assumes the environment is already wrapped and exposes the relevant spaces\n   and device.\n3. For IPPO\u002FMAPPO, simultaneous scopes, or Runner\u002FYAML component mapping, use\n   [multi-agent and Runner](..\u002Fmulti-agent-and-runner\u002FSKILL.md).\n4. Read [algorithm and component selection](references\u002Falgorithm-and-component-selection.md)\n   to select model keys, mixins, memory type, and the algorithm configuration.\n5. Follow [the workflow\u002FAPI contract](references\u002Fworkflow-and-api.md), then use\n   [JAX troubleshooting](references\u002Ftroubleshooting.md) when a construction,\n   initialization, device, or checkpoint check fails.\n\n## Minimal construction order\n\nFor a normal single-agent build, keep this order:\n\n1. Install the desired JAX\u002FJAXLIB variant first, then install `skrl[jax]`. The\n   package extra supplies JAX, JAXLIB, Flax, and Optax requirements, but the\n   public installation guidance warns that installing JAX first avoids\n   accidentally retaining its CPU build.\n2. Wrap the environment and take `observation_space`, optional `state_space`,\n   `action_space`, `num_envs`, and `device` from that wrapper.\n3. Define Flax models as `Mixin, Model` classes. Call `Model.__init__` before\n   the mixin constructor; give every custom constructor argument a default.\n   Implement Flax `__call__(inputs, role)` and return `(output, {})` or\n   `(output, extra_outputs)`.\n4. Call `model.init_state_dict(role=\u003Cagent-model-key>)` for **every** model\n   before constructing or using the agent. If needed, pass explicit sampled\n   `inputs` and a JAX PRNG `key`; otherwise initialization samples each\n   declared space using the model device and `config.jax.key`.\n5. Create `RandomMemory` for rollouts or replay data when the algorithm needs\n   memory. Instantiate the algorithm's `*_CFG`, set experiment\u002Fresource fields,\n   and construct the JAX agent with the exact model dictionary keys.\n6. Construct `SequentialTrainer` for the ordinary loop or `StepTrainer` when\n   the caller owns one-step train\u002Feval iteration. Call the trainer only after\n   the component\u002Finit smoke passes; this branch's bundled script deliberately\n   does not train.\n7. During evaluation, use the agent checkpoint rather than only a raw model\n   when optimizers or preprocessors matter. Set logging\u002Fcheckpoint intervals\n   to zero for a no-output evaluation smoke.\n\n## Bundled safe check\n\nFrom the generated skill root, run:\n\n```bash\npython sub-skills\u002Fjax-agent-training\u002Fscripts\u002Fjax_ppo_components.py --help\npython sub-skills\u002Fjax-agent-training\u002Fscripts\u002Fjax_ppo_components.py\n```\n\nThe helper constructs a small CPU Pendulum PPO stack, initializes Flax state,\ncreates the JAX memory and PPO agent, initializes a sequential trainer, and\nperforms one model action\u002Fvalue-shape check. It does **not** call `train()` or\n`eval()`, download anything, import a source checkout, create checkpoints, or\nwrite persistent output.\n\n## Scope boundary\n\nThis branch does not own generic framework routing, environment installation or\nwrapping, Torch\u002FWarp APIs, IPPO\u002FMAPPO, Runner YAML, or external simulator\nexecution. Keep those decisions in the linked sibling\u002Froot skills. See the\n[root troubleshooting guide](..\u002F..\u002Freferences\u002Ftroubleshooting.md) for\ncross-framework failures and the sibling skills for their own framework or\nmulti-agent contracts.\n\n## Evidence basis\n\nClaims in this branch are distilled from the public skrl 2.1.0 JAX package APIs,\nJAX model\u002Fagent\u002Fresource documentation, the maintained Gymnasium Pendulum PPO\nconstruction example, JAX configuration tests, JAX model-instantiator tests,\nand the installed public signatures. The exact evidence map and unresolved\nbackend limits are recorded in the branch handoff outside this runtime tree.\n",{"name":5,"license":15,"metadata":24,"description":7,"disable-model-invocation":26},{"disco-role":25},"operating",true,"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training",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":36,"prompt":46,"zipAvailable":26},[37,40,43],{"tool":38,"command":39},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill jax-agent-training",{"tool":41,"command":42},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill jax-agent-training",{"tool":44,"command":45},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill jax-agent-training","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training\u002FSKILL.md",[48,54,58],{"path":49,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Falgorithm-and-component-selection.md","REFERENCE","参考文档",9554,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training\u002Freferences\u002Falgorithm-and-component-selection.md",{"path":55,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":56,"sourceHtmlUrl":57},"references\u002Ftroubleshooting.md",11132,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training\u002Freferences\u002Ftroubleshooting.md",{"path":59,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":60,"sourceHtmlUrl":61},"references\u002Fworkflow-and-api.md",12261,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F05-技能与代理开发\u002Fskrl\u002Fsub-skills\u002Fjax-agent-training\u002Freferences\u002Fworkflow-and-api.md",[63,70,75,82,89,96],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"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.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":71,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"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":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"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":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"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":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"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":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"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",[],[]]