[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78THHRWCDFASDAFYV8N00HR":3,"skill-same:01M78THHRWCDFASDAFYV8N00HR":64},{"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},"01M78THHRWCDFASDAFYV8N00HR","llm-and-output",null,"Select Browser Use LLM adapters and configure provider credentials, ChatBrowserUse defaults, structured output, extraction LLMs, fallback models, token\u002Fcost tracking, and model troubleshooting.","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\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: llm-and-output\ndescription: \"Select Browser Use LLM adapters and configure provider\n  credentials, ChatBrowserUse defaults, structured output, extraction LLMs,\n  fallback models, token\u002Fcost tracking, and model troubleshooting.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# LLM and Output\n\nUse this sub-skill when the user is choosing or debugging the model layer for Browser Use: `ChatBrowserUse`, OpenAI, Anthropic, Google\u002FGemini, Azure, Groq, Mistral, Ollama, OpenRouter, Vercel, OCI, AWS\u002FBedrock, structured Pydantic output, page extraction models, fallback LLMs, or token\u002Fcost accounting.\n\n## Route Here\n\n- The request mentions `llm=...`, `ChatBrowserUse`, `ChatOpenAI`, `ChatGoogle`, `ChatAnthropic`, `ChatMistral`, `ChatOllama`, `ChatOpenRouter`, `ChatVercel`, `ChatAzureOpenAI`, `ChatLiteLLM`, or `browser_use.llm.models`.\n- The user asks which model to use, how to set provider API keys, how to pass `api_key`, `base_url`, retry, timeout, temperature, reasoning, or structured-output flags.\n- The user needs `output_model_schema`, `history.structured_output`, `history.get_structured_output(...)`, Pydantic validation, JSON schema repair, or final-result parsing.\n- The task uses `page_extraction_llm`, extraction schema, or a cheaper model for `extract` actions.\n- The task uses `fallback_llm`, model failover, `agent.is_using_fallback_llm`, or `agent.current_llm_model`.\n- The task asks for `calculate_cost`, `pricing_url`, token usage, usage summaries, or cost logs.\n\n## Route Elsewhere\n\n- General `Agent(...)`, task prompting, callbacks, history helpers, run-loop behavior, planning, flash mode, and timeouts belong in `..\u002Fagent-programming\u002FSKILL.md`.\n- Browser\u002Fprofile\u002FCDP\u002Fsession settings, cloud browser sessions, downloads, domains, screenshots, and low-level browser control belong in `..\u002Fbrowser-control\u002FSKILL.md`.\n- Custom actions, default tools, file tools, sensitive data handling, and `ActionResult` belong in `..\u002Ftools-and-actions\u002FSKILL.md`.\n- Terminal `browser-use` or `bu` CLI sessions belong in `..\u002Fcli-and-sessions\u002FSKILL.md`.\n- `@sandbox`, Browser Use Cloud production deployment, MCP, skills, telemetry, and external integrations belong in `..\u002Fproduction-integrations\u002FSKILL.md`.\n\n## Recommended Default\n\nFor new browser automation code, recommend `ChatBrowserUse()` first. It is the Browser Use maintained chat adapter for browser automation and defaults to the optimized `bu-2-0` model when `BROWSER_USE_API_KEY` is set.\n\n```python\nimport asyncio\nfrom browser_use import Agent, ChatBrowserUse\n\nasync def main():\n    agent = Agent(\n        task=\"Open https:\u002F\u002Fexample.com and report the page title.\",\n        llm=ChatBrowserUse(),\n    )\n    history = await agent.run(max_steps=10)\n    print(history.final_result())\n\nasyncio.run(main())\n```\n\nDo not replace model names in existing user code. Users may intentionally test new provider models that are unknown to this skill. If code already specifies a model, preserve it unless the user asks for a recommendation or the model is the direct cause of an error.\n\n## Credential Rules\n\n- Prefer environment variables loaded by the user's app or shell; do not print API keys.\n- `ChatBrowserUse()` reads `BROWSER_USE_API_KEY`; it can also accept `api_key=...` and `base_url=...` explicitly.\n- OpenAI-compatible adapters usually accept `api_key=...`, `base_url=...`, `timeout=...`, and `max_retries=...`.\n- Provider adapters read their conventional env vars when not passed explicitly, such as `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GOOGLE_API_KEY`, `MISTRAL_API_KEY`, `AZURE_OPENAI_API_KEY`, `AZURE_OPENAI_ENDPOINT`, `OPENROUTER_API_KEY`, or `CEREBRAS_API_KEY` depending on the chosen adapter.\n- Optional provider packages may be needed for less common adapters; if an import fails, install the documented extra or switch to a built-in adapter already available in the project environment.\n\nSee `references\u002Fmodels-and-credentials.md` for adapter snippets and credential checks.\n\n## Model Adapter Patterns\n\nUse top-level imports for common adapters:\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI, ChatGoogle, ChatAnthropic\n```\n\nUse `browser_use.llm` for the full adapter surface and lazy imports:\n\n```python\nfrom browser_use.llm import ChatMistral, ChatOllama, ChatOpenRouter\n```\n\nUse `browser_use.llm.models.get_llm_by_name(...)` only when the user wants named aliases resolved from environment variables:\n\n```python\nfrom browser_use.llm.models import get_llm_by_name\nllm = get_llm_by_name(\"bu_2_0\")\n```\n\nSupported adapter families include Browser Use, OpenAI, Azure OpenAI, Google\u002FGemini, Anthropic, AWS\u002FBedrock, DeepSeek, Groq, Mistral, Cerebras, OCI raw, Ollama, OpenRouter, Vercel AI Gateway, and LiteLLM\u002FLangChain compatibility paths.\n\n## Structured Output\n\nUse a Pydantic v2 model class with `output_model_schema=...` when the final answer must be machine-readable. Keep the model small and explicit.\n\n```python\nfrom pydantic import BaseModel\nfrom browser_use import Agent, ChatBrowserUse\n\nclass Product(BaseModel):\n    title: str\n    price: str | None = None\n    in_stock: bool\n\nagent = Agent(\n    task=\"Find the product title, price, and stock status. Return only the requested fields.\",\n    llm=ChatBrowserUse(),\n    output_model_schema=Product,\n)\nhistory = await agent.run(max_steps=25)\nproduct = history.structured_output or Product.model_validate_json(history.final_result())\n```\n\nBrowser Use wires `output_model_schema` into the final `done` action and auto-bridges it to `extraction_schema` when no explicit extraction schema is provided. See `references\u002Fstructured-output.md` for robust parsing, schema design, and validation repair.\n\n## Page Extraction LLM\n\nUse `page_extraction_llm` when normal planning needs a stronger model but page extraction should use a smaller, cheaper, or faster model.\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\n\nagent = Agent(\n    task=\"Find three pricing tiers and extract their names and monthly prices.\",\n    llm=ChatBrowserUse(),\n    page_extraction_llm=ChatOpenAI(model=\"gpt-4.1-mini\"),\n)\n```\n\nThis only changes extraction-style LLM calls used by tools such as `extract`; it does not replace the main planner. Custom tools that request `page_extraction_llm` are owned by `..\u002Ftools-and-actions\u002FSKILL.md`.\n\n## Fallback Models\n\nUse `fallback_llm` for resilience when the primary model fails with provider\u002Frate-limit\u002Fserver-style errors after the primary adapter's own retries are exhausted.\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\n\nagent = Agent(\n    task=\"Open https:\u002F\u002Fexample.com and summarize the page.\",\n    llm=ChatBrowserUse(),\n    fallback_llm=ChatOpenAI(model=\"gpt-4.1-mini\"),\n)\nhistory = await agent.run(max_steps=15)\nif agent.is_using_fallback_llm:\n    print(f\"Switched to {agent.current_llm_model}\")\n```\n\nSee `references\u002Fcost-and-fallbacks.md` for failure classes, retry implications, and cost-aware patterns.\n\n## Cost and Token Tracking\n\nSet `calculate_cost=True` on `Agent` to populate `history.usage` and log usage summaries where adapter usage data is available. Browser Use token costing can use built-in pricing, OpenRouter pricing, custom mappings, and a configurable pricing URL.\n\n```python\nagent = Agent(task=\"...\", llm=ChatBrowserUse(), calculate_cost=True)\nhistory = await agent.run(max_steps=20)\nif history.usage:\n    print(history.usage.total_tokens, history.usage.total_cost)\n```\n\nUse `pricing_url=...` only when the user has a trusted model-pricing source. See `references\u002Fcost-and-fallbacks.md` for fields and caveats.\n\n## Validation Checks\n\nRun these lightweight checks before debugging a full web task:\n\n```bash\npython - \u003C\u003C'PY'\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\nfrom browser_use.llm import ChatGoogle, ChatAnthropic\nprint(Agent, ChatBrowserUse, ChatOpenAI, ChatGoogle, ChatAnthropic)\nPY\n```\n\nCheck credentials without revealing values:\n\n```bash\npython - \u003C\u003C'PY'\nimport os\nfor name in [\"BROWSER_USE_API_KEY\", \"OPENAI_API_KEY\", \"ANTHROPIC_API_KEY\", \"GOOGLE_API_KEY\"]:\n    print(name, \"set\" if os.getenv(name) else \"missing\")\nPY\n```\n\nIf `ChatBrowserUse()` raises `BROWSER_USE_API_KEY is not set`, either set the key or choose a different adapter with its own credentials.\n\n## Troubleshooting First Steps\n\nStart with `references\u002Ftroubleshooting.md`, then narrow the failure:\n\n- Import failure: verify package install, adapter import path, and optional extras for the provider.\n- Authentication failure: confirm the right env var is set in the same shell or process running Python; do not log key contents.\n- Structured-output failure: simplify the Pydantic model, cap steps, inspect `history.final_result()`, then validate with `model_validate_json`.\n- Extraction failure: confirm `page_extraction_llm` is configured and the task tells the agent to use extraction where appropriate.\n- Fallback not used: check whether the primary model is still retrying internally, or whether the error is not a model provider\u002Frate-limit\u002Fserver error.\n- Cost missing: confirm `calculate_cost=True`, the adapter returns usage, and pricing data exists for the model.\n\n## Hard Case Patterns\n\nSupport multi-skill requests by routing only the model\u002Foutput portion here. Examples:\n\n- Authenticated production extraction with `Browser(use_cloud=True)`, `ChatBrowserUse`, a custom 2FA tool, and structured Pydantic output: use this sub-skill for model\u002Foutput schema and fallback; route cloud browser to `..\u002Fbrowser-control\u002FSKILL.md` or production deployment to `..\u002Fproduction-integrations\u002FSKILL.md`; route 2FA tool to `..\u002Ftools-and-actions\u002FSKILL.md`.\n- Failure recovery where a user reports `ValidationError` after `history.final_result()`: use this sub-skill to repair schema\u002Foutput parsing, then route run-loop or browser failures to the appropriate sibling only if model\u002Foutput fixes do not explain the symptom.\n","# LLM and Output\n\nUse this sub-skill when the user is choosing or debugging the model layer for Browser Use: `ChatBrowserUse`, OpenAI, Anthropic, Google\u002FGemini, Azure, Groq, Mistral, Ollama, OpenRouter, Vercel, OCI, AWS\u002FBedrock, structured Pydantic output, page extraction models, fallback LLMs, or token\u002Fcost accounting.\n\n## Route Here\n\n- The request mentions `llm=...`, `ChatBrowserUse`, `ChatOpenAI`, `ChatGoogle`, `ChatAnthropic`, `ChatMistral`, `ChatOllama`, `ChatOpenRouter`, `ChatVercel`, `ChatAzureOpenAI`, `ChatLiteLLM`, or `browser_use.llm.models`.\n- The user asks which model to use, how to set provider API keys, how to pass `api_key`, `base_url`, retry, timeout, temperature, reasoning, or structured-output flags.\n- The user needs `output_model_schema`, `history.structured_output`, `history.get_structured_output(...)`, Pydantic validation, JSON schema repair, or final-result parsing.\n- The task uses `page_extraction_llm`, extraction schema, or a cheaper model for `extract` actions.\n- The task uses `fallback_llm`, model failover, `agent.is_using_fallback_llm`, or `agent.current_llm_model`.\n- The task asks for `calculate_cost`, `pricing_url`, token usage, usage summaries, or cost logs.\n\n## Route Elsewhere\n\n- General `Agent(...)`, task prompting, callbacks, history helpers, run-loop behavior, planning, flash mode, and timeouts belong in `..\u002Fagent-programming\u002FSKILL.md`.\n- Browser\u002Fprofile\u002FCDP\u002Fsession settings, cloud browser sessions, downloads, domains, screenshots, and low-level browser control belong in `..\u002Fbrowser-control\u002FSKILL.md`.\n- Custom actions, default tools, file tools, sensitive data handling, and `ActionResult` belong in `..\u002Ftools-and-actions\u002FSKILL.md`.\n- Terminal `browser-use` or `bu` CLI sessions belong in `..\u002Fcli-and-sessions\u002FSKILL.md`.\n- `@sandbox`, Browser Use Cloud production deployment, MCP, skills, telemetry, and external integrations belong in `..\u002Fproduction-integrations\u002FSKILL.md`.\n\n## Recommended Default\n\nFor new browser automation code, recommend `ChatBrowserUse()` first. It is the Browser Use maintained chat adapter for browser automation and defaults to the optimized `bu-2-0` model when `BROWSER_USE_API_KEY` is set.\n\n```python\nimport asyncio\nfrom browser_use import Agent, ChatBrowserUse\n\nasync def main():\n    agent = Agent(\n        task=\"Open https:\u002F\u002Fexample.com and report the page title.\",\n        llm=ChatBrowserUse(),\n    )\n    history = await agent.run(max_steps=10)\n    print(history.final_result())\n\nasyncio.run(main())\n```\n\nDo not replace model names in existing user code. Users may intentionally test new provider models that are unknown to this skill. If code already specifies a model, preserve it unless the user asks for a recommendation or the model is the direct cause of an error.\n\n## Credential Rules\n\n- Prefer environment variables loaded by the user's app or shell; do not print API keys.\n- `ChatBrowserUse()` reads `BROWSER_USE_API_KEY`; it can also accept `api_key=...` and `base_url=...` explicitly.\n- OpenAI-compatible adapters usually accept `api_key=...`, `base_url=...`, `timeout=...`, and `max_retries=...`.\n- Provider adapters read their conventional env vars when not passed explicitly, such as `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GOOGLE_API_KEY`, `MISTRAL_API_KEY`, `AZURE_OPENAI_API_KEY`, `AZURE_OPENAI_ENDPOINT`, `OPENROUTER_API_KEY`, or `CEREBRAS_API_KEY` depending on the chosen adapter.\n- Optional provider packages may be needed for less common adapters; if an import fails, install the documented extra or switch to a built-in adapter already available in the project environment.\n\nSee `references\u002Fmodels-and-credentials.md` for adapter snippets and credential checks.\n\n## Model Adapter Patterns\n\nUse top-level imports for common adapters:\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI, ChatGoogle, ChatAnthropic\n```\n\nUse `browser_use.llm` for the full adapter surface and lazy imports:\n\n```python\nfrom browser_use.llm import ChatMistral, ChatOllama, ChatOpenRouter\n```\n\nUse `browser_use.llm.models.get_llm_by_name(...)` only when the user wants named aliases resolved from environment variables:\n\n```python\nfrom browser_use.llm.models import get_llm_by_name\nllm = get_llm_by_name(\"bu_2_0\")\n```\n\nSupported adapter families include Browser Use, OpenAI, Azure OpenAI, Google\u002FGemini, Anthropic, AWS\u002FBedrock, DeepSeek, Groq, Mistral, Cerebras, OCI raw, Ollama, OpenRouter, Vercel AI Gateway, and LiteLLM\u002FLangChain compatibility paths.\n\n## Structured Output\n\nUse a Pydantic v2 model class with `output_model_schema=...` when the final answer must be machine-readable. Keep the model small and explicit.\n\n```python\nfrom pydantic import BaseModel\nfrom browser_use import Agent, ChatBrowserUse\n\nclass Product(BaseModel):\n    title: str\n    price: str | None = None\n    in_stock: bool\n\nagent = Agent(\n    task=\"Find the product title, price, and stock status. Return only the requested fields.\",\n    llm=ChatBrowserUse(),\n    output_model_schema=Product,\n)\nhistory = await agent.run(max_steps=25)\nproduct = history.structured_output or Product.model_validate_json(history.final_result())\n```\n\nBrowser Use wires `output_model_schema` into the final `done` action and auto-bridges it to `extraction_schema` when no explicit extraction schema is provided. See `references\u002Fstructured-output.md` for robust parsing, schema design, and validation repair.\n\n## Page Extraction LLM\n\nUse `page_extraction_llm` when normal planning needs a stronger model but page extraction should use a smaller, cheaper, or faster model.\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\n\nagent = Agent(\n    task=\"Find three pricing tiers and extract their names and monthly prices.\",\n    llm=ChatBrowserUse(),\n    page_extraction_llm=ChatOpenAI(model=\"gpt-4.1-mini\"),\n)\n```\n\nThis only changes extraction-style LLM calls used by tools such as `extract`; it does not replace the main planner. Custom tools that request `page_extraction_llm` are owned by `..\u002Ftools-and-actions\u002FSKILL.md`.\n\n## Fallback Models\n\nUse `fallback_llm` for resilience when the primary model fails with provider\u002Frate-limit\u002Fserver-style errors after the primary adapter's own retries are exhausted.\n\n```python\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\n\nagent = Agent(\n    task=\"Open https:\u002F\u002Fexample.com and summarize the page.\",\n    llm=ChatBrowserUse(),\n    fallback_llm=ChatOpenAI(model=\"gpt-4.1-mini\"),\n)\nhistory = await agent.run(max_steps=15)\nif agent.is_using_fallback_llm:\n    print(f\"Switched to {agent.current_llm_model}\")\n```\n\nSee `references\u002Fcost-and-fallbacks.md` for failure classes, retry implications, and cost-aware patterns.\n\n## Cost and Token Tracking\n\nSet `calculate_cost=True` on `Agent` to populate `history.usage` and log usage summaries where adapter usage data is available. Browser Use token costing can use built-in pricing, OpenRouter pricing, custom mappings, and a configurable pricing URL.\n\n```python\nagent = Agent(task=\"...\", llm=ChatBrowserUse(), calculate_cost=True)\nhistory = await agent.run(max_steps=20)\nif history.usage:\n    print(history.usage.total_tokens, history.usage.total_cost)\n```\n\nUse `pricing_url=...` only when the user has a trusted model-pricing source. See `references\u002Fcost-and-fallbacks.md` for fields and caveats.\n\n## Validation Checks\n\nRun these lightweight checks before debugging a full web task:\n\n```bash\npython - \u003C\u003C'PY'\nfrom browser_use import Agent, ChatBrowserUse, ChatOpenAI\nfrom browser_use.llm import ChatGoogle, ChatAnthropic\nprint(Agent, ChatBrowserUse, ChatOpenAI, ChatGoogle, ChatAnthropic)\nPY\n```\n\nCheck credentials without revealing values:\n\n```bash\npython - \u003C\u003C'PY'\nimport os\nfor name in [\"BROWSER_USE_API_KEY\", \"OPENAI_API_KEY\", \"ANTHROPIC_API_KEY\", \"GOOGLE_API_KEY\"]:\n    print(name, \"set\" if os.getenv(name) else \"missing\")\nPY\n```\n\nIf `ChatBrowserUse()` raises `BROWSER_USE_API_KEY is not set`, either set the key or choose a different adapter with its own credentials.\n\n## Troubleshooting First Steps\n\nStart with `references\u002Ftroubleshooting.md`, then narrow the failure:\n\n- Import failure: verify package install, adapter import path, and optional extras for the provider.\n- Authentication failure: confirm the right env var is set in the same shell or process running Python; do not log key contents.\n- Structured-output failure: simplify the Pydantic model, cap steps, inspect `history.final_result()`, then validate with `model_validate_json`.\n- Extraction failure: confirm `page_extraction_llm` is configured and the task tells the agent to use extraction where appropriate.\n- Fallback not used: check whether the primary model is still retrying internally, or whether the error is not a model provider\u002Frate-limit\u002Fserver error.\n- Cost missing: confirm `calculate_cost=True`, the adapter returns usage, and pricing data exists for the model.\n\n## Hard Case Patterns\n\nSupport multi-skill requests by routing only the model\u002Foutput portion here. Examples:\n\n- Authenticated production extraction with `Browser(use_cloud=True)`, `ChatBrowserUse`, a custom 2FA tool, and structured Pydantic output: use this sub-skill for model\u002Foutput schema and fallback; route cloud browser to `..\u002Fbrowser-control\u002FSKILL.md` or production deployment to `..\u002Fproduction-integrations\u002FSKILL.md`; route 2FA tool to `..\u002Ftools-and-actions\u002FSKILL.md`.\n- Failure recovery where a user reports `ValidationError` after `history.final_result()`: use this sub-skill to repair schema\u002Foutput parsing, then route run-loop or browser failures to the appropriate sibling only if model\u002Foutput fixes do not explain the symptom.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output",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 llm-and-output",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill llm-and-output",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill llm-and-output","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002FSKILL.md",[46,52,56,60],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fcost-and-fallbacks.md","REFERENCE","参考文档",7780,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002Freferences\u002Fcost-and-fallbacks.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fmodels-and-credentials.md",7242,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002Freferences\u002Fmodels-and-credentials.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fstructured-output.md",7448,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002Freferences\u002Fstructured-output.md",{"path":61,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Ftroubleshooting.md",8885,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fbrowser-use\u002Fsub-skills\u002Fllm-and-output\u002Freferences\u002Ftroubleshooting.md",[65,73,80,87,94,101],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"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":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"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":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"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":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"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":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"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":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":69,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"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",[],[]]