[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M796SRGS96FEBPRB5YT2KTQW":3,"skill-same:01M796SRGS96FEBPRB5YT2KTQW":67},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M796SRGS96FEBPRB5YT2KTQW","agents-and-llm-orchestration",null,"Build txtai LLM, RAG, and Agent workflows with backend selection, tools, templates, teams, and troubleshooting guidance.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: agents-and-llm-orchestration\ndescription: \"Build txtai LLM, RAG, and Agent workflows with backend selection,\n  tools, templates, teams, and troubleshooting guidance.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# Agents and LLM Orchestration\n\nUse this sub-skill when the task is to connect txtai generation, retrieval, and agentic tool use. It covers `LLM`, `RAG`, and `Agent` construction, YAML configuration, tool definitions, prompt templates, agent teams, backend selection, and optional dependency troubleshooting.\n\nFor low-level embeddings store creation and query tuning, route to [embeddings-search](..\u002Fembeddings-search\u002FSKILL.md). For deterministic pipeline\u002Fworkflow task chains, route to [pipelines-and-workflows](..\u002Fpipelines-and-workflows\u002FSKILL.md). For serving agents\u002FRAG over API, OpenAI-compatible endpoints, or hosted app deployment, route to [api-and-deployment](..\u002Fapi-and-deployment\u002FSKILL.md).\n\n## Quick Routing\n\n| User goal | Use |\n| --- | --- |\n| Generate text directly | `LLM(path=None, method=None, **kwargs)` |\n| Answer with retrieved context | `RAG(similarity, path, template=..., context=..., output=...)` |\n| Let a model choose tools over multiple steps | `Agent(model=..., tools=[...], max_steps=...)` |\n| Build multi-agent research\u002Fdelegation | Agents as tool targets inside another `Agent` |\n| Define via YAML\u002FApplication | top-level `llm:`, `rag:`, and `agent:` sections |\n| Validate RAG config safely | `python scripts\u002Frag_config_template.py --write-template rag_template.py` |\n\n## Minimal Import Check\n\n```python\nfrom txtai import Agent, Embeddings, LLM, RAG\n\nprint(LLM)\nprint(RAG)\nprint(Agent)\n```\n\nIf constructing `Agent(...)` raises `ImportError: smolagents is not available - install \"agent\" extra to enable`, install an agent-capable environment such as `pip install \"txtai[agent]\"`. If LLM backends fail, see [backend-and-tooling](references\u002Fbackend-and-tooling.md).\n\n## Direct LLM Pattern\n\n```python\nfrom txtai import LLM\n\nllm = LLM(\"ibm-granite\u002Fgranite-4.0-350m\")\nanswer = llm(\"Summarize retrieval augmented generation\", maxlength=256)\n```\n\n`LLM` accepts a model `path`, optional `method`, and backend keyword arguments. It supports strings, lists of strings, chat-message dictionaries with `role` and `content`, streaming via `stream=True`, `stop` strings, `defaultrole`, and `stripthink` cleanup. Use `method=\"transformers\"`, `method=\"llama.cpp\"`, `method=\"litellm\"`, or a custom generation class path when inference cannot be inferred from the model path.\n\n## RAG Pattern\n\n```python\nfrom txtai import Embeddings, RAG\n\nembeddings = Embeddings(content=True)\nembeddings.index([\n    (\"install\", \"txtai supports semantic search, RAG, agents, workflows and APIs\"),\n    (\"agent\", \"txtai agents use tools such as Python functions, embeddings stores and web search\"),\n])\n\ntemplate = \"\"\"\nAnswer using only the context.\nQuestion: {question}\nContext: {context}\n\"\"\"\n\nrag = RAG(\n    embeddings,\n    \"ibm-granite\u002Fgranite-4.0-350m\",\n    template=template,\n    system=\"You answer from retrieved context.\",\n    context=3,\n    output=\"flatten\",\n)\n\nresult = rag(\"What can txtai agents use?\", maxlength=256)\n```\n\n`RAG` joins a similarity source with a generator or question-answering pipeline. The similarity source can be an `Embeddings` instance, a `Similarity` pipeline, a callable batch searcher, or an Application-provided embeddings index. Build and persist retrieval stores with [embeddings-search](..\u002Fembeddings-search\u002FSKILL.md) before focusing on prompt generation here.\n\n## Agent Pattern\n\n```python\nfrom datetime import datetime\nfrom txtai import Agent\n\ndef today() -> str:\n    \"\"\"\n    Gets the current date.\n\n    Returns:\n        current date in ISO format\n    \"\"\"\n    return datetime.today().isoformat()\n\nagent = Agent(\n    model=\"Qwen\u002FQwen3-4B-Instruct-2507\",\n    tools=[today, \"websearch\"],\n    max_steps=5,\n)\n\nanswer = agent(\"What is today's date and why might it matter for a news search?\", maxlength=2048)\n```\n\nUse agents for multi-step questions where tool choice matters. Prefer plain `RAG` or deterministic workflows for simple, repeatable flows.\n\n## Tool Shapes\n\n- Direct Python callable: add the function or callable instance to `tools`; type hints and Google-style docstrings improve generated tool schemas.\n- Explicit function dictionary: `{\"name\": \"lookup\", \"description\": \"...\", \"inputs\": {...}, \"target\": callable}`.\n- Embeddings dictionary: `{\"name\": \"kb\", \"description\": \"Searches the knowledge base\", \"path\": \"index-path\"}` or `{\"target\": embeddings}`; results are lists of dicts with `id`, `text`, and `score` when content is available.\n- Built-in shortcuts: `bash`, `defaults`, `edit`, `glob`, `grep`, `python`, `read`, `todowrite`, `websearch`, `webview`, and `write`.\n- MCP tools: HTTP strings such as `http:\u002F\u002Flocalhost:8000\u002Fmcp` load a Model Context Protocol tool collection.\n- Skill files: strings ending in `.md` load a `skill.md`-style tool with frontmatter `name` and `description`.\n\n## Agent Teams\n\nAgents can be tool targets for a supervisor agent.\n\n```python\nfrom txtai import Agent, LLM\n\nllm = LLM(\"Qwen\u002FQwen3-4B-Instruct-2507\")\nwebsearcher = Agent(model=llm, tools=[\"websearch\"], max_steps=3)\n\nsupervisor = Agent(\n    model=llm,\n    tools=[{\n        \"name\": \"websearcher\",\n        \"description\": \"Searches the web and returns concise findings\",\n        \"target\": websearcher,\n    }],\n    max_steps=6,\n)\n```\n\nShare a single `LLM` instance when team members should use the same backend configuration. Keep each agent tool description narrow so the supervisor can choose correctly.\n\n## YAML\u002FApplication Pattern\n\n```yaml\nllm:\n  path: Qwen\u002FQwen3-4B-Instruct-2507\n\nagent:\n  researcher:\n    max_steps: 5\n    tools:\n      - websearch\n      - name: today\n        description: Gets the current date\n        target: package.module.today\n\nrag:\n  path: ibm-granite\u002Fgranite-4.0-350m\n  template: |\n    Answer using only the context.\n    Question: {question}\n    Context: {context}\n  output: flatten\n  context: 3\n```\n\n`Application(config)` resolves configured pipelines first, then agents, then embeddings. A configured `rag:` pipeline can receive the Application embeddings index when its `similarity` is omitted. In `agent:` sections, `target` values are resolved as configured pipelines, workflows, or importable callables.\n\n## RAG Output Modes\n\n| `output` | Shape |\n| --- | --- |\n| `default` | `(name, answer)` tuples, or dicts with `answer` for dict\u002Fstring inputs |\n| `flatten` | answer strings only |\n| `reference` | `(name, answer, reference)` tuples or dicts with references |\n\nRAG input can be a string, a list of strings, tuples shaped like `(name, query, question, snippet)`, or dict rows with `name`, `query`, `question`, and `snippet`. Use `output=\"flatten\"` for simple chat answers and `output=\"reference\"` when the caller needs source ids from retrieved context.\n\n## References\n\n- [LLM, RAG, and Agent recipes](references\u002Fllm-rag-agent-recipes.md) for concrete Python and YAML patterns.\n- [Backend and tooling](references\u002Fbackend-and-tooling.md) for extras, credentials, backends, and tool schemas.\n- [Troubleshooting](references\u002Ftroubleshooting.md) for placeholder agents, template bugs, retrieval quality, output shapes, web\u002Fnetwork failures, and CPU\u002FGPU issues.\n\n## Validation Checklist\n\n- Confirm imports and constructor signatures for `LLM`, `RAG`, and `Agent` in the target environment.\n- For RAG, ensure the retrieval store is content-enabled and the prompt template includes both `{question}` and `{context}`.\n- For agents, verify optional extras by instantiating a tiny `Agent(model=..., tools=[], max_steps=1)` or by importing `smolagents` directly.\n- For hosted LLMs, verify credentials outside source files and run a one-prompt smoke test before wiring tools.\n- Keep source repository docs, examples, and absolute checkout paths out of generated runtime content.\n","# Agents and LLM Orchestration\n\nUse this sub-skill when the task is to connect txtai generation, retrieval, and agentic tool use. It covers `LLM`, `RAG`, and `Agent` construction, YAML configuration, tool definitions, prompt templates, agent teams, backend selection, and optional dependency troubleshooting.\n\nFor low-level embeddings store creation and query tuning, route to [embeddings-search](..\u002Fembeddings-search\u002FSKILL.md). For deterministic pipeline\u002Fworkflow task chains, route to [pipelines-and-workflows](..\u002Fpipelines-and-workflows\u002FSKILL.md). For serving agents\u002FRAG over API, OpenAI-compatible endpoints, or hosted app deployment, route to [api-and-deployment](..\u002Fapi-and-deployment\u002FSKILL.md).\n\n## Quick Routing\n\n| User goal | Use |\n| --- | --- |\n| Generate text directly | `LLM(path=None, method=None, **kwargs)` |\n| Answer with retrieved context | `RAG(similarity, path, template=..., context=..., output=...)` |\n| Let a model choose tools over multiple steps | `Agent(model=..., tools=[...], max_steps=...)` |\n| Build multi-agent research\u002Fdelegation | Agents as tool targets inside another `Agent` |\n| Define via YAML\u002FApplication | top-level `llm:`, `rag:`, and `agent:` sections |\n| Validate RAG config safely | `python scripts\u002Frag_config_template.py --write-template rag_template.py` |\n\n## Minimal Import Check\n\n```python\nfrom txtai import Agent, Embeddings, LLM, RAG\n\nprint(LLM)\nprint(RAG)\nprint(Agent)\n```\n\nIf constructing `Agent(...)` raises `ImportError: smolagents is not available - install \"agent\" extra to enable`, install an agent-capable environment such as `pip install \"txtai[agent]\"`. If LLM backends fail, see [backend-and-tooling](references\u002Fbackend-and-tooling.md).\n\n## Direct LLM Pattern\n\n```python\nfrom txtai import LLM\n\nllm = LLM(\"ibm-granite\u002Fgranite-4.0-350m\")\nanswer = llm(\"Summarize retrieval augmented generation\", maxlength=256)\n```\n\n`LLM` accepts a model `path`, optional `method`, and backend keyword arguments. It supports strings, lists of strings, chat-message dictionaries with `role` and `content`, streaming via `stream=True`, `stop` strings, `defaultrole`, and `stripthink` cleanup. Use `method=\"transformers\"`, `method=\"llama.cpp\"`, `method=\"litellm\"`, or a custom generation class path when inference cannot be inferred from the model path.\n\n## RAG Pattern\n\n```python\nfrom txtai import Embeddings, RAG\n\nembeddings = Embeddings(content=True)\nembeddings.index([\n    (\"install\", \"txtai supports semantic search, RAG, agents, workflows and APIs\"),\n    (\"agent\", \"txtai agents use tools such as Python functions, embeddings stores and web search\"),\n])\n\ntemplate = \"\"\"\nAnswer using only the context.\nQuestion: {question}\nContext: {context}\n\"\"\"\n\nrag = RAG(\n    embeddings,\n    \"ibm-granite\u002Fgranite-4.0-350m\",\n    template=template,\n    system=\"You answer from retrieved context.\",\n    context=3,\n    output=\"flatten\",\n)\n\nresult = rag(\"What can txtai agents use?\", maxlength=256)\n```\n\n`RAG` joins a similarity source with a generator or question-answering pipeline. The similarity source can be an `Embeddings` instance, a `Similarity` pipeline, a callable batch searcher, or an Application-provided embeddings index. Build and persist retrieval stores with [embeddings-search](..\u002Fembeddings-search\u002FSKILL.md) before focusing on prompt generation here.\n\n## Agent Pattern\n\n```python\nfrom datetime import datetime\nfrom txtai import Agent\n\ndef today() -> str:\n    \"\"\"\n    Gets the current date.\n\n    Returns:\n        current date in ISO format\n    \"\"\"\n    return datetime.today().isoformat()\n\nagent = Agent(\n    model=\"Qwen\u002FQwen3-4B-Instruct-2507\",\n    tools=[today, \"websearch\"],\n    max_steps=5,\n)\n\nanswer = agent(\"What is today's date and why might it matter for a news search?\", maxlength=2048)\n```\n\nUse agents for multi-step questions where tool choice matters. Prefer plain `RAG` or deterministic workflows for simple, repeatable flows.\n\n## Tool Shapes\n\n- Direct Python callable: add the function or callable instance to `tools`; type hints and Google-style docstrings improve generated tool schemas.\n- Explicit function dictionary: `{\"name\": \"lookup\", \"description\": \"...\", \"inputs\": {...}, \"target\": callable}`.\n- Embeddings dictionary: `{\"name\": \"kb\", \"description\": \"Searches the knowledge base\", \"path\": \"index-path\"}` or `{\"target\": embeddings}`; results are lists of dicts with `id`, `text`, and `score` when content is available.\n- Built-in shortcuts: `bash`, `defaults`, `edit`, `glob`, `grep`, `python`, `read`, `todowrite`, `websearch`, `webview`, and `write`.\n- MCP tools: HTTP strings such as `http:\u002F\u002Flocalhost:8000\u002Fmcp` load a Model Context Protocol tool collection.\n- Skill files: strings ending in `.md` load a `skill.md`-style tool with frontmatter `name` and `description`.\n\n## Agent Teams\n\nAgents can be tool targets for a supervisor agent.\n\n```python\nfrom txtai import Agent, LLM\n\nllm = LLM(\"Qwen\u002FQwen3-4B-Instruct-2507\")\nwebsearcher = Agent(model=llm, tools=[\"websearch\"], max_steps=3)\n\nsupervisor = Agent(\n    model=llm,\n    tools=[{\n        \"name\": \"websearcher\",\n        \"description\": \"Searches the web and returns concise findings\",\n        \"target\": websearcher,\n    }],\n    max_steps=6,\n)\n```\n\nShare a single `LLM` instance when team members should use the same backend configuration. Keep each agent tool description narrow so the supervisor can choose correctly.\n\n## YAML\u002FApplication Pattern\n\n```yaml\nllm:\n  path: Qwen\u002FQwen3-4B-Instruct-2507\n\nagent:\n  researcher:\n    max_steps: 5\n    tools:\n      - websearch\n      - name: today\n        description: Gets the current date\n        target: package.module.today\n\nrag:\n  path: ibm-granite\u002Fgranite-4.0-350m\n  template: |\n    Answer using only the context.\n    Question: {question}\n    Context: {context}\n  output: flatten\n  context: 3\n```\n\n`Application(config)` resolves configured pipelines first, then agents, then embeddings. A configured `rag:` pipeline can receive the Application embeddings index when its `similarity` is omitted. In `agent:` sections, `target` values are resolved as configured pipelines, workflows, or importable callables.\n\n## RAG Output Modes\n\n| `output` | Shape |\n| --- | --- |\n| `default` | `(name, answer)` tuples, or dicts with `answer` for dict\u002Fstring inputs |\n| `flatten` | answer strings only |\n| `reference` | `(name, answer, reference)` tuples or dicts with references |\n\nRAG input can be a string, a list of strings, tuples shaped like `(name, query, question, snippet)`, or dict rows with `name`, `query`, `question`, and `snippet`. Use `output=\"flatten\"` for simple chat answers and `output=\"reference\"` when the caller needs source ids from retrieved context.\n\n## References\n\n- [LLM, RAG, and Agent recipes](references\u002Fllm-rag-agent-recipes.md) for concrete Python and YAML patterns.\n- [Backend and tooling](references\u002Fbackend-and-tooling.md) for extras, credentials, backends, and tool schemas.\n- [Troubleshooting](references\u002Ftroubleshooting.md) for placeholder agents, template bugs, retrieval quality, output shapes, web\u002Fnetwork failures, and CPU\u002FGPU issues.\n\n## Validation Checklist\n\n- Confirm imports and constructor signatures for `LLM`, `RAG`, and `Agent` in the target environment.\n- For RAG, ensure the retrieval store is content-enabled and the prompt template includes both `{question}` and `{context}`.\n- For agents, verify optional extras by instantiating a tiny `Agent(model=..., tools=[], max_steps=1)` or by importing `smolagents` directly.\n- For hosted LLMs, verify credentials outside source files and run a one-prompt smoke test before wiring tools.\n- Keep source repository docs, examples, and absolute checkout paths out of generated runtime content.\n",{"name":5,"license":23,"metadata":24,"description":7,"disable-model-invocation":26},"Apache 2.0",{"disco-role":25},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":26},[36,39,42],{"tool":37,"command":38},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill agents-and-llm-orchestration",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill agents-and-llm-orchestration",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill agents-and-llm-orchestration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002FSKILL.md",[47,53,57,61],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fbackend-and-tooling.md","REFERENCE","参考文档",7964,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002Freferences\u002Fbackend-and-tooling.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fllm-rag-agent-recipes.md",10254,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002Freferences\u002Fllm-rag-agent-recipes.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",8572,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":63,"resourceTypeDesc":64,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Frag_config_template.py","SCRIPT","脚本",5447,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftxtai\u002Fsub-skills\u002Fagents-and-llm-orchestration\u002Fscripts\u002Frag_config_template.py",[68,75,82,89,96,103],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"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.","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":14,"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":14,"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":14,"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":14,"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":14,"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",[],[]]