[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M70FR610TDMB1K5ZCCJGH77H":3,"skill-same:01M70FR610TDMB1K5ZCCJGH77H":129},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":49,"commitSha":50,"skillDirectory":51,"forksCount":52,"repositoryHtmlUrl":53,"repositoryHomepageUrl":54,"ownerHtmlUrl":55,"ownerBlogUrl":56,"ownerDescription":57,"updatedAt":58,"install":59,"resources":72},"01M70FR610TDMB1K5ZCCJGH77H","ai-agent-frameworks",null,"Expert guidance on building production-ready multi-agent AI systems using CrewAI, LangChain, AutoGen, and custom architectures. Use when building agent systems, selecting frameworks, designing multi-agent workflows, debugging agent behavior, or deploying agents to production.","shengdabai\u002FTony-Claude-Code-Skills","shengdabai","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F116417419?v=4","01M70F6D2WY5PQV5A92BKYJQEW","Tony Sheng","01M70F6D3GAAAYZ5H4CC1G8VXM",3,"MIT","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"ai agent","agent frameworks","frameworks","ai agent frameworks","systems","behavior","agent workflows","building agent systems","multi","use","custom","architectures","langchain","multi-agent","when","workflows","debugging","production-ready","expert","ready","2026-11-10 10:12:30",false,"---\nname: ai-agent-frameworks\ndescription: Expert guidance on building production-ready multi-agent AI systems using CrewAI, LangChain, AutoGen, and custom architectures. Use when building agent systems, selecting frameworks, designing multi-agent workflows, debugging agent behavior, or deploying agents to production.\n---\n\n# AI Agent Frameworks Skill\n\n## When to Use This Skill\n\nInvoke this skill when working with multi-agent AI systems, including:\n\n- Selecting between CrewAI, LangChain, AutoGen, LangGraph, or custom agent architectures\n- Designing multi-agent collaboration patterns\n- Implementing agent workflows with proper observability\n- Debugging stuck agents, infinite loops, or runaway costs\n- Integrating agents with MCP servers and external tools\n- Deploying agent systems to production\n- Recognizing when agents add unnecessary complexity\n\n## Framework Selection Decision Process\n\n### Decision Tree\n\nStart by determining if agents are truly needed:\n\n**Question 1: Is multi-step reasoning or tool orchestration required?**\n- If NO → Use a single LLM call or simple chain\n- If YES → Continue\n\n**Question 2: Does the workflow map to clear team roles (research, write, edit)?**\n- If YES → Prefer **CrewAI** for its role-based simplicity\n- If NO → Continue\n\n**Question 3: Is RAG (Retrieval-Augmented Generation) central to the application?**\n- If YES → Prefer **LangChain** for best-in-class RAG capabilities\n- If NO → Continue\n\n**Question 4: Do agents need conversational debate or consensus-reaching?**\n- If YES → Consider **AutoGen** for conversation patterns\n- If NO → Continue\n\n**Question 5: Does the workflow require complex state machines with loops and branching?**\n- If YES → Consider **LangGraph** for stateful control\n- If NO → Continue\n\n**Question 6: Are minimal dependencies and maximum control essential?**\n- If YES → Build **custom** agent system\n- If NO → Default to **LangChain** (most versatile)\n\n### Quick Framework Comparison Reference\n\nFor detailed comparisons, refer to `references\u002Fframework-comparison.md`\n\n- **CrewAI**: Best for role-based teams, sequential workflows, quick prototypes\n- **LangChain**: Best for RAG, extensive integrations, production tooling (LangSmith)\n- **AutoGen**: Best for conversational agents, code execution, research projects\n- **LangGraph**: Best for complex stateful workflows requiring loops and retries\n- **Custom**: Best for simple use cases, performance-critical systems, unique requirements\n\n## Agent Architecture Patterns\n\n### Single-Agent vs Multi-Agent Criteria\n\nUse **single agent** when:\n- Task is straightforward with clear steps\n- No benefit from specialization\n- Cost and complexity must be minimized\n\nUse **multi-agent** when:\n- Tasks benefit from specialized expertise (research vs writing vs review)\n- Parallel execution would improve performance\n- Different agents need different models or tools\n- Workflow involves handoffs or collaboration\n\n### Common Orchestration Patterns\n\n**Sequential Pattern:**\n```\nAgent A → Agent B → Agent C\n```\n- Use when each step depends on previous output\n- Examples: Research → Write → Edit → Publish\n\n**Parallel Pattern:**\n```\n      → Agent A →\nInput → Agent B → Merge → Output\n      → Agent C →\n```\n- Use when tasks are independent and can run concurrently\n- Examples: Parallel research from multiple sources\n\n**Hierarchical Pattern:**\n```\n     Manager Agent\n    \u002F      |      \\\nAgent A  Agent B  Agent C\n```\n- Use when dynamic task allocation is needed\n- Manager delegates and coordinates\n\n**Conversational Pattern:**\n```\nAgent A ⟷ Agent B ⟷ Agent C\n```\n- Use for debate, consensus, or iterative refinement\n- Common in AutoGen workflows\n\n## Framework-Specific Guidance\n\n### When Working with CrewAI\n\nRefer to `references\u002Fcrewai-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Design agents with specific roles (not generic \"helper\")\n- Keep 2-5 agents per workflow (avoid over-engineering)\n- Use sequential process for predictable workflows\n- Use hierarchical process when dynamic delegation is needed\n- Set `max_iterations` to prevent runaway costs\n\n**Common anti-pattern to avoid:**\n```python\n# ❌ Don't: Vague agents with overlapping responsibilities\nagent = Agent(role=\"Helper\", goal=\"Help with stuff\")\n\n# ✅ Do: Specific agents with clear boundaries\nresearcher = Agent(\n    role=\"Technical Researcher\",\n    goal=\"Find accurate data on Python frameworks\",\n    tools=[search_tool, docs_tool]\n)\n```\n\n### When Working with LangChain\n\nRefer to `references\u002Flangchain-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Always set `max_iterations` on AgentExecutor to prevent loops\n- Use LCEL (`|` operator) for composing chains\n- Enable LangSmith tracing in production\n- Implement proper error handling with `handle_parsing_errors=True`\n- Consider using GPT-3.5-turbo for simple agents to reduce costs\n\n**RAG pattern (common use case):**\n```python\nfrom langchain.chains import RetrievalQA\n\nqa_chain = RetrievalQA.from_chain_type(\n    llm=llm,\n    retriever=vectorstore.as_retriever(search_kwargs={\"k\": 4}),\n    return_source_documents=True\n)\n```\n\n**Critical: Always limit iterations:**\n```python\nagent_executor = AgentExecutor(\n    agent=agent,\n    tools=tools,\n    max_iterations=10,  # REQUIRED\n    max_execution_time=60,\n    handle_parsing_errors=True\n)\n```\n\n### When Working with AutoGen\n\nRefer to `references\u002Fautogen-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Always use Docker for code execution in production (`use_docker=True`)\n- Set `max_round` on GroupChat to prevent infinite conversations\n- Use `human_input_mode` appropriately (NEVER, TERMINATE, or ALWAYS)\n- Keep group chats to 3-5 agents maximum\n- Track token usage as conversations can be expensive\n\n**Safe code execution:**\n```python\nuser_proxy = UserProxyAgent(\n    name=\"executor\",\n    code_execution_config={\n        \"use_docker\": True,  # CRITICAL for production\n        \"timeout\": 60\n    }\n)\n```\n\n### When Building Custom Agents\n\nRefer to `references\u002Fcustom-agent-guide.md` for complete implementations.\n\n**Build custom when:**\n- Single-purpose agent with minimal complexity\n- Framework overhead is unjustified\n- Need maximum performance control\n- Learning exercise to understand agents deeply\n\n**Minimal implementation pattern:**\n1. Prompt template\n2. LLM client with retry logic\n3. Tool execution with error handling\n4. Response parsing\n\n**Production-ready custom agent includes:**\n- Timeout protection\n- Retry logic with exponential backoff\n- Cost tracking\n- Comprehensive logging\n- Circuit breakers for external APIs\n\n## Debugging Stuck or Looping Agents\n\n### Detection Strategies\n\n**Loop detection:**\n- Track action history (last 3-5 actions)\n- Alert when same action repeats consecutively\n- Implement max_iterations as hard stop\n\n**Stuck detection:**\n- Set execution timeouts (30-60 seconds per agent)\n- Monitor for agents waiting indefinitely\n- Check for missing tool results\n\n**Tools available:**\n- Use `scripts\u002Fagent-debugger.py` to analyze trace files\n- Examine span durations to identify bottlenecks\n- Look for repeated patterns in execution logs\n\n### Common Causes and Solutions\n\n**Cause:** Agent keeps calling the same tool with same parameters\n**Solution:** Improve prompt to guide agent away from repetition, implement action history in context\n\n**Cause:** Agent waiting for tool that never completes\n**Solution:** Add timeouts to all tool calls, implement circuit breakers\n\n**Cause:** Agent confused about when to terminate\n**Solution:** Clear termination conditions, explicit TERMINATE signals\n\n## Tool Integration Best Practices\n\n### MCP Server Integration\n\nWhen integrating MCP servers with agent frameworks:\n\n**With LangChain:**\n```python\nfrom langchain.tools import Tool\n\n# Wrap MCP server function as LangChain tool\nmcp_tool = Tool(\n    name=\"mcp_server_query\",\n    func=mcp_server.query,\n    description=\"Query the MCP server for data\"\n)\n\n# Add to agent\nagent = create_openai_functions_agent(llm, tools=[mcp_tool], prompt=prompt)\n```\n\n**With CrewAI:**\n- Define MCP functions as @tool decorated functions\n- Register with specific agents that need MCP access\n\n**General principles:**\n- Always handle MCP server errors gracefully (return error strings, not exceptions)\n- Set timeouts on MCP calls (don't let agents hang)\n- Log all MCP interactions for debugging\n\n### Tool Error Handling Pattern\n\n```python\n@tool\ndef resilient_tool(query: str) -> str:\n    \"\"\"Tool with proper error handling\"\"\"\n    try:\n        result = external_api.call(query, timeout=10)\n        return json.dumps(result)\n    except TimeoutError:\n        return \"Error: Request timed out. Please try again.\"\n    except Exception as e:\n        logger.error(f\"Tool failed: {e}\")\n        return f\"Error: {str(e)}\"\n```\n\n**Never let tools crash the agent** - always return strings, even for errors.\n\n## Cost Optimization\n\n### Estimation Before Building\n\nUse `scripts\u002Fcost-estimator.py` to estimate costs before implementing:\n\n```bash\npython scripts\u002Fcost-estimator.py --config workflow.json --requests 1000 --optimize\n```\n\n### Optimization Strategies\n\n**1. Use cheaper models for simple agents:**\n- GPT-4 for complex reasoning (strategist, analyst)\n- GPT-3.5-turbo for simple tasks (formatter, router)\n\n**2. Limit iterations strictly:**\n- Set `max_iterations` to 5-10 (not 50)\n- Set `max_execution_time` to prevent runaway costs\n\n**3. Cache repeated work:**\n- Enable LLM response caching\n- Store results of expensive operations\n- Reuse research across similar queries\n\n**4. Implement budgets:**\n```python\ncost_tracker = CostTracker()\nif cost_tracker.total_cost > budget:\n    raise BudgetExceededError()\n```\n\n**5. Monitor costs in real-time:**\n- Track token usage per agent\n- Alert when costs approach thresholds\n- Analyze cost breakdowns to identify expensive agents\n\n## Production Deployment Considerations\n\nRefer to `references\u002Fproduction-patterns.md` for deployment architectures and `references\u002Fobservability-guide.md` for monitoring.\n\n### Essential Production Patterns\n\n**1. Async Task Queue (Recommended):**\n- Use Celery + Redis for long-running workflows\n- Return task ID immediately, poll for results\n- Enables horizontal scaling\n\n**2. Always Enable Observability:**\n- Structured logging (JSON format)\n- Distributed tracing (OpenTelemetry or LangSmith)\n- Cost tracking per request\n- Error monitoring (Sentry)\n\n**3. Implement Fault Tolerance:**\n- Retry with exponential backoff\n- Circuit breakers for external APIs\n- Graceful degradation (fallback to cheaper models)\n\n**4. Security:**\n- API authentication (JWT tokens)\n- Rate limiting (10-100 requests\u002Fminute per user)\n- Input validation (prevent prompt injection)\n- Sandboxed code execution (Docker for AutoGen)\n\n**5. Set Hard Limits:**\n```python\nLIMITS = {\n    \"max_iterations\": 10,\n    \"max_execution_time\": 60,  # seconds\n    \"max_cost_per_request\": 1.0,  # USD\n    \"max_concurrent_agents\": 5\n}\n```\n\n## Anti-Patterns to Avoid\n\n### ❌ Over-Engineering with Agents\n\n**Problem:** Using multiple agents when a single LLM call would suffice\n\n**Example:**\n```python\n# ❌ Overkill for simple task\nagents = [InputParser(), Processor(), OutputFormatter()]\n# Just to format text!\n```\n\n**Solution:** Start simple. One LLM call. Add agents only when truly beneficial.\n\n### ❌ No Termination Conditions\n\n**Problem:** Agents run forever or hit arbitrary iteration limits\n\n**Solution:**\n```python\n# ✅ Always set termination conditions\nmax_iterations=10\nis_termination_msg=lambda msg: \"TERMINATE\" in msg.get(\"content\", \"\")\n```\n\n### ❌ Vague Agent Roles\n\n**Problem:** Agents with unclear responsibilities cause overlap and confusion\n\n**Solution:** Specific roles with clear boundaries. \"Senior Python Developer\" not \"Helper\".\n\n### ❌ Ignoring Costs\n\n**Problem:** Running expensive models without tracking or budgets\n\n**Solution:** Track costs per request, set budgets, alert on anomalies\n\n### ❌ No Observability\n\n**Problem:** Can't debug when things go wrong\n\n**Solution:** Implement logging, tracing, and metrics from day one\n\n## When NOT to Use Agents\n\nRecognize situations where agents add unnecessary complexity:\n\n**Use a single LLM call instead of agents when:**\n- Task is straightforward (translation, summarization, Q&A)\n- No tool usage required\n- Single-step reasoning suffices\n- Cost and latency must be minimized\n\n**Example decision:**\n- \"Translate this text\" → Single LLM call\n- \"Research 10 sources and synthesize findings\" → Multi-agent system\n\n## Resource Reference Guide\n\nThis skill includes comprehensive reference materials. Load them as needed:\n\n### Framework Comparisons\n- **`references\u002Fframework-comparison.md`** - Detailed CrewAI vs LangChain vs AutoGen comparison with migration paths\n\n### Framework-Specific Patterns\n- **`references\u002Fcrewai-patterns.md`** - Role design, task patterns, processes, production tips\n- **`references\u002Flangchain-patterns.md`** - LCEL chains, RAG patterns, LangGraph workflows\n- **`references\u002Fautogen-patterns.md`** - Conversational agents, group chats, code execution\n\n### Advanced Topics\n- **`references\u002Fcustom-agent-guide.md`** - Building agents from scratch, minimal implementations\n- **`references\u002Fobservability-guide.md`** - Tracing, logging, metrics, debugging tools\n- **`references\u002Fproduction-patterns.md`** - Deployment architectures, scaling, security, monitoring\n\n### Helper Scripts\n- **`scripts\u002Fcost-estimator.py`** - Estimate workflow costs before building\n- **`scripts\u002Fagent-debugger.py`** - Analyze traces to identify loops and bottlenecks\n- **`scripts\u002Fagent-system-template\u002F`** - Boilerplate for new agent systems\n\n## Getting Started Workflow\n\nFor new agent system development:\n\n1. **Validate need for agents** - Can this be solved with a single LLM call?\n2. **Choose framework** - Use decision tree above\n3. **Estimate costs** - Run cost-estimator.py with expected workflow\n4. **Implement with observability** - Enable logging\u002Ftracing from start\n5. **Test for loops** - Use agent-debugger.py on traces\n6. **Set hard limits** - max_iterations, budgets, timeouts\n7. **Monitor in production** - Track costs, errors, performance\n\n## Framework-Specific Quick References\n\n### CrewAI Essentials\n- Agents need: role, goal, backstory, tools\n- Tasks need: description, expected_output, agent\n- Processes: Sequential (linear) or Hierarchical (manager delegates)\n- Limit: 2-5 agents per crew\n\n### LangChain Essentials\n- Always set max_iterations on AgentExecutor\n- Use LCEL for chain composition: `prompt | model | parser`\n- Enable LangSmith for production: `LANGCHAIN_TRACING_V2=true`\n- RAG: Use RetrievalQA or ConversationalRetrievalChain\n\n### AutoGen Essentials\n- UserProxyAgent executes code, AssistantAgent generates it\n- Always use Docker in production: `use_docker=True`\n- GroupChat needs max_round limit\n- Human input modes: NEVER (autonomous), TERMINATE (confirm end), ALWAYS (every step)\n\n### LangGraph Essentials\n- Define state, nodes (functions), and edges (transitions)\n- Use conditional edges for branching logic\n- Supports cycles (retries, loops)\n- Requires StateGraph setup and compilation\n\n## Summary\n\nMulti-agent systems add complexity. Use this skill to:\n\n✅ Make informed framework selections\n✅ Design effective agent architectures\n✅ Avoid common pitfalls (loops, runaway costs)\n✅ Implement production-ready observability\n✅ Recognize when agents are overkill\n\nPrioritize simplicity. Start with the minimum viable agent system. Add complexity only when justified by concrete benefits.\n","# AI Agent Frameworks Skill\n\n## When to Use This Skill\n\nInvoke this skill when working with multi-agent AI systems, including:\n\n- Selecting between CrewAI, LangChain, AutoGen, LangGraph, or custom agent architectures\n- Designing multi-agent collaboration patterns\n- Implementing agent workflows with proper observability\n- Debugging stuck agents, infinite loops, or runaway costs\n- Integrating agents with MCP servers and external tools\n- Deploying agent systems to production\n- Recognizing when agents add unnecessary complexity\n\n## Framework Selection Decision Process\n\n### Decision Tree\n\nStart by determining if agents are truly needed:\n\n**Question 1: Is multi-step reasoning or tool orchestration required?**\n- If NO → Use a single LLM call or simple chain\n- If YES → Continue\n\n**Question 2: Does the workflow map to clear team roles (research, write, edit)?**\n- If YES → Prefer **CrewAI** for its role-based simplicity\n- If NO → Continue\n\n**Question 3: Is RAG (Retrieval-Augmented Generation) central to the application?**\n- If YES → Prefer **LangChain** for best-in-class RAG capabilities\n- If NO → Continue\n\n**Question 4: Do agents need conversational debate or consensus-reaching?**\n- If YES → Consider **AutoGen** for conversation patterns\n- If NO → Continue\n\n**Question 5: Does the workflow require complex state machines with loops and branching?**\n- If YES → Consider **LangGraph** for stateful control\n- If NO → Continue\n\n**Question 6: Are minimal dependencies and maximum control essential?**\n- If YES → Build **custom** agent system\n- If NO → Default to **LangChain** (most versatile)\n\n### Quick Framework Comparison Reference\n\nFor detailed comparisons, refer to `references\u002Fframework-comparison.md`\n\n- **CrewAI**: Best for role-based teams, sequential workflows, quick prototypes\n- **LangChain**: Best for RAG, extensive integrations, production tooling (LangSmith)\n- **AutoGen**: Best for conversational agents, code execution, research projects\n- **LangGraph**: Best for complex stateful workflows requiring loops and retries\n- **Custom**: Best for simple use cases, performance-critical systems, unique requirements\n\n## Agent Architecture Patterns\n\n### Single-Agent vs Multi-Agent Criteria\n\nUse **single agent** when:\n- Task is straightforward with clear steps\n- No benefit from specialization\n- Cost and complexity must be minimized\n\nUse **multi-agent** when:\n- Tasks benefit from specialized expertise (research vs writing vs review)\n- Parallel execution would improve performance\n- Different agents need different models or tools\n- Workflow involves handoffs or collaboration\n\n### Common Orchestration Patterns\n\n**Sequential Pattern:**\n```\nAgent A → Agent B → Agent C\n```\n- Use when each step depends on previous output\n- Examples: Research → Write → Edit → Publish\n\n**Parallel Pattern:**\n```\n      → Agent A →\nInput → Agent B → Merge → Output\n      → Agent C →\n```\n- Use when tasks are independent and can run concurrently\n- Examples: Parallel research from multiple sources\n\n**Hierarchical Pattern:**\n```\n     Manager Agent\n    \u002F      |      \\\nAgent A  Agent B  Agent C\n```\n- Use when dynamic task allocation is needed\n- Manager delegates and coordinates\n\n**Conversational Pattern:**\n```\nAgent A ⟷ Agent B ⟷ Agent C\n```\n- Use for debate, consensus, or iterative refinement\n- Common in AutoGen workflows\n\n## Framework-Specific Guidance\n\n### When Working with CrewAI\n\nRefer to `references\u002Fcrewai-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Design agents with specific roles (not generic \"helper\")\n- Keep 2-5 agents per workflow (avoid over-engineering)\n- Use sequential process for predictable workflows\n- Use hierarchical process when dynamic delegation is needed\n- Set `max_iterations` to prevent runaway costs\n\n**Common anti-pattern to avoid:**\n```python\n# ❌ Don't: Vague agents with overlapping responsibilities\nagent = Agent(role=\"Helper\", goal=\"Help with stuff\")\n\n# ✅ Do: Specific agents with clear boundaries\nresearcher = Agent(\n    role=\"Technical Researcher\",\n    goal=\"Find accurate data on Python frameworks\",\n    tools=[search_tool, docs_tool]\n)\n```\n\n### When Working with LangChain\n\nRefer to `references\u002Flangchain-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Always set `max_iterations` on AgentExecutor to prevent loops\n- Use LCEL (`|` operator) for composing chains\n- Enable LangSmith tracing in production\n- Implement proper error handling with `handle_parsing_errors=True`\n- Consider using GPT-3.5-turbo for simple agents to reduce costs\n\n**RAG pattern (common use case):**\n```python\nfrom langchain.chains import RetrievalQA\n\nqa_chain = RetrievalQA.from_chain_type(\n    llm=llm,\n    retriever=vectorstore.as_retriever(search_kwargs={\"k\": 4}),\n    return_source_documents=True\n)\n```\n\n**Critical: Always limit iterations:**\n```python\nagent_executor = AgentExecutor(\n    agent=agent,\n    tools=tools,\n    max_iterations=10,  # REQUIRED\n    max_execution_time=60,\n    handle_parsing_errors=True\n)\n```\n\n### When Working with AutoGen\n\nRefer to `references\u002Fautogen-patterns.md` for comprehensive patterns.\n\n**Key principles:**\n- Always use Docker for code execution in production (`use_docker=True`)\n- Set `max_round` on GroupChat to prevent infinite conversations\n- Use `human_input_mode` appropriately (NEVER, TERMINATE, or ALWAYS)\n- Keep group chats to 3-5 agents maximum\n- Track token usage as conversations can be expensive\n\n**Safe code execution:**\n```python\nuser_proxy = UserProxyAgent(\n    name=\"executor\",\n    code_execution_config={\n        \"use_docker\": True,  # CRITICAL for production\n        \"timeout\": 60\n    }\n)\n```\n\n### When Building Custom Agents\n\nRefer to `references\u002Fcustom-agent-guide.md` for complete implementations.\n\n**Build custom when:**\n- Single-purpose agent with minimal complexity\n- Framework overhead is unjustified\n- Need maximum performance control\n- Learning exercise to understand agents deeply\n\n**Minimal implementation pattern:**\n1. Prompt template\n2. LLM client with retry logic\n3. Tool execution with error handling\n4. Response parsing\n\n**Production-ready custom agent includes:**\n- Timeout protection\n- Retry logic with exponential backoff\n- Cost tracking\n- Comprehensive logging\n- Circuit breakers for external APIs\n\n## Debugging Stuck or Looping Agents\n\n### Detection Strategies\n\n**Loop detection:**\n- Track action history (last 3-5 actions)\n- Alert when same action repeats consecutively\n- Implement max_iterations as hard stop\n\n**Stuck detection:**\n- Set execution timeouts (30-60 seconds per agent)\n- Monitor for agents waiting indefinitely\n- Check for missing tool results\n\n**Tools available:**\n- Use `scripts\u002Fagent-debugger.py` to analyze trace files\n- Examine span durations to identify bottlenecks\n- Look for repeated patterns in execution logs\n\n### Common Causes and Solutions\n\n**Cause:** Agent keeps calling the same tool with same parameters\n**Solution:** Improve prompt to guide agent away from repetition, implement action history in context\n\n**Cause:** Agent waiting for tool that never completes\n**Solution:** Add timeouts to all tool calls, implement circuit breakers\n\n**Cause:** Agent confused about when to terminate\n**Solution:** Clear termination conditions, explicit TERMINATE signals\n\n## Tool Integration Best Practices\n\n### MCP Server Integration\n\nWhen integrating MCP servers with agent frameworks:\n\n**With LangChain:**\n```python\nfrom langchain.tools import Tool\n\n# Wrap MCP server function as LangChain tool\nmcp_tool = Tool(\n    name=\"mcp_server_query\",\n    func=mcp_server.query,\n    description=\"Query the MCP server for data\"\n)\n\n# Add to agent\nagent = create_openai_functions_agent(llm, tools=[mcp_tool], prompt=prompt)\n```\n\n**With CrewAI:**\n- Define MCP functions as @tool decorated functions\n- Register with specific agents that need MCP access\n\n**General principles:**\n- Always handle MCP server errors gracefully (return error strings, not exceptions)\n- Set timeouts on MCP calls (don't let agents hang)\n- Log all MCP interactions for debugging\n\n### Tool Error Handling Pattern\n\n```python\n@tool\ndef resilient_tool(query: str) -> str:\n    \"\"\"Tool with proper error handling\"\"\"\n    try:\n        result = external_api.call(query, timeout=10)\n        return json.dumps(result)\n    except TimeoutError:\n        return \"Error: Request timed out. Please try again.\"\n    except Exception as e:\n        logger.error(f\"Tool failed: {e}\")\n        return f\"Error: {str(e)}\"\n```\n\n**Never let tools crash the agent** - always return strings, even for errors.\n\n## Cost Optimization\n\n### Estimation Before Building\n\nUse `scripts\u002Fcost-estimator.py` to estimate costs before implementing:\n\n```bash\npython scripts\u002Fcost-estimator.py --config workflow.json --requests 1000 --optimize\n```\n\n### Optimization Strategies\n\n**1. Use cheaper models for simple agents:**\n- GPT-4 for complex reasoning (strategist, analyst)\n- GPT-3.5-turbo for simple tasks (formatter, router)\n\n**2. Limit iterations strictly:**\n- Set `max_iterations` to 5-10 (not 50)\n- Set `max_execution_time` to prevent runaway costs\n\n**3. Cache repeated work:**\n- Enable LLM response caching\n- Store results of expensive operations\n- Reuse research across similar queries\n\n**4. Implement budgets:**\n```python\ncost_tracker = CostTracker()\nif cost_tracker.total_cost > budget:\n    raise BudgetExceededError()\n```\n\n**5. Monitor costs in real-time:**\n- Track token usage per agent\n- Alert when costs approach thresholds\n- Analyze cost breakdowns to identify expensive agents\n\n## Production Deployment Considerations\n\nRefer to `references\u002Fproduction-patterns.md` for deployment architectures and `references\u002Fobservability-guide.md` for monitoring.\n\n### Essential Production Patterns\n\n**1. Async Task Queue (Recommended):**\n- Use Celery + Redis for long-running workflows\n- Return task ID immediately, poll for results\n- Enables horizontal scaling\n\n**2. Always Enable Observability:**\n- Structured logging (JSON format)\n- Distributed tracing (OpenTelemetry or LangSmith)\n- Cost tracking per request\n- Error monitoring (Sentry)\n\n**3. Implement Fault Tolerance:**\n- Retry with exponential backoff\n- Circuit breakers for external APIs\n- Graceful degradation (fallback to cheaper models)\n\n**4. Security:**\n- API authentication (JWT tokens)\n- Rate limiting (10-100 requests\u002Fminute per user)\n- Input validation (prevent prompt injection)\n- Sandboxed code execution (Docker for AutoGen)\n\n**5. Set Hard Limits:**\n```python\nLIMITS = {\n    \"max_iterations\": 10,\n    \"max_execution_time\": 60,  # seconds\n    \"max_cost_per_request\": 1.0,  # USD\n    \"max_concurrent_agents\": 5\n}\n```\n\n## Anti-Patterns to Avoid\n\n### ❌ Over-Engineering with Agents\n\n**Problem:** Using multiple agents when a single LLM call would suffice\n\n**Example:**\n```python\n# ❌ Overkill for simple task\nagents = [InputParser(), Processor(), OutputFormatter()]\n# Just to format text!\n```\n\n**Solution:** Start simple. One LLM call. Add agents only when truly beneficial.\n\n### ❌ No Termination Conditions\n\n**Problem:** Agents run forever or hit arbitrary iteration limits\n\n**Solution:**\n```python\n# ✅ Always set termination conditions\nmax_iterations=10\nis_termination_msg=lambda msg: \"TERMINATE\" in msg.get(\"content\", \"\")\n```\n\n### ❌ Vague Agent Roles\n\n**Problem:** Agents with unclear responsibilities cause overlap and confusion\n\n**Solution:** Specific roles with clear boundaries. \"Senior Python Developer\" not \"Helper\".\n\n### ❌ Ignoring Costs\n\n**Problem:** Running expensive models without tracking or budgets\n\n**Solution:** Track costs per request, set budgets, alert on anomalies\n\n### ❌ No Observability\n\n**Problem:** Can't debug when things go wrong\n\n**Solution:** Implement logging, tracing, and metrics from day one\n\n## When NOT to Use Agents\n\nRecognize situations where agents add unnecessary complexity:\n\n**Use a single LLM call instead of agents when:**\n- Task is straightforward (translation, summarization, Q&A)\n- No tool usage required\n- Single-step reasoning suffices\n- Cost and latency must be minimized\n\n**Example decision:**\n- \"Translate this text\" → Single LLM call\n- \"Research 10 sources and synthesize findings\" → Multi-agent system\n\n## Resource Reference Guide\n\nThis skill includes comprehensive reference materials. Load them as needed:\n\n### Framework Comparisons\n- **`references\u002Fframework-comparison.md`** - Detailed CrewAI vs LangChain vs AutoGen comparison with migration paths\n\n### Framework-Specific Patterns\n- **`references\u002Fcrewai-patterns.md`** - Role design, task patterns, processes, production tips\n- **`references\u002Flangchain-patterns.md`** - LCEL chains, RAG patterns, LangGraph workflows\n- **`references\u002Fautogen-patterns.md`** - Conversational agents, group chats, code execution\n\n### Advanced Topics\n- **`references\u002Fcustom-agent-guide.md`** - Building agents from scratch, minimal implementations\n- **`references\u002Fobservability-guide.md`** - Tracing, logging, metrics, debugging tools\n- **`references\u002Fproduction-patterns.md`** - Deployment architectures, scaling, security, monitoring\n\n### Helper Scripts\n- **`scripts\u002Fcost-estimator.py`** - Estimate workflow costs before building\n- **`scripts\u002Fagent-debugger.py`** - Analyze traces to identify loops and bottlenecks\n- **`scripts\u002Fagent-system-template\u002F`** - Boilerplate for new agent systems\n\n## Getting Started Workflow\n\nFor new agent system development:\n\n1. **Validate need for agents** - Can this be solved with a single LLM call?\n2. **Choose framework** - Use decision tree above\n3. **Estimate costs** - Run cost-estimator.py with expected workflow\n4. **Implement with observability** - Enable logging\u002Ftracing from start\n5. **Test for loops** - Use agent-debugger.py on traces\n6. **Set hard limits** - max_iterations, budgets, timeouts\n7. **Monitor in production** - Track costs, errors, performance\n\n## Framework-Specific Quick References\n\n### CrewAI Essentials\n- Agents need: role, goal, backstory, tools\n- Tasks need: description, expected_output, agent\n- Processes: Sequential (linear) or Hierarchical (manager delegates)\n- Limit: 2-5 agents per crew\n\n### LangChain Essentials\n- Always set max_iterations on AgentExecutor\n- Use LCEL for chain composition: `prompt | model | parser`\n- Enable LangSmith for production: `LANGCHAIN_TRACING_V2=true`\n- RAG: Use RetrievalQA or ConversationalRetrievalChain\n\n### AutoGen Essentials\n- UserProxyAgent executes code, AssistantAgent generates it\n- Always use Docker in production: `use_docker=True`\n- GroupChat needs max_round limit\n- Human input modes: NEVER (autonomous), TERMINATE (confirm end), ALWAYS (every step)\n\n### LangGraph Essentials\n- Define state, nodes (functions), and edges (transitions)\n- Use conditional edges for branching logic\n- Supports cycles (retries, loops)\n- Requires StateGraph setup and compilation\n\n## Summary\n\nMulti-agent systems add complexity. Use this skill to:\n\n✅ Make informed framework selections\n✅ Design effective agent architectures\n✅ Avoid common pitfalls (loops, runaway costs)\n✅ Implement production-ready observability\n✅ Recognize when agents are overkill\n\nPrioritize simplicity. Start with the minimum viable agent system. Add complexity only when justified by concrete benefits.\n",{"name":5,"description":7},"main","7850740bd837adb32f0a109ab2af82482196c6f2","skills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks",1,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills","https:\u002F\u002Fzturnsgo.com","https:\u002F\u002Fgithub.com\u002Fshengdabai","https:\u002F\u002Fzturnsgo.com\u002F","AI × education creator — producer, not consumer. Taught 6,000+ learners; shipping learning tools & daily essays with Claude Code, practicing @xiaolai's way 🥋","2026-11-01 21:28:23",{"source":8,"skillName":5,"commands":60,"prompt":70,"zipAvailable":71},[61,64,67],{"tool":62,"command":63},"npx","npx skills add shengdabai\u002FTony-Claude-Code-Skills --skill ai-agent-frameworks",{"tool":65,"command":66},"bunx","bunx skills add shengdabai\u002FTony-Claude-Code-Skills --skill ai-agent-frameworks",{"tool":68,"command":69},"pnpm","pnpm dlx skills add shengdabai\u002FTony-Claude-Code-Skills --skill ai-agent-frameworks","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002FSKILL.md",true,[73,79,83,87,91,95,99,103,109,113,117,121,125],{"path":74,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":77,"sourceHtmlUrl":78},"references\u002Fautogen-patterns.md","REFERENCE","参考文档",18134,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fautogen-patterns.md",{"path":80,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":81,"sourceHtmlUrl":82},"references\u002Fcrewai-patterns.md",15066,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fcrewai-patterns.md",{"path":84,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":85,"sourceHtmlUrl":86},"references\u002Fcustom-agent-guide.md",23325,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fcustom-agent-guide.md",{"path":88,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":89,"sourceHtmlUrl":90},"references\u002Fframework-comparison.md",10760,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fframework-comparison.md",{"path":92,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":93,"sourceHtmlUrl":94},"references\u002Flangchain-patterns.md",18224,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Flangchain-patterns.md",{"path":96,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":97,"sourceHtmlUrl":98},"references\u002Fobservability-guide.md",20140,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fobservability-guide.md",{"path":100,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":101,"sourceHtmlUrl":102},"references\u002Fproduction-patterns.md",19617,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Freferences\u002Fproduction-patterns.md",{"path":104,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":107,"sourceHtmlUrl":108},"scripts\u002Fagent-debugger.py","SCRIPT","脚本",10627,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fagent-debugger.py",{"path":110,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":111,"sourceHtmlUrl":112},"scripts\u002Fagent-system-template\u002Fconfig.yaml",460,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fagent-system-template\u002Fconfig.yaml",{"path":114,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":115,"sourceHtmlUrl":116},"scripts\u002Fagent-system-template\u002Fmain.py",1296,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fagent-system-template\u002Fmain.py",{"path":118,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":119,"sourceHtmlUrl":120},"scripts\u002Fagent-system-template\u002FREADME.md",1123,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fagent-system-template\u002FREADME.md",{"path":122,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":123,"sourceHtmlUrl":124},"scripts\u002Fagent-system-template\u002Frequirements.txt",65,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fagent-system-template\u002Frequirements.txt",{"path":126,"resourceType":105,"resourceTypeDesc":106,"sizeBytes":127,"sourceHtmlUrl":128},"scripts\u002Fcost-estimator.py",6948,"https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002F_collections\u002Fdot_claude\u002Fskills\u002Fai-agent-frameworks\u002Fscripts\u002Fcost-estimator.py",[130,139,147,155,163,171],{"skillNo":131,"name":132,"nameZh":6,"description":133,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":134,"sourceHtmlUrl":135,"categories":136,"keywords":137,"publishedAt":138},"01M73H7XKBHE1ETQDZT94MF9HH","seedance-prompt-en","Write effective prompts for Jimeng Seedance 2.0 multimodal AI video generation. Use when users want to create video prompts using text, images, videos, and audio inputs with the @ reference system. Covers camera movements, effects replication, video extension, editing, music beat-matching, e-commerce ads, short dramas, and educational content.","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fseedance-prompt-en\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fseedance-prompt-en\u002FSKILL.md",[],[],"2026-11-11 15:11:14",{"skillNo":140,"name":141,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":143,"sourceHtmlUrl":144,"categories":145,"keywords":146,"publishedAt":138},"01M73GRE9998Q1DZWC71N41JWG","onepage-pdf","Convert an HTML page into a single continuous-page PDF (one tall page, no pagination breaks), preserving desktop layout, backgrounds and selectable text. Use when the user asks to turn an HTML file\u002Freport\u002Fproposal into a \"single-page PDF\", \"long PDF\", \"不分页 PDF\", \"单页 PDF\", \"长图式 PDF\", or complains that an HTML-to-PDF export breaks into pages or cuts content. Supports optional string redaction with leak verification.","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fonepage-pdf\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fonepage-pdf\u002FSKILL.md",[],[],{"skillNo":148,"name":149,"nameZh":6,"description":150,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":151,"sourceHtmlUrl":152,"categories":153,"keywords":154,"publishedAt":138},"01M73GQG1KZDADVPV3MHAT1J5A","native-feel-cross-platform-desktop","Use when the user is designing, prototyping, or rewriting a desktop app that must run on multiple OSes (macOS + Windows, optionally Linux) AND feel indistinguishable from a native app to its users — fast launch, native windowing, native input handling, native materials. Trigger words include \"cross-platform desktop\", \"Electron alternative\", \"Tauri vs native\", \"WebView wrapper\", \"near-native performance\", \"Raycast architecture\", \"WebKit\u002FWebView2 quirks\", \"WKWebView\", \"system tray app\", \"global hotkey app\", \"launcher app\". Do NOT trigger this skill for pure web apps, pure mobile apps, or for greenfield projects that have no native-feel requirement.","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fnative-feel-cross-platform-desktop\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fnative-feel-cross-platform-desktop\u002FSKILL.md",[],[],{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":159,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":138},"01M73G87PDNSMKG2FHWCYER6N8","gstack","Router for the gstack skill suite. (gstack)","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fgstack\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fgstack\u002FSKILL.md",[],[],{"skillNo":164,"name":165,"nameZh":6,"description":166,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":167,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":138},"01M73G3EAQWKCM8DB8D0N0QSVG","fireworks-tech-graph","Use when the user wants to create any technical diagram - architecture, data flow, flowchart, sequence, agent\u002Fmemory, or concept map - and export as SVG+PNG. Trigger on: \"画图\" \"帮我画\" \"生成图\" \"做个图\" \"架构图\" \"流程图\" \"可视化一下\" \"出图\" \"generate diagram\" \"draw diagram\" \"visualize\" or any system\u002Fflow description the user wants illustrated.","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Ffireworks-tech-graph\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Ffireworks-tech-graph\u002FSKILL.md",[],[],{"skillNo":172,"name":173,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":138},"01M73G086AXAH9DW1H5G5N3CPB","dbs-video-extract","接收抖音、小红书、微信视频号等短视频链接或分享文案，查询作品／账号数据并提取语音文字稿，按作者和标题归档为 Markdown。用户希望完整解析短视频、取得数据或文字稿时使用；运行前检查 TikHub 与轻抖使用凭证，缺少时用新手能理解的方式引导购买和安全配置。","https:\u002F\u002Fgithub.com\u002Fshengdabai\u002FTony-Claude-Code-Skills\u002Fblob\u002Fmain\u002Fskills\u002Fdbs-video-extract\u002FSKILL.md",[],[]]