[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6C52C8MH87HMHPPNGHKZ90K":3,"skill-same:01M6C52C8MH87HMHPPNGHKZ90K":57},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":25,"linkOnly":26,"markdownContent":27,"markdownBody":28,"frontMatter":29,"refName":30,"commitSha":31,"skillDirectory":32,"forksCount":14,"repositoryHtmlUrl":33,"repositoryHomepageUrl":6,"ownerHtmlUrl":34,"ownerBlogUrl":35,"ownerDescription":6,"updatedAt":36,"install":37,"resources":50},"01M6C52C8MH87HMHPPNGHKZ90K","deep-agents-orchestration",null,"INVOKE THIS SKILL when using subagents, task planning, or human approval in Deep Agents. Covers SubAgentMiddleware, TodoList for planning, and HITL interrupts.","CoralShades\u002Facm-ai","CoralShades","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F220297585?v=4","01M6C5211QSJHZ8HXSDDWCR507","Coral Shades AI Agency","01M6C52161FNQDRQ0FRR8DAA9D",0,"MIT","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fdeep-agents-orchestration\u002FSKILL.md",[],[19,20,21,22,23,24],"orchestration","deep agents orchestration","planning","task planning","approval","subagents","2026-11-10 10:12:30",false,"---\nname: deep-agents-orchestration\ndescription: \"INVOKE THIS SKILL when using subagents, task planning, or human approval in Deep Agents. Covers SubAgentMiddleware, TodoList for planning, and HITL interrupts.\"\n---\n\n\u003Coverview>\nDeep Agents include three orchestration capabilities:\n\n1. **SubAgentMiddleware**: Delegate work via `task` tool to specialized agents\n2. **TodoListMiddleware**: Plan and track tasks via `write_todos` tool\n3. **HumanInTheLoopMiddleware**: Require approval before sensitive operations\n\nAll three are automatically included in `create_deep_agent()`.\n\u003C\u002Foverview>\n\n---\n\n## Subagents (Task Delegation)\n\n\u003Cwhen-to-use-subagents>\n\n| Use Subagents When | Use Main Agent When |\n|-------------------|-------------------|\n| Task needs specialized tools | General-purpose tools sufficient |\n| Want to isolate complex work | Single-step operation |\n| Need clean context for main agent | Context bloat acceptable |\n\n\u003C\u002Fwhen-to-use-subagents>\n\n\u003Chow-subagents-work>\nMain agent has `task` tool -> creates fresh subagent -> subagent executes autonomously -> returns final report.\n\n**Default subagent**: \"general-purpose\" - automatically available with same tools\u002Fconfig as main agent.\n\u003C\u002Fhow-subagents-work>\n\n\u003Cex-custom-subagents>\n\u003Cpython>\nCreate a custom \"researcher\" subagent with specialized tools for academic paper search.\n```python\nfrom deepagents import create_deep_agent\nfrom langchain.tools import tool\n\n@tool\ndef search_papers(query: str) -> str:\n    \"\"\"Search academic papers.\"\"\"\n    return f\"Found 10 papers about {query}\"\n\nagent = create_deep_agent(\n    subagents=[\n        {\n            \"name\": \"researcher\",\n            \"description\": \"Conduct web research and compile findings\",\n            \"system_prompt\": \"Search thoroughly, return concise summary\",\n            \"tools\": [search_papers],\n        }\n    ]\n)\n\n# Main agent delegates: task(agent=\"researcher\", instruction=\"Research AI trends\")\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nCreate a custom \"researcher\" subagent with specialized tools for academic paper search.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { tool } from \"@langchain\u002Fcore\u002Ftools\";\nimport { z } from \"zod\";\n\nconst searchPapers = tool(\n  async ({ query }) => `Found 10 papers about ${query}`,\n  { name: \"search_papers\", description: \"Search papers\", schema: z.object({ query: z.string() }) }\n);\n\nconst agent = await createDeepAgent({\n  subagents: [\n    {\n      name: \"researcher\",\n      description: \"Conduct web research and compile findings\",\n      systemPrompt: \"Search thoroughly, return concise summary\",\n      tools: [searchPapers],\n    }\n  ]\n});\n\n\u002F\u002F Main agent delegates: task(agent=\"researcher\", instruction=\"Research AI trends\")\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-custom-subagents>\n\n\u003Cex-subagent-with-hitl>\n\u003Cpython>\nConfigure a subagent with HITL approval for sensitive operations.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\n\nagent = create_deep_agent(\n    subagents=[\n        {\n            \"name\": \"code-deployer\",\n            \"description\": \"Deploy code to production\",\n            \"system_prompt\": \"You deploy code after tests pass.\",\n            \"tools\": [run_tests, deploy_to_prod],\n            \"interrupt_on\": {\"deploy_to_prod\": True},  # Require approval\n        }\n    ],\n    checkpointer=MemorySaver()  # Required for interrupts\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-subagent-with-hitl>\n\n\u003Cfix-subagents-are-stateless>\n\u003Cpython>\nSubagents are stateless - provide complete instructions in a single call.\n```python\n# WRONG: Subagents don't remember previous calls\n# task(agent='research', instruction='Find data')\n# task(agent='research', instruction='What did you find?')  # Starts fresh!\n\n# CORRECT: Complete instructions upfront\n# task(agent='research', instruction='Find data on AI, save to \u002Fresearch\u002F, return summary')\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nSubagents are stateless - provide complete instructions in a single call.\n```typescript\n\u002F\u002F WRONG: Subagents don't remember previous calls\n\u002F\u002F task research: Find data\n\u002F\u002F task research: What did you find?  \u002F\u002F Starts fresh!\n\n\u002F\u002F CORRECT: Complete instructions upfront\n\u002F\u002F task research: Find data on AI, save to \u002Fresearch\u002F, return summary\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-subagents-are-stateless>\n\n\u003Cfix-custom-subagents-dont-inherit-skills>\n\u003Cpython>\nCustom subagents don't inherit skills from the main agent.\n```python\n# WRONG: Custom subagent won't have main agent's skills\nagent = create_deep_agent(\n    skills=[\"\u002Fmain-skills\u002F\"],\n    subagents=[{\"name\": \"helper\", ...}]  # No skills inherited\n)\n\n# CORRECT: Provide skills explicitly (general-purpose subagent DOES inherit)\nagent = create_deep_agent(\n    skills=[\"\u002Fmain-skills\u002F\"],\n    subagents=[{\"name\": \"helper\", \"skills\": [\"\u002Fhelper-skills\u002F\"], ...}]\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-custom-subagents-dont-inherit-skills>\n\n---\n\n## TodoList (Task Planning)\n\n\u003Cwhen-to-use-todolist>\n\n| Use TodoList When | Skip TodoList When |\n|------------------|-------------------|\n| Complex multi-step tasks | Simple single-action tasks |\n| Long-running operations | Quick operations (\u003C 3 steps) |\n\n\u003C\u002Fwhen-to-use-todolist>\n\n\u003Ctodolist-tool>\n```\nwrite_todos(todos: list[dict]) -> None\n```\n\nEach todo item has:\n- `content`: Description of the task\n- `status`: One of `\"pending\"`, `\"in_progress\"`, `\"completed\"`\n\u003C\u002Ftodolist-tool>\n\n\u003Cex-todolist-usage>\n\u003Cpython>\nInvoke an agent that automatically creates a todo list for a multi-step task.\n```python\nfrom deepagents import create_deep_agent\n\nagent = create_deep_agent()  # TodoListMiddleware included by default\n\nresult = agent.invoke({\n    \"messages\": [{\"role\": \"user\", \"content\": \"Create a REST API: design models, implement CRUD, add auth, write tests\"}]\n}, config={\"configurable\": {\"thread_id\": \"session-1\"}})\n\n# Agent's planning via write_todos:\n# [\n#   {\"content\": \"Design data models\", \"status\": \"in_progress\"},\n#   {\"content\": \"Implement CRUD endpoints\", \"status\": \"pending\"},\n#   {\"content\": \"Add authentication\", \"status\": \"pending\"},\n#   {\"content\": \"Write tests\", \"status\": \"pending\"}\n# ]\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nInvoke an agent that automatically creates a todo list for a multi-step task.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\n\nconst agent = await createDeepAgent();  \u002F\u002F TodoListMiddleware included\n\nconst result = await agent.invoke({\n  messages: [{ role: \"user\", content: \"Create a REST API: design models, implement CRUD, add auth, write tests\" }]\n}, { configurable: { thread_id: \"session-1\" } });\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-todolist-usage>\n\n\u003Cex-access-todo-state>\n\u003Cpython>\nAccess the todo list from the agent's final state after invocation.\n```python\nresult = agent.invoke({...}, config={\"configurable\": {\"thread_id\": \"session-1\"}})\n\n# Access todo list from final state\ntodos = result.get(\"todos\", [])\nfor todo in todos:\n    print(f\"[{todo['status']}] {todo['content']}\")\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-access-todo-state>\n\n\u003Cfix-todolist-requires-thread-id>\n\u003Cpython>\nTodo list state requires a thread_id for persistence across invocations.\n```python\n# WRONG: Fresh state each time without thread_id\nagent.invoke({\"messages\": [...]})\n\n# CORRECT: Use thread_id\nconfig = {\"configurable\": {\"thread_id\": \"user-session\"}}\nagent.invoke({\"messages\": [...]}, config=config)  # Todos preserved\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-todolist-requires-thread-id>\n\n---\n\n## Human-in-the-Loop (Approval Workflows)\n\n\u003Cwhen-to-use-hitl>\n\n| Use HITL When | Skip HITL When |\n|--------------|---------------|\n| High-stakes operations (DB writes, deployments) | Read-only operations |\n| Compliance requires human oversight | Fully automated workflows |\n\n\u003C\u002Fwhen-to-use-hitl>\n\n\u003Cex-hitl-setup>\n\u003Cpython>\nConfigure which tools require human approval before execution.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\n\nagent = create_deep_agent(\n    interrupt_on={\n        \"write_file\": True,  # All decisions allowed\n        \"execute_sql\": {\"allowed_decisions\": [\"approve\", \"reject\"]},\n        \"read_file\": False,  # No interrupts\n    },\n    checkpointer=MemorySaver()  # REQUIRED for interrupts\n)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nConfigure which tools require human approval before execution.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { MemorySaver } from \"@langchain\u002Flanggraph\";\n\nconst agent = await createDeepAgent({\n  interruptOn: {\n    write_file: true,\n    execute_sql: { allowedDecisions: [\"approve\", \"reject\"] },\n    read_file: false,\n  },\n  checkpointer: new MemorySaver()  \u002F\u002F REQUIRED\n});\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-hitl-setup>\n\n\u003Cex-approval-workflow>\n\u003Cpython>\nComplete workflow: trigger an interrupt, check state, approve action, and resume execution.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\nfrom langgraph.types import Command\n\nagent = create_deep_agent(\n    interrupt_on={\"write_file\": True},\n    checkpointer=MemorySaver()\n)\n\nconfig = {\"configurable\": {\"thread_id\": \"session-1\"}}\n\n# Step 1: Agent proposes write_file - execution pauses\nresult = agent.invoke({\n    \"messages\": [{\"role\": \"user\", \"content\": \"Write config to \u002Fprod.yaml\"}]\n}, config=config)\n\n# Step 2: Check for interrupts\nstate = agent.get_state(config)\nif state.next:\n    print(f\"Pending action\")\n\n# Step 3: Approve and resume\nresult = agent.invoke(Command(resume={\"decisions\": [{\"type\": \"approve\"}]}), config=config)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nComplete workflow: trigger an interrupt, check state, approve action, and resume execution.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { MemorySaver, Command } from \"@langchain\u002Flanggraph\";\n\nconst agent = await createDeepAgent({\n  interruptOn: { write_file: true },\n  checkpointer: new MemorySaver()\n});\n\nconst config = { configurable: { thread_id: \"session-1\" } };\n\n\u002F\u002F Step 1: Agent proposes write_file - execution pauses\nlet result = await agent.invoke({\n  messages: [{ role: \"user\", content: \"Write config to \u002Fprod.yaml\" }]\n}, config);\n\n\u002F\u002F Step 2: Check for interrupts\nconst state = await agent.getState(config);\nif (state.next) {\n  console.log(\"Pending action\");\n}\n\n\u002F\u002F Step 3: Approve and resume\nresult = await agent.invoke(\n  new Command({ resume: { decisions: [{ type: \"approve\" }] } }), config\n);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-approval-workflow>\n\n\u003Cex-reject-with-feedback>\n\u003Cpython>\nReject a pending action with feedback, prompting the agent to try a different approach.\n```python\nresult = agent.invoke(\n    Command(resume={\"decisions\": [{\"type\": \"reject\", \"message\": \"Run tests first\"}]}),\n    config=config,\n)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nReject a pending action with feedback, prompting the agent to try a different approach.\n```typescript\nconst result = await agent.invoke(\n  new Command({ resume: { decisions: [{ type: \"reject\", message: \"Run tests first\" }] } }),\n  config,\n);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-reject-with-feedback>\n\n\u003Cex-edit-before-execution>\n\u003Cpython>\nEdit the proposed action arguments before allowing execution.\n```python\nresult = agent.invoke(\n    Command(resume={\"decisions\": [{\n        \"type\": \"edit\",\n        \"edited_action\": {\n            \"name\": \"execute_sql\",\n            \"args\": {\"query\": \"DELETE FROM users WHERE last_login \u003C '2020-01-01' LIMIT 100\"},\n        },\n    }]}),\n    config=config,\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-edit-before-execution>\n\n\u003Cboundaries>\n### What Agents CAN Configure\n\n- Subagent names, tools, models, system prompts\n- Which tools require approval\n- Allowed decision types per tool\n- TodoList content and structure\n\n### What Agents CANNOT Configure\n\n- Tool names (`task`, `write_todos`)\n- HITL protocol (approve\u002Fedit\u002Freject structure)\n- Skip checkpointer requirement for interrupts\n- Make subagents stateful (they're ephemeral)\n\u003C\u002Fboundaries>\n\n\u003Cfix-checkpointer-required>\n\u003Cpython>\nCheckpointer is required when using interrupt_on for HITL workflows.\n```python\n# WRONG\nagent = create_deep_agent(interrupt_on={\"write_file\": True})\n\n# CORRECT\nagent = create_deep_agent(interrupt_on={\"write_file\": True}, checkpointer=MemorySaver())\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nCheckpointer is required when using interruptOn for HITL workflows.\n```typescript\n\u002F\u002F WRONG\nconst agent = await createDeepAgent({ interruptOn: { write_file: true } });\n\n\u002F\u002F CORRECT\nconst agent = await createDeepAgent({ interruptOn: { write_file: true }, checkpointer: new MemorySaver() });\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-checkpointer-required>\n\n\u003Cfix-thread-id-required-for-resumption>\n\u003Cpython>\nA consistent thread_id is required to resume interrupted workflows.\n```python\n# WRONG: Can't resume without thread_id\nagent.invoke({\"messages\": [...]})\n\n# CORRECT\nconfig = {\"configurable\": {\"thread_id\": \"session-1\"}}\nagent.invoke({...}, config=config)\n# Resume with Command using same config\nagent.invoke(Command(resume={\"decisions\": [{\"type\": \"approve\"}]}), config=config)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nA consistent thread_id is required to resume interrupted workflows.\n```typescript\n\u002F\u002F WRONG: Can't resume without thread_id\nawait agent.invoke({ messages: [...] });\n\n\u002F\u002F CORRECT\nconst config = { configurable: { thread_id: \"session-1\" } };\nawait agent.invoke({ messages: [...] }, config);\n\u002F\u002F Resume with Command using same config\nawait agent.invoke(new Command({ resume: { decisions: [{ type: \"approve\" }] } }), config);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-thread-id-required-for-resumption>\n\n\u003Cfix-interrupt-checks-between-invocations>\n\u003Cpython>\nInterrupts happen BETWEEN invoke() calls, not mid-execution.\n```python\nresult = agent.invoke({...}, config=config)       # Step 1: triggers interrupt\nif \"__interrupt__\" in result:                      # Step 2: check for interrupt\n    result = agent.invoke(                         # Step 3: resume\n        Command(resume={\"decisions\": [{\"type\": \"approve\"}]}),\n        config=config,\n    )\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-interrupt-checks-between-invocations>\n","\u003Coverview>\nDeep Agents include three orchestration capabilities:\n\n1. **SubAgentMiddleware**: Delegate work via `task` tool to specialized agents\n2. **TodoListMiddleware**: Plan and track tasks via `write_todos` tool\n3. **HumanInTheLoopMiddleware**: Require approval before sensitive operations\n\nAll three are automatically included in `create_deep_agent()`.\n\u003C\u002Foverview>\n\n---\n\n## Subagents (Task Delegation)\n\n\u003Cwhen-to-use-subagents>\n\n| Use Subagents When | Use Main Agent When |\n|-------------------|-------------------|\n| Task needs specialized tools | General-purpose tools sufficient |\n| Want to isolate complex work | Single-step operation |\n| Need clean context for main agent | Context bloat acceptable |\n\n\u003C\u002Fwhen-to-use-subagents>\n\n\u003Chow-subagents-work>\nMain agent has `task` tool -> creates fresh subagent -> subagent executes autonomously -> returns final report.\n\n**Default subagent**: \"general-purpose\" - automatically available with same tools\u002Fconfig as main agent.\n\u003C\u002Fhow-subagents-work>\n\n\u003Cex-custom-subagents>\n\u003Cpython>\nCreate a custom \"researcher\" subagent with specialized tools for academic paper search.\n```python\nfrom deepagents import create_deep_agent\nfrom langchain.tools import tool\n\n@tool\ndef search_papers(query: str) -> str:\n    \"\"\"Search academic papers.\"\"\"\n    return f\"Found 10 papers about {query}\"\n\nagent = create_deep_agent(\n    subagents=[\n        {\n            \"name\": \"researcher\",\n            \"description\": \"Conduct web research and compile findings\",\n            \"system_prompt\": \"Search thoroughly, return concise summary\",\n            \"tools\": [search_papers],\n        }\n    ]\n)\n\n# Main agent delegates: task(agent=\"researcher\", instruction=\"Research AI trends\")\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nCreate a custom \"researcher\" subagent with specialized tools for academic paper search.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { tool } from \"@langchain\u002Fcore\u002Ftools\";\nimport { z } from \"zod\";\n\nconst searchPapers = tool(\n  async ({ query }) => `Found 10 papers about ${query}`,\n  { name: \"search_papers\", description: \"Search papers\", schema: z.object({ query: z.string() }) }\n);\n\nconst agent = await createDeepAgent({\n  subagents: [\n    {\n      name: \"researcher\",\n      description: \"Conduct web research and compile findings\",\n      systemPrompt: \"Search thoroughly, return concise summary\",\n      tools: [searchPapers],\n    }\n  ]\n});\n\n\u002F\u002F Main agent delegates: task(agent=\"researcher\", instruction=\"Research AI trends\")\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-custom-subagents>\n\n\u003Cex-subagent-with-hitl>\n\u003Cpython>\nConfigure a subagent with HITL approval for sensitive operations.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\n\nagent = create_deep_agent(\n    subagents=[\n        {\n            \"name\": \"code-deployer\",\n            \"description\": \"Deploy code to production\",\n            \"system_prompt\": \"You deploy code after tests pass.\",\n            \"tools\": [run_tests, deploy_to_prod],\n            \"interrupt_on\": {\"deploy_to_prod\": True},  # Require approval\n        }\n    ],\n    checkpointer=MemorySaver()  # Required for interrupts\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-subagent-with-hitl>\n\n\u003Cfix-subagents-are-stateless>\n\u003Cpython>\nSubagents are stateless - provide complete instructions in a single call.\n```python\n# WRONG: Subagents don't remember previous calls\n# task(agent='research', instruction='Find data')\n# task(agent='research', instruction='What did you find?')  # Starts fresh!\n\n# CORRECT: Complete instructions upfront\n# task(agent='research', instruction='Find data on AI, save to \u002Fresearch\u002F, return summary')\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nSubagents are stateless - provide complete instructions in a single call.\n```typescript\n\u002F\u002F WRONG: Subagents don't remember previous calls\n\u002F\u002F task research: Find data\n\u002F\u002F task research: What did you find?  \u002F\u002F Starts fresh!\n\n\u002F\u002F CORRECT: Complete instructions upfront\n\u002F\u002F task research: Find data on AI, save to \u002Fresearch\u002F, return summary\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-subagents-are-stateless>\n\n\u003Cfix-custom-subagents-dont-inherit-skills>\n\u003Cpython>\nCustom subagents don't inherit skills from the main agent.\n```python\n# WRONG: Custom subagent won't have main agent's skills\nagent = create_deep_agent(\n    skills=[\"\u002Fmain-skills\u002F\"],\n    subagents=[{\"name\": \"helper\", ...}]  # No skills inherited\n)\n\n# CORRECT: Provide skills explicitly (general-purpose subagent DOES inherit)\nagent = create_deep_agent(\n    skills=[\"\u002Fmain-skills\u002F\"],\n    subagents=[{\"name\": \"helper\", \"skills\": [\"\u002Fhelper-skills\u002F\"], ...}]\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-custom-subagents-dont-inherit-skills>\n\n---\n\n## TodoList (Task Planning)\n\n\u003Cwhen-to-use-todolist>\n\n| Use TodoList When | Skip TodoList When |\n|------------------|-------------------|\n| Complex multi-step tasks | Simple single-action tasks |\n| Long-running operations | Quick operations (\u003C 3 steps) |\n\n\u003C\u002Fwhen-to-use-todolist>\n\n\u003Ctodolist-tool>\n```\nwrite_todos(todos: list[dict]) -> None\n```\n\nEach todo item has:\n- `content`: Description of the task\n- `status`: One of `\"pending\"`, `\"in_progress\"`, `\"completed\"`\n\u003C\u002Ftodolist-tool>\n\n\u003Cex-todolist-usage>\n\u003Cpython>\nInvoke an agent that automatically creates a todo list for a multi-step task.\n```python\nfrom deepagents import create_deep_agent\n\nagent = create_deep_agent()  # TodoListMiddleware included by default\n\nresult = agent.invoke({\n    \"messages\": [{\"role\": \"user\", \"content\": \"Create a REST API: design models, implement CRUD, add auth, write tests\"}]\n}, config={\"configurable\": {\"thread_id\": \"session-1\"}})\n\n# Agent's planning via write_todos:\n# [\n#   {\"content\": \"Design data models\", \"status\": \"in_progress\"},\n#   {\"content\": \"Implement CRUD endpoints\", \"status\": \"pending\"},\n#   {\"content\": \"Add authentication\", \"status\": \"pending\"},\n#   {\"content\": \"Write tests\", \"status\": \"pending\"}\n# ]\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nInvoke an agent that automatically creates a todo list for a multi-step task.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\n\nconst agent = await createDeepAgent();  \u002F\u002F TodoListMiddleware included\n\nconst result = await agent.invoke({\n  messages: [{ role: \"user\", content: \"Create a REST API: design models, implement CRUD, add auth, write tests\" }]\n}, { configurable: { thread_id: \"session-1\" } });\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-todolist-usage>\n\n\u003Cex-access-todo-state>\n\u003Cpython>\nAccess the todo list from the agent's final state after invocation.\n```python\nresult = agent.invoke({...}, config={\"configurable\": {\"thread_id\": \"session-1\"}})\n\n# Access todo list from final state\ntodos = result.get(\"todos\", [])\nfor todo in todos:\n    print(f\"[{todo['status']}] {todo['content']}\")\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-access-todo-state>\n\n\u003Cfix-todolist-requires-thread-id>\n\u003Cpython>\nTodo list state requires a thread_id for persistence across invocations.\n```python\n# WRONG: Fresh state each time without thread_id\nagent.invoke({\"messages\": [...]})\n\n# CORRECT: Use thread_id\nconfig = {\"configurable\": {\"thread_id\": \"user-session\"}}\nagent.invoke({\"messages\": [...]}, config=config)  # Todos preserved\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-todolist-requires-thread-id>\n\n---\n\n## Human-in-the-Loop (Approval Workflows)\n\n\u003Cwhen-to-use-hitl>\n\n| Use HITL When | Skip HITL When |\n|--------------|---------------|\n| High-stakes operations (DB writes, deployments) | Read-only operations |\n| Compliance requires human oversight | Fully automated workflows |\n\n\u003C\u002Fwhen-to-use-hitl>\n\n\u003Cex-hitl-setup>\n\u003Cpython>\nConfigure which tools require human approval before execution.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\n\nagent = create_deep_agent(\n    interrupt_on={\n        \"write_file\": True,  # All decisions allowed\n        \"execute_sql\": {\"allowed_decisions\": [\"approve\", \"reject\"]},\n        \"read_file\": False,  # No interrupts\n    },\n    checkpointer=MemorySaver()  # REQUIRED for interrupts\n)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nConfigure which tools require human approval before execution.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { MemorySaver } from \"@langchain\u002Flanggraph\";\n\nconst agent = await createDeepAgent({\n  interruptOn: {\n    write_file: true,\n    execute_sql: { allowedDecisions: [\"approve\", \"reject\"] },\n    read_file: false,\n  },\n  checkpointer: new MemorySaver()  \u002F\u002F REQUIRED\n});\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-hitl-setup>\n\n\u003Cex-approval-workflow>\n\u003Cpython>\nComplete workflow: trigger an interrupt, check state, approve action, and resume execution.\n```python\nfrom deepagents import create_deep_agent\nfrom langgraph.checkpoint.memory import MemorySaver\nfrom langgraph.types import Command\n\nagent = create_deep_agent(\n    interrupt_on={\"write_file\": True},\n    checkpointer=MemorySaver()\n)\n\nconfig = {\"configurable\": {\"thread_id\": \"session-1\"}}\n\n# Step 1: Agent proposes write_file - execution pauses\nresult = agent.invoke({\n    \"messages\": [{\"role\": \"user\", \"content\": \"Write config to \u002Fprod.yaml\"}]\n}, config=config)\n\n# Step 2: Check for interrupts\nstate = agent.get_state(config)\nif state.next:\n    print(f\"Pending action\")\n\n# Step 3: Approve and resume\nresult = agent.invoke(Command(resume={\"decisions\": [{\"type\": \"approve\"}]}), config=config)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nComplete workflow: trigger an interrupt, check state, approve action, and resume execution.\n```typescript\nimport { createDeepAgent } from \"deepagents\";\nimport { MemorySaver, Command } from \"@langchain\u002Flanggraph\";\n\nconst agent = await createDeepAgent({\n  interruptOn: { write_file: true },\n  checkpointer: new MemorySaver()\n});\n\nconst config = { configurable: { thread_id: \"session-1\" } };\n\n\u002F\u002F Step 1: Agent proposes write_file - execution pauses\nlet result = await agent.invoke({\n  messages: [{ role: \"user\", content: \"Write config to \u002Fprod.yaml\" }]\n}, config);\n\n\u002F\u002F Step 2: Check for interrupts\nconst state = await agent.getState(config);\nif (state.next) {\n  console.log(\"Pending action\");\n}\n\n\u002F\u002F Step 3: Approve and resume\nresult = await agent.invoke(\n  new Command({ resume: { decisions: [{ type: \"approve\" }] } }), config\n);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-approval-workflow>\n\n\u003Cex-reject-with-feedback>\n\u003Cpython>\nReject a pending action with feedback, prompting the agent to try a different approach.\n```python\nresult = agent.invoke(\n    Command(resume={\"decisions\": [{\"type\": \"reject\", \"message\": \"Run tests first\"}]}),\n    config=config,\n)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nReject a pending action with feedback, prompting the agent to try a different approach.\n```typescript\nconst result = await agent.invoke(\n  new Command({ resume: { decisions: [{ type: \"reject\", message: \"Run tests first\" }] } }),\n  config,\n);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Fex-reject-with-feedback>\n\n\u003Cex-edit-before-execution>\n\u003Cpython>\nEdit the proposed action arguments before allowing execution.\n```python\nresult = agent.invoke(\n    Command(resume={\"decisions\": [{\n        \"type\": \"edit\",\n        \"edited_action\": {\n            \"name\": \"execute_sql\",\n            \"args\": {\"query\": \"DELETE FROM users WHERE last_login \u003C '2020-01-01' LIMIT 100\"},\n        },\n    }]}),\n    config=config,\n)\n```\n\u003C\u002Fpython>\n\u003C\u002Fex-edit-before-execution>\n\n\u003Cboundaries>\n### What Agents CAN Configure\n\n- Subagent names, tools, models, system prompts\n- Which tools require approval\n- Allowed decision types per tool\n- TodoList content and structure\n\n### What Agents CANNOT Configure\n\n- Tool names (`task`, `write_todos`)\n- HITL protocol (approve\u002Fedit\u002Freject structure)\n- Skip checkpointer requirement for interrupts\n- Make subagents stateful (they're ephemeral)\n\u003C\u002Fboundaries>\n\n\u003Cfix-checkpointer-required>\n\u003Cpython>\nCheckpointer is required when using interrupt_on for HITL workflows.\n```python\n# WRONG\nagent = create_deep_agent(interrupt_on={\"write_file\": True})\n\n# CORRECT\nagent = create_deep_agent(interrupt_on={\"write_file\": True}, checkpointer=MemorySaver())\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nCheckpointer is required when using interruptOn for HITL workflows.\n```typescript\n\u002F\u002F WRONG\nconst agent = await createDeepAgent({ interruptOn: { write_file: true } });\n\n\u002F\u002F CORRECT\nconst agent = await createDeepAgent({ interruptOn: { write_file: true }, checkpointer: new MemorySaver() });\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-checkpointer-required>\n\n\u003Cfix-thread-id-required-for-resumption>\n\u003Cpython>\nA consistent thread_id is required to resume interrupted workflows.\n```python\n# WRONG: Can't resume without thread_id\nagent.invoke({\"messages\": [...]})\n\n# CORRECT\nconfig = {\"configurable\": {\"thread_id\": \"session-1\"}}\nagent.invoke({...}, config=config)\n# Resume with Command using same config\nagent.invoke(Command(resume={\"decisions\": [{\"type\": \"approve\"}]}), config=config)\n```\n\u003C\u002Fpython>\n\u003Ctypescript>\nA consistent thread_id is required to resume interrupted workflows.\n```typescript\n\u002F\u002F WRONG: Can't resume without thread_id\nawait agent.invoke({ messages: [...] });\n\n\u002F\u002F CORRECT\nconst config = { configurable: { thread_id: \"session-1\" } };\nawait agent.invoke({ messages: [...] }, config);\n\u002F\u002F Resume with Command using same config\nawait agent.invoke(new Command({ resume: { decisions: [{ type: \"approve\" }] } }), config);\n```\n\u003C\u002Ftypescript>\n\u003C\u002Ffix-thread-id-required-for-resumption>\n\n\u003Cfix-interrupt-checks-between-invocations>\n\u003Cpython>\nInterrupts happen BETWEEN invoke() calls, not mid-execution.\n```python\nresult = agent.invoke({...}, config=config)       # Step 1: triggers interrupt\nif \"__interrupt__\" in result:                      # Step 2: check for interrupt\n    result = agent.invoke(                         # Step 3: resume\n        Command(resume={\"decisions\": [{\"type\": \"approve\"}]}),\n        config=config,\n    )\n```\n\u003C\u002Fpython>\n\u003C\u002Ffix-interrupt-checks-between-invocations>\n",{"name":5,"description":7},"main","1ca81b0a7de199942400f965edd4d7544077ac5d",".agent\u002Fskills\u002Fdeep-agents-orchestration","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai","https:\u002F\u002Fgithub.com\u002FCoralShades","coralshades.com.au","2026-04-18 00:45:07",{"source":8,"skillName":5,"commands":38,"prompt":48,"zipAvailable":49},[39,42,45],{"tool":40,"command":41},"npx","npx skills add CoralShades\u002Facm-ai --skill deep-agents-orchestration",{"tool":43,"command":44},"bunx","bunx skills add CoralShades\u002Facm-ai --skill deep-agents-orchestration",{"tool":46,"command":47},"pnpm","pnpm dlx skills add CoralShades\u002Facm-ai --skill deep-agents-orchestration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fdeep-agents-orchestration\u002FSKILL.md",true,[51],{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"agents\u002Fopenai.yaml","REFERENCE","参考文档",254,"https:\u002F\u002Fgithub.com\u002FTimothyMeadows\u002FHephaera\u002Fblob\u002Fmain\u002Fskills\u002Fbabylonjs-build-ui\u002Fagents\u002Fopenai.yaml",[58,82,95,111,132,142],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":69,"publishedAt":25},"01M6C63HKXA890HHV47WBAX0TB","uncodixfy","Prevents generic AI\u002FCodex UI patterns when generating frontend code. Use this skill whenever generating HTML, CSS, React, Vue, Svelte, or any frontend UI code to enforce clean, human-designed aesthetics inspired by Linear, Raycast, Stripe, and GitHub instead of typical AI-generated UI.","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Funcodixfy\u002FSKILL.md",[64],{"categoryNo":65,"categoryName":66,"taxonomyType":67,"taxonomyTypeDesc":68},"01JBSEEDCAT000000000000009","前端开发","SCENARIO","使用场景",[60,70,71,72,73,74,75,76,77,78,79,80,81],"github","frontend","css","codex","react","code","linear","ui","vue","generated","use","html",{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":86,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":25},"01M6C63FJ2GEBBGZ5XSMV40KXB","runpodctl","Runpod CLI to manage your GPU workloads.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Frunpodctl\u002FSKILL.md",[],[84,90,91,92,93,94],"macos","cli","linux","gpu","Bash",{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":25},"01M6C63DPAV46TFVQ01SFYPJ7X","langgraph-persistence","INVOKE THIS SKILL when your LangGraph needs to persist state, remember conversations, travel through history, or configure subgraph checkpointer scoping. Covers checkpointers, thread_id, time travel, Store, and subgraph persistence modes.","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Flanggraph-persistence\u002FSKILL.md",[],[102,103,104,105,106,107,108,109,110],"langgraph persistence","langgraph","persistence","travel","history","scoping","remember","state","configure",{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":122,"publishedAt":25},"01M6C63BNS9CQEBTB8RNSSBFPA","langgraph-human-in-the-loop","INVOKE THIS SKILL when implementing human-in-the-loop patterns, pausing for approval, or handling errors in LangGraph. Covers interrupt(), Command(resume=...), approval\u002Fvalidation workflows, and the 4-tier error handling strategy.","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Flanggraph-human-in-the-loop\u002FSKILL.md",[117],{"categoryNo":118,"categoryName":119,"taxonomyType":120,"taxonomyTypeDesc":121},"01JBSEEDCAT000000000000038","人力资源","PROFESSION","适用职业",[123,124,103,125,126,127,128,129,130,23,131],"langgraph human in the loop","loop","workflows","error handling","validation","human-in-the-loop","command","strategy","resume",{"skillNo":133,"name":134,"nameZh":6,"description":135,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":136,"categories":137,"keywords":138,"publishedAt":25},"01M6C639GVYZFX3NFPV5FBDR4F","langgraph-fundamentals","INVOKE THIS SKILL when writing ANY LangGraph code. Covers StateGraph, state schemas, nodes, edges, Command, Send, invoke, streaming, and error handling.","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Flanggraph-fundamentals\u002FSKILL.md",[],[139,103,126,109,140,141,129,75],"langgraph fundamentals","streaming","writing",{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":146,"categories":147,"keywords":148,"publishedAt":25},"01M6C6375N34YX2GZ3SD3R7XR2","langchain-rag","INVOKE THIS SKILL when building ANY retrieval-augmented generation (RAG) system. Covers document loaders, RecursiveCharacterTextSplitter, embeddings (OpenAI), and vector stores (Chroma, FAISS, Pinecone).","https:\u002F\u002Fgithub.com\u002FCoralShades\u002Facm-ai\u002Fblob\u002Fmain\u002F.qwen\u002Fskills\u002Flangchain-rag\u002FSKILL.md",[],[149,150,151,152,153,154,155,156,157,158],"rag","langchain rag","langchain","embeddings","vector","openai","chroma","faiss","vector stores","system"]