[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M76SHGAA823AFEMDRGRAPPMY":3,"skill-same:01M76SHGAA823AFEMDRGRAPPMY":75},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M76SHGAA823AFEMDRGRAPPMY","agents-managers",null,"Expert guidance for creating, building, and using Claude Code agents and the Task tool. Use when working with agents, setting up agent configurations, understanding how agents work, or using the Task tool to launch specialized agents.","zuzu59\u002Fz-skills","zuzu59","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F15345233?v=4","01M756RFXRYFCM8DCAHTP0F5YW","Christian Zufferey","01M756RFYZRYH7F070CKQRMW38",0,"MIT","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: agents-managers\ndescription: Expert guidance for creating, building, and using Claude Code agents and the Task tool. Use when working with agents, setting up agent configurations, understanding how agents work, or using the Task tool to launch specialized agents.\n---\n\n\u003Cobjective>\nAgents are specialized Claude instances that run in isolated contexts with focused roles and limited tool access. This skill teaches you how to create effective agents, write strong system prompts, configure tool access, and orchestrate multi-agent workflows using the Task tool.\n\nAgents enable delegation of complex tasks to specialized agents that operate autonomously without user interaction, returning their final output to the main conversation.\n\u003C\u002Fobjective>\n\n\u003Cquick_start>\n\u003Cworkflow>\n\n1. Run `\u002Fagents` command\n2. Select \"Create New Agent\"\n3. Choose project-level (`.claude\u002Fagents\u002F`) or user-level (`~\u002F.claude\u002Fagents\u002F`)\n4. Define the agent:\n   - **name**: lowercase-with-hyphens\n   - **description**: When should this agent be used?\n   - **tools**: Optional comma-separated list (inherits all if omitted)\n   - **model**: Optional (`sonnet`, `opus`, `haiku`, or `inherit`)\n5. Write the system prompt (the agent's instructions)\n   \u003C\u002Fworkflow>\n\n\u003Cexample>\n```markdown\n---\nname: code-reviewer\ndescription: Expert code reviewer. Use proactively after code changes to review for quality, security, and best practices.\ntools: Read, Grep, Glob, Bash\nmodel: sonnet\n---\n\n\u003Crole>\nYou are a senior code reviewer focused on quality, security, and best practices.\n\u003C\u002Frole>\n\n\u003Cfocus_areas>\n\n- Code quality and maintainability\n- Security vulnerabilities\n- Performance issues\n- Best practices adherence\n  \u003C\u002Ffocus_areas>\n\n\u003Coutput_format>\nProvide specific, actionable feedback with file:line references.\n\u003C\u002Foutput_format>\n\n```\n\u003C\u002Fexample>\n\u003C\u002Fquick_start>\n\n\u003Cfile_structure>\n| Type | Location | Scope | Priority |\n|------|----------|-------|----------|\n| **Project** | `.claude\u002Fagents\u002F` | Current project only | Highest |\n| **User** | `~\u002F.claude\u002Fagents\u002F` | All projects | Lower |\n| **Plugin** | Plugin's `agents\u002F` dir | All projects | Lowest |\n\nProject-level agents override user-level when names conflict.\n\u003C\u002Ffile_structure>\n\n\u003Cconfiguration>\n\u003Cfield name=\"name\">\n- Lowercase letters and hyphens only\n- Must be unique\n\u003C\u002Ffield>\n\n\u003Cfield name=\"description\">\n- Natural language description of purpose\n- Include when Claude should invoke this agent\n- Used for automatic agent selection\n\u003C\u002Ffield>\n\n\u003Cfield name=\"tools\">\n- Comma-separated list: `Read, Write, Edit, Bash, Grep`\n- If omitted: inherits all tools from main thread\n- Use `\u002Fagents` interface to see all available tools\n\u003C\u002Ffield>\n\n\u003Cfield name=\"model\">\n- `sonnet`, `opus`, `haiku`, or `inherit`\n- `inherit`: uses same model as main conversation\n- If omitted: defaults to configured agent model (usually sonnet)\n\u003C\u002Ffield>\n\u003C\u002Fconfiguration>\n\n\u003Cexecution_model>\n\u003Ccritical_constraint>\n**Agents are black boxes that cannot interact with users.**\n\nAgents run in isolated contexts and return their final output to the main conversation. They:\n- ✅ Can use tools like Read, Write, Edit, Bash, Grep, Glob\n- ✅ Can access MCP servers and other non-interactive tools\n- ❌ **Cannot use AskUserQuestion** or any tool requiring user interaction\n- ❌ **Cannot present options or wait for user input**\n- ❌ **User never sees agent's intermediate steps**\n\nThe main conversation sees only the agent's final report\u002Foutput.\n\u003C\u002Fcritical_constraint>\n\n\u003Cworkflow_design>\n**Designing workflows with agents:**\n\nUse **main chat** for:\n- Gathering requirements from user (AskUserQuestion)\n- Presenting options or decisions to user\n- Any task requiring user confirmation\u002Finput\n- Work where user needs visibility into progress\n\nUse **agents** for:\n- Research tasks (API documentation lookup, code analysis)\n- Code generation based on pre-defined requirements\n- Analysis and reporting (security review, test coverage)\n- Context-heavy operations that don't need user interaction\n\n**Example workflow pattern:**\n```\n\nMain Chat: Ask user for requirements (AskUserQuestion)\n↓\nAgent: Research API and create documentation (no user interaction)\n↓\nMain Chat: Review research with user, confirm approach\n↓\nAgent: Generate code based on confirmed plan\n↓\nMain Chat: Present results, handle testing\u002Fdeployment\n\n````\n\u003C\u002Fworkflow_design>\n\u003C\u002Fexecution_model>\n\n\u003Csystem_prompt_guidelines>\n\u003Cprinciple name=\"be_specific\">\nClearly define the agent's role, capabilities, and constraints.\n\u003C\u002Fprinciple>\n\n\u003Cprinciple name=\"use_pure_xml_structure\">\nStructure the system prompt with pure XML tags. Remove ALL markdown headings from the body.\n\n```markdown\n---\nname: security-reviewer\ndescription: Reviews code for security vulnerabilities\ntools: Read, Grep, Glob, Bash\nmodel: sonnet\n---\n\n\u003Crole>\nYou are a senior code reviewer specializing in security.\n\u003C\u002Frole>\n\n\u003Cfocus_areas>\n- SQL injection vulnerabilities\n- XSS attack vectors\n- Authentication\u002Fauthorization issues\n- Sensitive data exposure\n\u003C\u002Ffocus_areas>\n\n\u003Cworkflow>\n1. Read the modified files\n2. Identify security risks\n3. Provide specific remediation steps\n4. Rate severity (Critical\u002FHigh\u002FMedium\u002FLow)\n\u003C\u002Fworkflow>\n````\n\n\u003C\u002Fprinciple>\n\n\u003Cprinciple name=\"task_specific\">\nTailor instructions to the specific task domain. Don't create generic \"helper\" agents.\n\n❌ Bad: \"You are a helpful assistant that helps with code\"\n\u0005✅ Good: \"You are a React component refactoring specialist. Analyze components for hooks best practices, performance anti-patterns, and accessibility issues.\"\n\u003C\u002Fprinciple>\n\u003C\u002Fsystem_prompt_guidelines>\n\n\u003Cagent_xml_structure>\nAgent.md files are system prompts consumed only by Claude. Like skills, they should use pure XML structure for optimal parsing and token efficiency.\n\n\u003Crecommended_tags>\nCommon tags for agent structure:\n\n- `\u003Crole>` - Who the agent is and what it does\n- `\u003Cconstraints>` - Hard rules (NEVER\u002FMUST\u002FALWAYS)\n- `\u003Cfocus_areas>` - What to prioritize\n- `\u003Cworkflow>` - Step-by-step process\n- `\u003Coutput_format>` - How to structure deliverables\n- `\u003Csuccess_criteria>` - Completion criteria\n- `\u003Cvalidation>` - How to verify work\n  \u003C\u002Frecommended_tags>\n\n\u003Cintelligence_rules>\n**Simple agents** (single focused task):\n\n- Use role + constraints + workflow minimum\n- Example: code-reviewer, test-runner\n\n**Medium agents** (multi-step process):\n\n- Add workflow steps, output_format, success_criteria\n- Example: api-researcher, documentation-generator\n\n**Complex agents** (research + generation + validation):\n\n- Add all tags as appropriate including validation, examples\n- Example: mcp-api-researcher, comprehensive-auditor\n  \u003C\u002Fintelligence_rules>\n\n\u003Ccritical_rule>\n**Remove ALL markdown headings (##, ###) from agent body.** Use semantic XML tags instead.\n\nKeep markdown formatting WITHIN content (bold, italic, lists, code blocks, links).\n\nFor XML structure principles and token efficiency details, see `..\u002Fskill-creator\u002Freferences\u002Fxml-tag-guide.md` - the same principles apply to agents.\n\u003C\u002Fcritical_rule>\n\u003C\u002Fagent_xml_structure>\n\n\u003Cinvocation>\n\u003Cautomatic>\nClaude automatically selects agents based on the `description` field when it matches the current task.\n\u003C\u002Fautomatic>\n\n\u003Cexplicit>\nYou can explicitly invoke a agent:\n\n```\n> Use the code-reviewer agent to check my recent changes\n```\n\n```\n> Have the test-writer agent create tests for the new API endpoints\n```\n\n\u003C\u002Fexplicit>\n\u003C\u002Finvocation>\n\n\u003Cbackground_execution>\nAgents can run in the background using the `run_in_background` parameter, allowing parallel execution while the main conversation continues.\n\n\u003Chow_it_works>\n**Starting a background agent:**\nThe Task tool accepts `run_in_background: true` to launch agents asynchronously:\n\n```\nTask tool call:\n- description: \"Analyze security vulnerabilities\"\n- prompt: \"Review all authentication code for security issues...\"\n- subagent_type: \"security-reviewer\"\n- run_in_background: true\n```\n\nThe agent starts immediately and returns an `agent_id` for tracking.\n\u003C\u002Fhow_it_works>\n\n\u003Cretrieving_results>\n**Getting results with TaskOutput:**\nUse the `TaskOutput` tool to retrieve results from background agents:\n\n```\nTaskOutput tool call:\n- task_id: \"agent-12345\"  # The agent_id from the Task call\n- block: true            # Wait for completion (default)\n- timeout: 30000         # Max wait time in ms\n```\n\n**Parameters:**\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `task_id` | Required | The agent ID returned from Task tool |\n| `block` | `true` | Wait for completion or check current status |\n| `timeout` | `30000` | Max wait time in milliseconds (up to 600000) |\n\n**Non-blocking check:**\nSet `block: false` to check status without waiting:\n\n```\nTaskOutput tool call:\n- task_id: \"agent-12345\"\n- block: false\n```\n\nReturns current status: running, completed, or the final result.\n\u003C\u002Fretrieving_results>\n\n\u003Cparallel_agents>\n**Launching multiple agents in parallel:**\n\nTo maximize performance, launch multiple independent agents simultaneously:\n\n```\nSingle message with multiple Task tool calls:\n\nTask 1:\n- description: \"Review code quality\"\n- prompt: \"Check code quality...\"\n- subagent_type: \"code-reviewer\"\n- run_in_background: true\n\nTask 2:\n- description: \"Run security scan\"\n- prompt: \"Scan for vulnerabilities...\"\n- subagent_type: \"security-scanner\"\n- run_in_background: true\n\nTask 3:\n- description: \"Check test coverage\"\n- prompt: \"Analyze test coverage...\"\n- subagent_type: \"test-analyzer\"\n- run_in_background: true\n```\n\nThen retrieve all results:\n\n```\nTaskOutput calls for each agent_id\n```\n\n\u003C\u002Fparallel_agents>\n\n\u003Cwhen_to_use_background>\n**Use background agents for:**\n\n- Long-running analysis (security review, comprehensive code analysis)\n- Multiple independent tasks that can run in parallel\n- Tasks where you want to continue working while waiting\n- Research tasks that may take significant time\n\n**Don't use background for:**\n\n- Quick operations (\u003C 10 seconds)\n- Tasks that depend on each other sequentially\n- Tasks where immediate results are needed for next step\n- Simple single-file operations\n\n**Pattern: Parallel Analysis Pipeline**\n\n```\n1. Launch multiple analysis agents in background\n2. Continue with other work or wait\n3. Collect all results\n4. Synthesize findings in main conversation\n```\n\n\u003C\u002Fwhen_to_use_background>\n\n\u003Cresuming_agents>\n**Resuming agents:**\nAgents can be resumed using the `resume` parameter with their agent ID:\n\n```\nTask tool call:\n- description: \"Continue security review\"\n- prompt: \"Please continue with the remaining files...\"\n- subagent_type: \"security-reviewer\"\n- resume: \"agent-12345\"  # Previous agent ID\n```\n\nThe agent continues with its full previous context preserved.\n\u003C\u002Fresuming_agents>\n\u003C\u002Fbackground_execution>\n\n\u003Cmanagement>\n\u003Cusing_agents_command>\nRun `\u002Fagents` for an interactive interface to:\n- View all available agents\n- Create new agents\n- Edit existing agents\n- Delete custom agents\n\u003C\u002Fusing_agents_command>\n\n\u003Cmanual_editing>\nYou can also edit agent files directly:\n\n- Project: `.claude\u002Fagents\u002Fagent-name.md`\n- User: `~\u002F.claude\u002Fagents\u002Fagent-name.md`\n  \u003C\u002Fmanual_editing>\n  \u003C\u002Fmanagement>\n\n\u003Creference>\n**Core references**:\n\n**Agent usage and configuration**: [references\u002Fagents.md](references\u002Fagents.md)\n\n- File format and configuration\n- Model selection (Sonnet 4.5 + Haiku 4.5 orchestration)\n- Tool security and least privilege\n- Prompt caching optimization\n- **Background execution** (run_in_background, TaskOutput, parallel agents)\n- Complete examples\n\n**Writing effective prompts**: [references\u002Fwriting-agent-prompts.md](references\u002Fwriting-agent-prompts.md)\n\n- Core principles and XML structure\n- Description field optimization for routing\n- Extended thinking for complex reasoning\n- Security constraints and strong modal verbs\n- Success criteria definition\n\n**Advanced topics**:\n\n**Evaluation and testing**: [references\u002Fevaluation-and-testing.md](references\u002Fevaluation-and-testing.md)\n\n- Evaluation metrics (task completion, tool correctness, robustness)\n- Testing strategies (offline, simulation, online monitoring)\n- Evaluation-driven development\n- G-Eval for custom criteria\n\n**Error handling and recovery**: [references\u002Ferror-handling-and-recovery.md](references\u002Ferror-handling-and-recovery.md)\n\n- Common failure modes and causes\n- Recovery strategies (graceful degradation, retry, circuit breakers)\n- Structured communication and observability\n- Anti-patterns to avoid\n\n**Context management**: [references\u002Fcontext-management.md](references\u002Fcontext-management.md)\n\n- Memory architecture (STM, LTM, working memory)\n- Context strategies (summarization, sliding window, scratchpads)\n- Managing long-running tasks\n- Prompt caching interaction\n\n**Orchestration patterns**: [references\u002Forchestration-patterns.md](references\u002Forchestration-patterns.md)\n\n- Sequential, parallel, hierarchical, coordinator patterns\n- Sonnet + Haiku orchestration for cost\u002Fperformance\n- Multi-agent coordination\n- Pattern selection guidance\n\n**Debugging and troubleshooting**: [references\u002Fdebugging-agents.md](references\u002Fdebugging-agents.md)\n\n- Logging, tracing, and correlation IDs\n- Common failure types (hallucinations, format errors, tool misuse)\n- Diagnostic procedures\n- Continuous monitoring\n  \u003C\u002Freference>\n\n\u003Csuccess_criteria>\nA well-configured agent has:\n\n- Valid YAML frontmatter (name matches file, description includes triggers)\n- Clear role definition in system prompt\n- Appropriate tool restrictions (least privilege)\n- XML-structured system prompt with role, approach, and constraints\n- Description field optimized for automatic routing\n- Successfully tested on representative tasks\n- Model selection appropriate for task complexity (Sonnet for reasoning, Haiku for simple tasks)\n  \u003C\u002Fsuccess_criteria>\n","\u003Cobjective>\nAgents are specialized Claude instances that run in isolated contexts with focused roles and limited tool access. This skill teaches you how to create effective agents, write strong system prompts, configure tool access, and orchestrate multi-agent workflows using the Task tool.\n\nAgents enable delegation of complex tasks to specialized agents that operate autonomously without user interaction, returning their final output to the main conversation.\n\u003C\u002Fobjective>\n\n\u003Cquick_start>\n\u003Cworkflow>\n\n1. Run `\u002Fagents` command\n2. Select \"Create New Agent\"\n3. Choose project-level (`.claude\u002Fagents\u002F`) or user-level (`~\u002F.claude\u002Fagents\u002F`)\n4. Define the agent:\n   - **name**: lowercase-with-hyphens\n   - **description**: When should this agent be used?\n   - **tools**: Optional comma-separated list (inherits all if omitted)\n   - **model**: Optional (`sonnet`, `opus`, `haiku`, or `inherit`)\n5. Write the system prompt (the agent's instructions)\n   \u003C\u002Fworkflow>\n\n\u003Cexample>\n```markdown\n---\nname: code-reviewer\ndescription: Expert code reviewer. Use proactively after code changes to review for quality, security, and best practices.\ntools: Read, Grep, Glob, Bash\nmodel: sonnet\n---\n\n\u003Crole>\nYou are a senior code reviewer focused on quality, security, and best practices.\n\u003C\u002Frole>\n\n\u003Cfocus_areas>\n\n- Code quality and maintainability\n- Security vulnerabilities\n- Performance issues\n- Best practices adherence\n  \u003C\u002Ffocus_areas>\n\n\u003Coutput_format>\nProvide specific, actionable feedback with file:line references.\n\u003C\u002Foutput_format>\n\n```\n\u003C\u002Fexample>\n\u003C\u002Fquick_start>\n\n\u003Cfile_structure>\n| Type | Location | Scope | Priority |\n|------|----------|-------|----------|\n| **Project** | `.claude\u002Fagents\u002F` | Current project only | Highest |\n| **User** | `~\u002F.claude\u002Fagents\u002F` | All projects | Lower |\n| **Plugin** | Plugin's `agents\u002F` dir | All projects | Lowest |\n\nProject-level agents override user-level when names conflict.\n\u003C\u002Ffile_structure>\n\n\u003Cconfiguration>\n\u003Cfield name=\"name\">\n- Lowercase letters and hyphens only\n- Must be unique\n\u003C\u002Ffield>\n\n\u003Cfield name=\"description\">\n- Natural language description of purpose\n- Include when Claude should invoke this agent\n- Used for automatic agent selection\n\u003C\u002Ffield>\n\n\u003Cfield name=\"tools\">\n- Comma-separated list: `Read, Write, Edit, Bash, Grep`\n- If omitted: inherits all tools from main thread\n- Use `\u002Fagents` interface to see all available tools\n\u003C\u002Ffield>\n\n\u003Cfield name=\"model\">\n- `sonnet`, `opus`, `haiku`, or `inherit`\n- `inherit`: uses same model as main conversation\n- If omitted: defaults to configured agent model (usually sonnet)\n\u003C\u002Ffield>\n\u003C\u002Fconfiguration>\n\n\u003Cexecution_model>\n\u003Ccritical_constraint>\n**Agents are black boxes that cannot interact with users.**\n\nAgents run in isolated contexts and return their final output to the main conversation. They:\n- ✅ Can use tools like Read, Write, Edit, Bash, Grep, Glob\n- ✅ Can access MCP servers and other non-interactive tools\n- ❌ **Cannot use AskUserQuestion** or any tool requiring user interaction\n- ❌ **Cannot present options or wait for user input**\n- ❌ **User never sees agent's intermediate steps**\n\nThe main conversation sees only the agent's final report\u002Foutput.\n\u003C\u002Fcritical_constraint>\n\n\u003Cworkflow_design>\n**Designing workflows with agents:**\n\nUse **main chat** for:\n- Gathering requirements from user (AskUserQuestion)\n- Presenting options or decisions to user\n- Any task requiring user confirmation\u002Finput\n- Work where user needs visibility into progress\n\nUse **agents** for:\n- Research tasks (API documentation lookup, code analysis)\n- Code generation based on pre-defined requirements\n- Analysis and reporting (security review, test coverage)\n- Context-heavy operations that don't need user interaction\n\n**Example workflow pattern:**\n```\n\nMain Chat: Ask user for requirements (AskUserQuestion)\n↓\nAgent: Research API and create documentation (no user interaction)\n↓\nMain Chat: Review research with user, confirm approach\n↓\nAgent: Generate code based on confirmed plan\n↓\nMain Chat: Present results, handle testing\u002Fdeployment\n\n````\n\u003C\u002Fworkflow_design>\n\u003C\u002Fexecution_model>\n\n\u003Csystem_prompt_guidelines>\n\u003Cprinciple name=\"be_specific\">\nClearly define the agent's role, capabilities, and constraints.\n\u003C\u002Fprinciple>\n\n\u003Cprinciple name=\"use_pure_xml_structure\">\nStructure the system prompt with pure XML tags. Remove ALL markdown headings from the body.\n\n```markdown\n---\nname: security-reviewer\ndescription: Reviews code for security vulnerabilities\ntools: Read, Grep, Glob, Bash\nmodel: sonnet\n---\n\n\u003Crole>\nYou are a senior code reviewer specializing in security.\n\u003C\u002Frole>\n\n\u003Cfocus_areas>\n- SQL injection vulnerabilities\n- XSS attack vectors\n- Authentication\u002Fauthorization issues\n- Sensitive data exposure\n\u003C\u002Ffocus_areas>\n\n\u003Cworkflow>\n1. Read the modified files\n2. Identify security risks\n3. Provide specific remediation steps\n4. Rate severity (Critical\u002FHigh\u002FMedium\u002FLow)\n\u003C\u002Fworkflow>\n````\n\n\u003C\u002Fprinciple>\n\n\u003Cprinciple name=\"task_specific\">\nTailor instructions to the specific task domain. Don't create generic \"helper\" agents.\n\n❌ Bad: \"You are a helpful assistant that helps with code\"\n\u0005✅ Good: \"You are a React component refactoring specialist. Analyze components for hooks best practices, performance anti-patterns, and accessibility issues.\"\n\u003C\u002Fprinciple>\n\u003C\u002Fsystem_prompt_guidelines>\n\n\u003Cagent_xml_structure>\nAgent.md files are system prompts consumed only by Claude. Like skills, they should use pure XML structure for optimal parsing and token efficiency.\n\n\u003Crecommended_tags>\nCommon tags for agent structure:\n\n- `\u003Crole>` - Who the agent is and what it does\n- `\u003Cconstraints>` - Hard rules (NEVER\u002FMUST\u002FALWAYS)\n- `\u003Cfocus_areas>` - What to prioritize\n- `\u003Cworkflow>` - Step-by-step process\n- `\u003Coutput_format>` - How to structure deliverables\n- `\u003Csuccess_criteria>` - Completion criteria\n- `\u003Cvalidation>` - How to verify work\n  \u003C\u002Frecommended_tags>\n\n\u003Cintelligence_rules>\n**Simple agents** (single focused task):\n\n- Use role + constraints + workflow minimum\n- Example: code-reviewer, test-runner\n\n**Medium agents** (multi-step process):\n\n- Add workflow steps, output_format, success_criteria\n- Example: api-researcher, documentation-generator\n\n**Complex agents** (research + generation + validation):\n\n- Add all tags as appropriate including validation, examples\n- Example: mcp-api-researcher, comprehensive-auditor\n  \u003C\u002Fintelligence_rules>\n\n\u003Ccritical_rule>\n**Remove ALL markdown headings (##, ###) from agent body.** Use semantic XML tags instead.\n\nKeep markdown formatting WITHIN content (bold, italic, lists, code blocks, links).\n\nFor XML structure principles and token efficiency details, see `..\u002Fskill-creator\u002Freferences\u002Fxml-tag-guide.md` - the same principles apply to agents.\n\u003C\u002Fcritical_rule>\n\u003C\u002Fagent_xml_structure>\n\n\u003Cinvocation>\n\u003Cautomatic>\nClaude automatically selects agents based on the `description` field when it matches the current task.\n\u003C\u002Fautomatic>\n\n\u003Cexplicit>\nYou can explicitly invoke a agent:\n\n```\n> Use the code-reviewer agent to check my recent changes\n```\n\n```\n> Have the test-writer agent create tests for the new API endpoints\n```\n\n\u003C\u002Fexplicit>\n\u003C\u002Finvocation>\n\n\u003Cbackground_execution>\nAgents can run in the background using the `run_in_background` parameter, allowing parallel execution while the main conversation continues.\n\n\u003Chow_it_works>\n**Starting a background agent:**\nThe Task tool accepts `run_in_background: true` to launch agents asynchronously:\n\n```\nTask tool call:\n- description: \"Analyze security vulnerabilities\"\n- prompt: \"Review all authentication code for security issues...\"\n- subagent_type: \"security-reviewer\"\n- run_in_background: true\n```\n\nThe agent starts immediately and returns an `agent_id` for tracking.\n\u003C\u002Fhow_it_works>\n\n\u003Cretrieving_results>\n**Getting results with TaskOutput:**\nUse the `TaskOutput` tool to retrieve results from background agents:\n\n```\nTaskOutput tool call:\n- task_id: \"agent-12345\"  # The agent_id from the Task call\n- block: true            # Wait for completion (default)\n- timeout: 30000         # Max wait time in ms\n```\n\n**Parameters:**\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `task_id` | Required | The agent ID returned from Task tool |\n| `block` | `true` | Wait for completion or check current status |\n| `timeout` | `30000` | Max wait time in milliseconds (up to 600000) |\n\n**Non-blocking check:**\nSet `block: false` to check status without waiting:\n\n```\nTaskOutput tool call:\n- task_id: \"agent-12345\"\n- block: false\n```\n\nReturns current status: running, completed, or the final result.\n\u003C\u002Fretrieving_results>\n\n\u003Cparallel_agents>\n**Launching multiple agents in parallel:**\n\nTo maximize performance, launch multiple independent agents simultaneously:\n\n```\nSingle message with multiple Task tool calls:\n\nTask 1:\n- description: \"Review code quality\"\n- prompt: \"Check code quality...\"\n- subagent_type: \"code-reviewer\"\n- run_in_background: true\n\nTask 2:\n- description: \"Run security scan\"\n- prompt: \"Scan for vulnerabilities...\"\n- subagent_type: \"security-scanner\"\n- run_in_background: true\n\nTask 3:\n- description: \"Check test coverage\"\n- prompt: \"Analyze test coverage...\"\n- subagent_type: \"test-analyzer\"\n- run_in_background: true\n```\n\nThen retrieve all results:\n\n```\nTaskOutput calls for each agent_id\n```\n\n\u003C\u002Fparallel_agents>\n\n\u003Cwhen_to_use_background>\n**Use background agents for:**\n\n- Long-running analysis (security review, comprehensive code analysis)\n- Multiple independent tasks that can run in parallel\n- Tasks where you want to continue working while waiting\n- Research tasks that may take significant time\n\n**Don't use background for:**\n\n- Quick operations (\u003C 10 seconds)\n- Tasks that depend on each other sequentially\n- Tasks where immediate results are needed for next step\n- Simple single-file operations\n\n**Pattern: Parallel Analysis Pipeline**\n\n```\n1. Launch multiple analysis agents in background\n2. Continue with other work or wait\n3. Collect all results\n4. Synthesize findings in main conversation\n```\n\n\u003C\u002Fwhen_to_use_background>\n\n\u003Cresuming_agents>\n**Resuming agents:**\nAgents can be resumed using the `resume` parameter with their agent ID:\n\n```\nTask tool call:\n- description: \"Continue security review\"\n- prompt: \"Please continue with the remaining files...\"\n- subagent_type: \"security-reviewer\"\n- resume: \"agent-12345\"  # Previous agent ID\n```\n\nThe agent continues with its full previous context preserved.\n\u003C\u002Fresuming_agents>\n\u003C\u002Fbackground_execution>\n\n\u003Cmanagement>\n\u003Cusing_agents_command>\nRun `\u002Fagents` for an interactive interface to:\n- View all available agents\n- Create new agents\n- Edit existing agents\n- Delete custom agents\n\u003C\u002Fusing_agents_command>\n\n\u003Cmanual_editing>\nYou can also edit agent files directly:\n\n- Project: `.claude\u002Fagents\u002Fagent-name.md`\n- User: `~\u002F.claude\u002Fagents\u002Fagent-name.md`\n  \u003C\u002Fmanual_editing>\n  \u003C\u002Fmanagement>\n\n\u003Creference>\n**Core references**:\n\n**Agent usage and configuration**: [references\u002Fagents.md](references\u002Fagents.md)\n\n- File format and configuration\n- Model selection (Sonnet 4.5 + Haiku 4.5 orchestration)\n- Tool security and least privilege\n- Prompt caching optimization\n- **Background execution** (run_in_background, TaskOutput, parallel agents)\n- Complete examples\n\n**Writing effective prompts**: [references\u002Fwriting-agent-prompts.md](references\u002Fwriting-agent-prompts.md)\n\n- Core principles and XML structure\n- Description field optimization for routing\n- Extended thinking for complex reasoning\n- Security constraints and strong modal verbs\n- Success criteria definition\n\n**Advanced topics**:\n\n**Evaluation and testing**: [references\u002Fevaluation-and-testing.md](references\u002Fevaluation-and-testing.md)\n\n- Evaluation metrics (task completion, tool correctness, robustness)\n- Testing strategies (offline, simulation, online monitoring)\n- Evaluation-driven development\n- G-Eval for custom criteria\n\n**Error handling and recovery**: [references\u002Ferror-handling-and-recovery.md](references\u002Ferror-handling-and-recovery.md)\n\n- Common failure modes and causes\n- Recovery strategies (graceful degradation, retry, circuit breakers)\n- Structured communication and observability\n- Anti-patterns to avoid\n\n**Context management**: [references\u002Fcontext-management.md](references\u002Fcontext-management.md)\n\n- Memory architecture (STM, LTM, working memory)\n- Context strategies (summarization, sliding window, scratchpads)\n- Managing long-running tasks\n- Prompt caching interaction\n\n**Orchestration patterns**: [references\u002Forchestration-patterns.md](references\u002Forchestration-patterns.md)\n\n- Sequential, parallel, hierarchical, coordinator patterns\n- Sonnet + Haiku orchestration for cost\u002Fperformance\n- Multi-agent coordination\n- Pattern selection guidance\n\n**Debugging and troubleshooting**: [references\u002Fdebugging-agents.md](references\u002Fdebugging-agents.md)\n\n- Logging, tracing, and correlation IDs\n- Common failure types (hallucinations, format errors, tool misuse)\n- Diagnostic procedures\n- Continuous monitoring\n  \u003C\u002Freference>\n\n\u003Csuccess_criteria>\nA well-configured agent has:\n\n- Valid YAML frontmatter (name matches file, description includes triggers)\n- Clear role definition in system prompt\n- Appropriate tool restrictions (least privilege)\n- XML-structured system prompt with role, approach, and constraints\n- Description field optimized for automatic routing\n- Successfully tested on representative tasks\n- Model selection appropriate for task complexity (Sonnet for reasoning, Haiku for simple tasks)\n  \u003C\u002Fsuccess_criteria>\n",{"name":5,"description":7},"master","46a82833a3f7f6f5f408511246a55e0485c5f7ab","skills\u002Fagents-managers",1,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills","https:\u002F\u002Fgithub.com\u002Fzuzu59","2026-09-30 17:58:42",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add zuzu59\u002Fz-skills --skill agents-managers",{"tool":37,"command":38},"bunx","bunx skills add zuzu59\u002Fz-skills --skill agents-managers",{"tool":40,"command":41},"pnpm","pnpm dlx skills add zuzu59\u002Fz-skills --skill agents-managers","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002FSKILL.md",true,[45,51,55,59,63,67,71],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"references\u002Fagents.md","REFERENCE","参考文档",17152,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Fagents.md",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fcontext-management.md",15477,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Fcontext-management.md",{"path":56,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Fdebugging-agents.md",18838,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Fdebugging-agents.md",{"path":60,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Ferror-handling-and-recovery.md",14806,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Ferror-handling-and-recovery.md",{"path":64,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Fevaluation-and-testing.md",11049,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Fevaluation-and-testing.md",{"path":68,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":69,"sourceHtmlUrl":70},"references\u002Forchestration-patterns.md",17169,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Forchestration-patterns.md",{"path":72,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":73,"sourceHtmlUrl":74},"references\u002Fwriting-agent-prompts.md",14116,"https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fagents-managers\u002Freferences\u002Fwriting-agent-prompts.md",[76,83,90,97,104,111],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M76SQGVSKRYNQX4AA4RXV3K0","writing-great-skills","Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fwriting-great-skills\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M76SQDSAZSA1272D56JY8Q2G","use-style","Apply named visual style guides to landing pages, app shells, and UI. Use for $use-style, \u002Fuseskill, list styles, or styles like ios-app, grid, vercel, black-grid, stripe, linear, raycast, gumroad, dusk, or luma.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fuse-style\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M76SQAXDVX02V4JH4PF1MC9G","use-goal","Use when the user asks to create, draft, set, start, or refine a Codex or Claude Code `\u002Fgoal` objective for persistent multi-turn work.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fuse-goal\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":19},"01M76SQ8B11GTTGQ5CZHHGXFXM","use-fable","Orchestrator mode for Fable: Fable thinks, plans, and reviews while Codex CLI and subagents execute all heavy work. Use when the user invokes \u002Fuse-fable, says 'use fable', 'fable mode', 'orchestrator mode', or wants to save Fable tokens\u002Frate limits.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fuse-fable\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":19},"01M76SQ4SB7Q4CXEGGPA75KGFK","use-delegate","Delegation mode: the host agent (Claude or Codex) plans and reviews while heavy work runs on cheap executors: OpenCode Kimi K3, Codex GPT-5.6 terra\u002Fsol. Use when the user invokes \u002Fuse-delegate, says 'use delegate', 'delegate mode', 'orchestrator mode', or wants to save tokens\u002Frate limits.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fuse-delegate\u002FSKILL.md",[],[],{"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":117,"publishedAt":19},"01M76SQ24836GDGMGGPV6Z2B7N","use-artifacts","Create Claude-style local HTML artifacts under ~\u002F.agents\u002Fartifacts. Use for plans, prototypes, visualizations, dashboards, diagrams, options, or substantial reusable content, with a portable local runtime and no Anthropic API.","https:\u002F\u002Fgithub.com\u002Fzuzu59\u002Fz-skills\u002Fblob\u002Fmaster\u002Fskills\u002Fuse-artifacts\u002FSKILL.md",[],[]]