菜单
Use when deciding whether to launch an agent, selecting which agent to use, or coordinating multiple agents. Covers delegation criteria, external-model patterns, task isolation, and agent selection strategies.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/MadAppGang/claude-code --skill agent-coordination-discipline命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use when starting isolated feature work or before executing implementation plans. Manages full worktree lifecycle from creation through cleanup with safety checks and error recovery.
Configuration reference and troubleshooting for the statusline plugin — sections, themes, bar widths, and script architecture
Common agent patterns and templates for Claude Code. Use when implementing agents to follow proven patterns for Tasks integration, quality checks, and external model invocation via claudish CLI.
正在显示 SKILL.md
基于 SOC 职业分类
| name | agent-coordination-discipline |
| description | Use when deciding whether to launch an agent, selecting which agent to use, or coordinating multiple agents. Covers delegation criteria, external-model patterns, task isolation, and agent selection strategies. |
| keywords | ["agent-coordination","external-model","task-isolation","delegation-criteria","multi-agent","external-model","claudish","orchestration","agent-selection","Task-tool","developer-agent","architect-agent","grok-code-fast","sonnet-4-5","thinking-budget"] |
| created | "2026-01-20T00:00:00.000Z" |
| updated | "2026-01-20T00:00:00.000Z" |
| plugin | dev |
| type | discipline |
| difficulty | intermediate |
Iron Law: "NO AGENT LAUNCH WITHOUT CLEAR DELEGATION CRITERIA"
Use this skill when:
This skill prevents premature agent launches, redundant agent usage, and poor task isolation that wastes thinking budget and causes coordination failures.
Does the task require:
├─ Single tool call (grep, read, edit)?
│ └─ ✗ NO AGENT - Use native tool directly
├─ 2-3 sequential tool calls?
│ └─ ✗ NO AGENT - Use tools directly in sequence
├─ Multi-step investigation with branching logic?
│ └─ ✓ AGENT - Task tool with developer/architect agent
├─ External model expertise (Grok, DeepSeek, etc.)?
│ └─ ✓ AGENT - external-model pattern with model specification
├─ Parallel exploration of multiple code paths?
│ └─ ✓ AGENT - Multiple Task calls with coordination
└─ High-risk change needing isolation?
└─ ✓ AGENT - Task tool with sandbox/review focus
Every agent task must be independently executable:
Bad (not isolated):
Task: "Fix the bug we discussed earlier"
Good (properly isolated):
Task: "Debug the TypeError in src/components/UserProfile.tsx line 42.
Context: User reports 'Cannot read property name of undefined' when viewing profile page.
Evidence: Error occurs after recent commit abc123 that changed user data structure.
Success criteria: Identify root cause, propose fix, verify with test scenario."
When delegating to external models via claudish CLI:
Structure:
claudish --model {model_id} --stdin --quiet <<EOF > output.md
{Task Description}
Context:
- {Relevant file paths}
- {Current state}
- {Related decisions}
Success Criteria:
- {What constitutes success}
- {Expected output format}
Constraints:
- {Time limits}
- {Tool restrictions}
- {Quality requirements}
EOF
Example:
claudish --model x-ai/grok-code-fast-1 --stdin --quiet <<EOF > analysis.md
Analyze the React component rendering performance issue in Dashboard.tsx.
Context:
- File: src/components/Dashboard.tsx (247 lines)
- Issue: Component re-renders 40+ times on data updates
- Recent changes: Added real-time WebSocket updates in commit f4a2c1b
Success Criteria:
- Identify unnecessary re-renders (provide line numbers)
- Propose memoization strategy
- Estimate performance improvement
Constraints:
- Max 3 minutes analysis time
- Focus on React 19 compiler-friendly patterns
EOF
Trigger: Task requires 5+ tool calls with conditional branching Agent: developer, architect Example: "Trace data flow through 3 layers to find where user.email becomes null"
Trigger: Need specialized model capabilities (code speed, vision, reasoning) Agent: external-model with specific model Example: "Use Grok Code Fast to refactor 15 files for consistency in < 2 minutes"
Trigger: Multiple independent tasks that can run simultaneously Agent: Multiple Task calls with result aggregation Example: "Analyze frontend performance (Task 1) while auditing API security (Task 2)"
Trigger: High-risk changes needing review before merging to main workflow Agent: review-focused agent with checkpoint Example: "Evaluate if this database migration will cause downtime"
Trigger: Current agent lacks specific skill that another agent has Agent: specialist agent (security, performance, accessibility) Example: "Launch accessibility agent to audit ARIA compliance"
Instead: Use native Grep or Glob tool directly
# ✗ DON'T
Task: "Find all files using the deprecated API"
# ✓ DO
Grep("oldApiCall", output_mode: "files_with_matches", type: "js")
Instead: Use tool directly
# ✗ DON'T
Task: "Read the config file and tell me the API URL"
# ✓ DO
Read("/path/to/config.json")
// Parse and extract apiUrl field
Instead: Execute the decision
# ✗ DON'T
Task: "I think we should use React Query. What do you think?"
# ✓ DO
// Just implement React Query since decision is made
Write("src/hooks/useApiQuery.ts", reactQueryCode)
Instead: Chain tools directly
# ✗ DON'T
Task: "Find the function, read it, and edit it"
# ✓ DO
Grep("functionName", output_mode: "files_with_matches")
// => result: src/utils/helper.ts
Read("src/utils/helper.ts")
Edit("src/utils/helper.ts", old_string, new_string)
Instead: Handle in current agent
# ✗ DON'T
Task: "Based on our earlier discussion about performance vs. maintainability trade-offs, decide if we should cache this"
# ✓ DO
// Current agent already has context, make decision directly
if (performanceIsCritical) {
implementCaching()
}
| Task Type | Best Agent | Model | Reasoning |
|---|---|---|---|
| Debugging errors | developer | sonnet-4-5 | Deep reasoning, context retention |
| Design review | architect | sonnet-4-5 | System thinking, trade-off evaluation |
| Code generation | developer | grok-code-fast | Speed for repetitive patterns |
| Multi-codebase analysis | developer | sonnet-4-5 | Cross-repo understanding |
| Performance profiling | developer + external-model | grok-code-fast | Fast scanning + specific optimization |
| Security audit | security (if available) | sonnet-4-5 | Nuanced threat modeling |
| Documentation generation | developer | grok-code-fast | Fast, straightforward task |
| Refactoring (large scope) | developer | sonnet-4-5 | Maintain consistency across changes |
Fast Execution (< 2 min):
x-ai/grok-code-fast-1 - Code generation, refactoring, simple analysisanthropic/claude-3-5-haiku - Quick decisions, data transformationDeep Reasoning (> 2 min):
anthropic/claude-sonnet-4-5 - Complex debugging, architecture designgoogle/gemini-2.0-flash-thinking-exp-01-21 - Extended thinking budgetSpecialized:
Minimal (< 1000 tokens):
Moderate (1000-5000 tokens):
Full (5000+ tokens):
Must include:
Example:
Success Criteria:
- Output: JSON array of {file, line, issue, suggestion}
- Completeness: All React components in src/ analyzed
- Quality: Each suggestion must include before/after code
- Constraints: Complete within 5 minutes, use only Read/Grep tools
Pattern:
1. Launch agent with external-model
2. Capture result in variable or file
3. Validate result against success criteria
4. Route to next step:
- If success: Use result in main workflow
- If partial: Request clarification
- If failure: Fall back to native tools
Example:
result = Task("external-model: x-ai/grok-code-fast-1\n\nRefactor 10 components for React 19...")
if (result.contains("Refactored successfully")) {
// Apply changes to codebase
applyRefactorings(result.changes)
} else {
// Fall back to manual refactoring
manualRefactor()
}
Before launching an agent, verify:
# ✗ VIOLATION: Agent for single grep
Task: "Find all files importing the old database client"
# ✓ CORRECT: Use native tool
Grep("from old_db_client import", type: "py", output_mode: "files_with_matches")
// ✓ CORRECT: Multi-step investigation with agent
Task: "Debug the race condition in WebSocket message handling.
Context:
- File: src/services/websocket.ts (342 lines)
- Issue: Messages arrive out of order 5% of the time
- Environment: Production only (not reproducible in dev)
- Recent changes: Added message batching in commit a3f9c21
Success Criteria:
- Identify race condition root cause (provide line numbers)
- Propose synchronization strategy
- Verify solution handles edge cases
Constraints:
- Max 10 minutes analysis
- Use Read, Grep, and Bash tools only
- No code changes (diagnosis only)"
// ✓ CORRECT: Fast refactoring with Grok
external-model: x-ai/grok-code-fast-1
Refactor 15 handler functions in handlers/ to use consistent error handling pattern.
Context:
- Directory: internal/handlers/ (15 files, ~200 lines each)
- Current state: Inconsistent error responses (some use Error(), some use Errorf(), some return raw errors)
- Target pattern: Use custom AppError type with status codes and messages
Success Criteria:
- All 15 handlers use AppError consistently
- Preserve existing business logic (only change error handling)
- Provide git diff summary
Constraints:
- Complete within 3 minutes
- Use Read and Grep tools for analysis
- Return refactored code for all 15 files
Works with:
Prevents:
| Anti-Pattern | ✗ Without Discipline | ✓ With Discipline |
|---|---|---|
| Trivial task delegation | Launch agent to run single grep | Use Grep tool directly |
| Missing isolation | "Fix the bug we discussed" | "Debug TypeError in UserProfile.tsx line 42: 'Cannot read property name of undefined'. Context: ..." |
| No success criteria | "Analyze the performance issue" | "Identify re-render causes (line numbers), propose memoization, estimate improvement %" |
| Wrong model selection | Use sonnet-4-5 for simple refactoring | Use grok-code-fast for speed |
| No result validation | Launch agent, assume success | Check result against success criteria, have fallback plan |
| Coordination failure | Launch 3 agents, hope they coordinate | Define result routing: Agent 1 → validate → Agent 2 → aggregate |
Detection:
Correction:
Validation:
Agent Task Checklist (all must be true):
✓ Task requires 5+ tool calls OR external model expertise
✓ Success criteria defined (output format, completeness, quality bar)
✓ Context is self-contained (no references to earlier discussion)
✓ Model selection justified (speed vs. reasoning trade-off considered)
✓ Result routing planned (validation + next steps)
✓ Error handling defined (fallback if agent fails)
✓ Native tools attempted first (or explicitly not applicable)
Related Skills:
verification-before-completion - Validate agent resultssystematic-debugging - Multi-step debugging investigationsorchestration/multi-agent-orchestration - Complex coordination patternsVersion: 1.0.0 Last Updated: 2026-01-20