Workflow Enforcer Skill

1. Check for agent: Is there a specialized agent for this task?
2. Announce: "I'm using [agent] to [action]"
3. Create todos: Use TodoWrite for multi-step work
4. Track: Mark in_progress → completed

1. Check for specialized agent - Is there an agent for this task type?
2. Announce usage - "I'm using [agent] to [action]"
3. Create todos/tasks - Use TaskCreate (preferred) or TodoWrite for multi-step work
4. Track progress - Mark in_progress before starting, completed after finishing

Quick Reference

For the complete 70+ agent catalog, see reference/agents-catalog.md.

How to Use

Step 1: Identify Task Type

Categorize the request:

• Debugging? → debugging agents
• Code review? → review agents
• Testing? → test automation agents
• Performance? → performance engineers
• Security? → security auditors
• Deployment? → deployment engineers

Step 2: Announce

Before starting work:

"I'm using the [agent-name] to [what you're doing]"

Examples:

• "I'm using debugging-toolkit:debugger to trace this authentication error"
• "I'm using python-development:python-pro to refactor this async code"

Step 3: Create Todos / Tasks

For multi-step tasks, use TaskCreate (preferred — renders live UI spinners) or TodoWrite (fallback):

TaskCreate (native progress UI):

1. Break work into specific items
2. Use TaskCreate({ subject: "...", activeForm: "..." }) for each item
3. Use TaskUpdate({ taskId, status: "in_progress" }) before starting → shows spinner
4. Use TaskUpdate({ taskId, status: "completed" }) after finishing → shows checkmark
5. Use addBlockedBy for sequential dependencies between tasks

TodoWrite (simpler, text-based):

1. Break work into specific items
2. Use TodoWrite to create the list
3. Mark in_progress before starting, completed after finishing

Step 4: Follow Agent Discipline

Each agent type has its own methodology:

• TDD agents → Write tests first
• Debugging agents → Systematic root cause analysis
• Code review agents → Follow review checklist
• Deployment agents → Follow deployment protocols

Common Rationalizations to Avoid

If you think any of these, STOP and use the appropriate agent:

• "This is simple, I don't need an agent"
• "Let me just quickly fix this"
• "I can debug this manually"
• "This doesn't need a formal review"
• "I'll skip the test for now"

The right thought: "What specialized agent should I use for this task?"

Guidelines

• This skill applies to EVERY session, EVERY project, EVERY task
• No exceptions, no rationalizations, no shortcuts
• When in doubt, check reference/agents-catalog.md

Emit Outcome Sidecar

As the final step, write to ~/.claude/skill-analytics/last-outcome-workflow-enforcer.json:

{"ts":"[UTC ISO8601]","skill":"workflow-enforcer","version":"1.0.0","variant":"default",
 "status":"[success|partial|error]","runtime_ms":[estimated ms from start],
 "metrics":{"checks_performed":[n],"agents_recommended":[n],"violations_caught":[n]},
 "error":null,"session_id":"[YYYY-MM-DD]"}

Use status "partial" if some stages failed but results were produced. Use "error" only if no output was generated.