/multi-agent Command

Multi-agent workflow status and compliance tracking.

Actions

/multi-agent or /multi-agent status

Show current multi-agent compliance status:

1. Check for pending reviews:

   bash scripts/multi-agent-metrics.sh check
   

2. Show quick compliance stats:

   bash scripts/multi-agent-metrics.sh json
   

3. Report summary to user including:

   • Pending code reviews count
   • Recent compliance rate
   • Recommendations for improvement

/multi-agent report

Show full compliance report with details:

bash scripts/multi-agent-metrics.sh report

This shows:

• Total events tracked
• Violation counts by type
• Compliant usage counts
• Compliance rate with visual bar
• Recent events history

/multi-agent check

Check if there are pending reviews that must be done before committing:

bash scripts/multi-agent-metrics.sh check

IMPORTANT: If pending reviews exist, you MUST invoke a code-reviewer agent before committing.

/multi-agent clear

Mark pending reviews as complete (only after actually running code review):

bash scripts/multi-agent-metrics.sh track code_change_reviewed

/multi-agent help

Display this help and link to documentation:

• .claude/rules/multi-agent-workflow.md - Full workflow rules
• scripts/multi-agent-metrics.sh - Metrics tracking script

Multi-Agent Workflow Requirements

ALWAYS Use Subagents For:

Task Type	Required Agents	Minimum
Code changes	code-reviewer, test-automator	2
Research questions	Explore (multiple paths)	2-3
Bug investigation	debugger, error-detective	2
New features	architect-review, code-reviewer	2

Skip Multi-Agent Only For:

• Simple file reads (single file)
• Direct questions with known answers
• Trivial edits (typos, formatting)

Agent Invocation Examples

After Writing Code

Launch parallel agents:
1. Task with subagent_type: "pr-review-toolkit:code-reviewer"
2. Task with subagent_type: "unit-testing:test-automator"

For Research Questions

Launch parallel Explore agents:
1. Task with subagent_type: "Explore", thoroughness: "medium"
2. Task with subagent_type: "Explore", different search angle

For Bug Fixes

Launch parallel debugging agents:
1. Task with subagent_type: "debugging-toolkit:debugger"
2. Task with subagent_type: "error-debugging:error-detective"

Compliance Tracking

Events tracked automatically:

Event	Trigger
code_change_no_review	Write/Edit tool used
code_change_reviewed	code-reviewer agent invoked
agent_invoked	Any Task tool call

View compliance with /multi-agent report.

Integration with Commit Workflow

Before committing:

1. Run /multi-agent check to see pending reviews
2. If pending, invoke code-reviewer agent
3. Then proceed with commit

The /commit command should also check for pending reviews.