Workflow Router Skill

You are the workflow-router skill. Your role is to determine the next action in the BAZINGA workflow by calling workflow_router.py, which reads from the state machine configuration.

Overview

This skill determines what to do after receiving an agent response by:

• Reading transitions from database (seeded from transitions.json)
• Applying testing mode rules (skip QA if disabled/minimal)
• Applying escalation rules (escalate to SSE after N failures)
• Applying security rules (force SSE for security tasks)
• Returning JSON with next action

Prerequisites

• Database must be initialized (bazinga/bazinga.db exists)
• Config must be seeded (run config-seeder skill at session start)

When to Invoke This Skill

• AFTER receiving any agent response
• When orchestrator needs to decide what to do next based on agent's status code
• Examples:

- Developer returned "READYFORQA" → Route to QA Expert - QA returned "PASS" → Route to Tech Lead - Tech Lead returned "APPROVED" → Check phase, maybe route to PM

Your Task

When invoked, you must:

Step 1: Extract Parameters

Parse from the agent's response and current state:

• currentagent: Agent that just responded (developer, qaexpert, tech_lead, etc.)
• status: Status code from response (READYFORQA, PASS, APPROVED, etc.)
• session_id: Current session ID
• group_id: Current group ID
• testing_mode: full, minimal, or disabled

Step 2: Call the Python Script

python3 .claude/skills/workflow-router/scripts/workflow_router.py \
  --current-agent "{current_agent}" \
  --status "{status}" \
  --session-id "{session_id}" \
  --group-id "{group_id}" \
  --testing-mode "{testing_mode}"

Step 3: Parse and Return Result

The script outputs JSON:

{
  "success": true,
  "current_agent": "developer",
  "response_status": "READY_FOR_QA",
  "next_agent": "qa_expert",
  "action": "spawn",
  "model": "sonnet",
  "group_id": "AUTH",
  "session_id": "bazinga_xxx",
  "include_context": ["dev_output", "test_results"]
}

Return this JSON to the orchestrator.

Actions Explained

Validator Workflow

When PM sends BAZINGA, the orchestrator invokes bazinga-validator. After validator returns:

Orchestrator calls workflow-router skill:

workflow-router, determine next action:
Current agent: validator
Status: ACCEPT  # or REJECT
Session ID: {session_id}

Then invoke: Skill(command: "workflow-router")

Note: Validator is session-scoped - omit group_id (not needed for routing).

Transitions defined in workflow/transitions.json → validator section:

• ACCEPT → end_session action → Complete shutdown protocol
• REJECT → spawn action with nextagent: projectmanager → PM fixes issues

🔴 CRITICAL: After validator REJECT, orchestrator MUST spawn PM with the rejection details. Do NOT stop!

Special Flags in Result

Output Format

JSON object with routing decision. Parse and execute the action.

Error Handling

If routing fails, use the fallback_action in the error response.