Workflow Runner

Display available workflows with descriptions:

Available workflows:
  feature  — Full feature implementation with TDD and review
  hotfix   — Quick fix with minimal review
  refactor — Code improvement without behavior changes

3. Execute Workflow

For the selected workflow, execute steps in order:

Sequential steps — run one after another, passing context forward. Parallel steps — use parallel_with to run simultaneously (launch subagents in parallel). Audit steps — run ops/audit-task.sh for automated checks. Memory steps — trigger memory consolidation via memory-consolidator agent.

Step Execution Protocol

For each step:

1. Announce: "▶ Step N: {agent} — {description}"
2. Execute: Invoke the agent as a subagent with the task context
3. Capture: Note key outputs/decisions from the agent
4. Update memory: Write a brief entry to .claude/memory/ with what was done
5. Check: If step failed, STOP and report (unless continue_on_failure: true)

4. Audit Gate

If the workflow has audit: true steps:

bash ops/audit-task.sh . "workflow: {workflow_name}"

Report results. If audit FAILS:

• Show findings
• Ask developer: "Audit failed. Fix issues and retry? [y/n]"
• Do NOT proceed to next steps until audit passes

5. Summary

After all steps complete:

── Workflow Complete: {name} ──
  Steps executed: N/N
  Duration: ~Xm
  Audit: PASS/FAIL
  Memory: Updated

Pipeline YAML Schema