| name | workflow |
| description | Orchestrate multi-step workflows by decomposing into subtasks, dispatching them, and monitoring via foreground polling loops. Prevents session timeout kills during long-running orchestration. Use when coordinating multiple agents or running multi-phase work that takes more than a few minutes. |
Workflow Orchestrator
Read the full workflow from .claude/commands/workflow.md and execute it.
Quick Summary
- Decompose — break the user's request into discrete subtasks with dependencies
- Dispatch — send subtasks to other agents via
dispatch_task (with /do instructions) and verify each task started with the intended title, profile, and constraints
- Poll — foreground
sleep 300 + get_task_details loop keeps the session alive
- React — dispatch dependent tasks as predecessors complete, retry failures
- Complete — summarize results when all subtasks finish
Why This Exists
When Claude Code dispatches subtasks and waits passively, the ACP session appears idle and the control plane kills it. This skill uses explicit foreground polling (Bash sleep + MCP tool calls) to maintain visible session activity throughout the orchestration.
Staging Debugging Access
All agents have access to $CF_TOKEN for direct Cloudflare API queries against staging. When monitoring subtasks that deploy to staging, use the CF API to verify their work landed correctly — query D1 for data state, read KV for feature flags, check DNS for routing. See .claude/rules/32-cf-api-debugging.md for the full cheat sheet.
State Persistence
Maintain .workflow-state.md (gitignored) as external memory. Re-read it before every poll cycle. This survives context compaction. See .claude/commands/workflow.md for the full state file format.
