MoonMcp: Skill for Claude Code

Run Coordination

Drive assessment work through persistent state rather than chat memory. Keep the coordinator responsible for task boundaries, policy, evidence integration, and stopping conditions—not for doing every specialist task itself.

Adapted for security runs from obra/superpowers' MIT-licensed executing-plans, dispatching-parallel-agents, and finishing-a-development-branch workflows.

Resume before acting

1. Call run_status(run_id, include_events=true).
2. Recall memory_brief(target) and inspect runs://recent when the run ID is unknown.
3. Verify the run is running, scope/program identity is still correct, and budgets remain.
4. Inspect unfinished tasks, dependencies, coverage, and recent artifacts.
5. Do not repeat completed work unless the evidence is stale or invalid.

Dispatch safely

Use run_next for the highest-priority unblocked task. Delegate only tasks that are independent in target state and do not mutate the same session or workflow.

Each worker receives:

• run ID and task ID;
• exact target and scope constraints;
• role and falsifiable objective;
• relevant curated memory and untrusted evidence references;
• permitted risk tier and remaining budget;
• required output/evidence format;
• explicit stop conditions.

Parallelize passive research or independent hosts. Serialize authentication flows, shared-account mutations, rate-limit tests, purchases, password resets, races, and any task whose result changes another task's baseline.

Task lifecycle

1. Set the selected task to running with run_task_update.
2. Execute only its bounded objective.
3. Store material evidence with run_artifact_add.
4. Update relevant coverage with the evidence reference.
5. Consume measured usage through run_usage.
6. Set the task to completed, blocked, failed, or cancelled; include a concise result.
7. Create follow-up tasks only when new evidence justifies them. Deduplicate against existing tasks and memory.

Never mark a task complete merely because a tool returned successfully. Completion means its hypothesis or deliverable was resolved.