---

name: simulation-run description: "Starts, monitors, stops, inspects, interviews, and finalizes MiroFish OASIS simulation runs across parallel, Twitter, Reddit, IPC, and optional Zep graph-memory modes." disable-model-invocation: true metadata: disco-role: operating license: AGPL 3.0

simulation-run

Use this sub-skill when a task asks to start a prepared MiroFish simulation, choose parallel/twitter/reddit, poll run status, inspect timelines/actions/posts/comments/stats, interview agents, close the interactive environment, stop a process, or reason about graph-memory update finalization.

Route first

• If the request is about building or resetting the Zep graph, route to the sibling graph-build sub-skill.
• If the request is about creating a simulation, generating profiles, or producing simulation_config.json, route to the sibling simulation-setup sub-skill.
• If the run has reached a terminal successful or stopped state and the user asks for an analysis/report, route to the sibling reporting sub-skill after confirming graph-memory finalization is not still pending.

Read or run the bundled material

• Read references/workflows.md for start modes, monitor loops, stop versus close-env decisions, interviews, graph-memory finalization, and report handoff.
• Read references/api-reference.md when constructing backend API calls, interpreting response fields, or diagnosing status codes.
• Read references/runtime-artifacts.md when inspecting state.json, run_state.json, platform actions.jsonl, SQLite platform databases, logs, or IPC directories.
• Read references/troubleshooting.md when a start/stop/restart/interview/memory-update operation fails, hangs, or returns a pending barrier.
• Run python scripts/action_logger.py --help to inspect the bundled JSONL helper, or python scripts/action_logger.py --self-test to verify the helper writes and reads the MiroFish action-log shape without starting OASIS.

Minimal safe operating loop

1. Confirm the backend is reachable and you have a simulation_id whose preparation is complete.
2. Start with POST /api/simulation/start using a real JSON boolean for force and enable_graph_memory_update when present.
3. Poll GET /api/simulation/<simulation_id>/run-status for lightweight progress; use detail/actions/timeline endpoints only when you need event data.
4. Use POST /api/simulation/env-status before any interview. Interviews require a live wait-mode environment.
5. Prefer POST /api/simulation/close-env for a completed interactive environment; use POST /api/simulation/stop for running, stuck, or abort-required processes.
6. Treat stopping as a non-terminal graph-ingestion barrier. Do not generate reports until it resolves to completed or stopped; investigate or retry if it resolves to failed.