---

name: simulation-setup description: "Create and prepare MiroFish simulation setup artifacts from a completed Zep graph, including entity filtering, OASIS profiles, simulation_config.json, realtime setup status, and setup downloads." disable-model-invocation: true metadata: disco-role: operating license: AGPL 3.0

simulation-setup

Use this sub-skill when a task asks to prepare a MiroFish simulation after graph build: create a simulation record, inspect/filter graph entities, generate OASIS Agent profiles/personas, produce simulation_config.json, watch realtime Step 2 setup progress, download setup artifacts, or debug "no matching entities" and partial setup artifacts.

Route first

• If the user still needs seed documents uploaded, ontology generated, graph build started/polled, or graph reset/delete, route to the sibling graph-build sub-skill.
• If the user asks to start/stop a simulation, monitor runtime rounds/actions/timelines/posts/comments/stats, interview agents, or finalize a live environment, route to the sibling simulation-run sub-skill after setup is ready.
• If the user asks for generated reports or report graph reads, route to the sibling reporting sub-skill after runtime completion.

Read or run the bundled material

• Read references/workflows.md for the end-to-end Step 2 flow: create simulation, optional entity inspection, prepare, realtime profile/config polling, artifact validation, downloads, and handoff to simulation-run.
• Read references/api-reference.md when constructing /api/simulation setup, entity, profile, config, download, environment-status, or close-env calls and interpreting response fields/statuses.
• Read references/profile-formats.md before validating twitter_profiles.csv, reddit_profiles.json, standalone /generate-profiles output, or simulation_config.json shape and platform defaults.
• Read references/troubleshooting.md when setup fails, stalls, reuses stale artifacts, reports zero entities, receives an invalid platform, has partial JSON/CSV writes, or lacks required keys.
• Run scripts/profile-format-smoke.py with python scripts/profile-format-smoke.py --self-test to smoke-check the bundled current profile-file formats without importing MiroFish or starting OASIS. Use --help for options.

Minimal safe operating loop

1. Confirm there is a completed project graph: a project_id, a usable graph_id, and a non-empty project simulation_requirement.
2. Create or reuse a simulation record with POST /api/simulation/create; the UI default enables both Twitter and Reddit.
3. If entity readiness is uncertain, inspect GET /api/simulation/entities/<graph_id>?enrich=false first, then narrow with entity_types only after you see available custom labels.
4. Start preparation with POST /api/simulation/prepare. Do not pass runtime max_rounds here; setup generates time/config parameters from the simulation requirement.
5. Poll POST /api/simulation/prepare/status plus realtime profiles/realtime and config/realtime until config_generated is true and the simulation state is ready or the task fails with an explicit error.
6. Validate profile count, profile file format, agent_configs, initial_posts[*].poster_agent_id, enabled platform configs, and state.json flags before proceeding.
7. For POST /api/simulation/start, runtime status, interviews, close/stop, and graph-memory updates, switch to simulation-run.