Runtime Copilot

Use this skill when working with the Mini Data Engine MCP server from Codex.

The goal is to treat the runtime as a self-describing operational surface, not as a pile of scripts.

Quick start

1. Call project_tool_catalog to see the current MCP surface.
2. Call project_get_defaults to load default roots and runtime assumptions.
3. Use health_check for a fast smoke test.
4. Use project_run_regression when the user wants a release-style verdict.
5. Use explain_run and similar_incidents when diagnosing failures.

Working rules

• Prefer MCP tools before raw shell commands when the capability already exists in MCP.
• Use shell commands only for setup gaps, file inspection, or changes outside the MCP surface.
• When summarizing runtime state, mention the exact MCP tools used.
• If the user asks what is available, answer from project_tool_catalog, not from memory.
• If the user asks about defaults, answer from project_get_defaults, not from scattered code paths.

Recommended tool order

For discovery:

• project_tool_catalog
• project_get_defaults
• project_manifest
• project_capabilities

For health and diagnostics:

• health_check
• benchmark_calls
• scenario_load_test
• explain_run
• similar_incidents

For regression and release confidence:

• project_run_regression
• project_capture_baseline
• project_compare_baseline
• decision_gate

Example requests that should trigger this skill

• "Show me what this MCP runtime can do."
• "Check whether the runtime is healthy."
• "Run the release gate and tell me if this is safe."
• "Explain why this run failed."
• "Compare current runtime behavior to the last baseline."

Default answer shape

Keep answers operational and short:

• what was checked
• what passed or failed
• what changed
• what the next action is