mike-runner

Run MIKE+ simulations headless via the mike-plus MCP server and report result files + engine status. Use when an agent must execute a MIKE+ .sqlite model's active (or a named) simulation without openi

SKILL.md

GitHub

MIKE+ Runner

Executes a MIKE+ model headless through the mike-plus MCP server (engine: MIKE 1D).

When to use

After a model is ready and you need to actually run it — the active simulation, or a named setup from msm_Project. Do not use this to read results (that's mike-results) or to plot (mike-plot). The whole-workflow rules (run directory, copy policy, stop rules, QA gates) are in mike-end-to-end; a baseline-vs-variant pair of runs is compared with mike-compare.

Tool

  • mike_run — args: sqlite (path), simulation (optional setup id), timeout_s (optional). Returns {ok, active_simulation, result_files[], elapsed_s}, a parsed QA gate {completed, errors, warnings, issues, status, log_file}, and _log_tail (engine console).

Conventions (evidence boundaries)

  • License + install required. mike_run loads the local MIKE 1D engine; it fails on a machine without MIKE+ + a valid license. (Reading/plotting do not.)
  • Always run on a COPY. mikeplus has no undo — copy the model folder first and run the copy, never the user's original.
  • Gate on the QA fields, not just ok. ok only means the call returned. Check status: ok / completed_with_warnings are usable; completed_with_errors or incomplete mean the run is untrustworthy — surface errors/warnings and point at log_file for detail, don't bury them. status="unknown" means no engine log was found.
  • Do not fabricate metrics. Pass the returned result_files to mike-results / mike-plot, and the whole return value to mike_manifest_write (mike-audit).

Orchestration

mike_model_info  -> confirm the active simulation / pick one (mike_set_scenario if a GUI scenario is wanted)
mike_run         -> result_files (.res1d) + status
        -> mike-results (summary / flooding / read)  -> mike-plot  -> mike-audit