mike-end-to-end

Standard operating procedure for an auditable, headless MIKE+ workflow via the mike-plus MCP server — how to plan, copy, run, QA-gate, compare, plot, audit, and safely stop. Use FIRST whenever an agen

SKILL.md

GitHub

MIKE+ End-to-End (SOP)

The other mike-* skills describe one tool group each. This one says how they fit together, what must be true before a number or figure is reported, and when to stop.

1. Pick an operating mode (state it in the first reply)

Mode When Chain License
A. Read / plot only a .res1d already exists, or no MIKE+ install mike-results -> mike-plot -> mike-audit none
B. Run an existing model a .sqlite and a goal ("run the 1-year storm") copy -> mike-model -> (mike_set_scenario) -> mike-runner -> mike-results -> mike-plot -> mike-audit yes
C. What-if / compare "what if the pipe were 0.5 m", "before vs after" baseline copy + variant copy -> mike-params (variant only) -> mike-runner x2 -> mike-compare -> mike-audit yes
D. Import from SWMM an EPA SWMM .inp (SWMMCanada, agentic-swmm) mike-import -> then B or C yes
E. New rainfall a CSV storm / gauge record to drive the model mike-rain -> point the model's rainfall boundary at the .dfs0 (mike-params) -> B rain: none, run: yes

If the goal only needs Mode A, do not touch mikeplus at all: it is faster and needs no license.

2. Standard run directory (create it, use it everywhere)

runs/<case>/
  00_model/        COPY of the whole model folder (.sqlite, .mupp, .dfs0, hotstart .res1d ...)
  01_inputs/       rain CSV/.dfs0 you produced, notes
  02_results/      the engine writes next to the model: 00_model/<name>_m1d - Result Files/*.res1d
  03_plots/        every PNG
  manifest.json    mike_manifest_write, written last

For Mode C use 00_model_baseline/ and 00_model_<variant>/: two independent copies.

3. Hard rules

  1. Copy first, always the whole folder. mikeplus has no undo and the MIKE 1D engine writes result files (and a .log) next to the model; external files (rain .dfs0, hotstart, .mupp) are referenced relatively, so copying only the .sqlite breaks the run.
  2. Gate on status, not ok. mike_run returns ok:true whenever the call returned. Usable = status in {ok, completed_with_warnings}; completed_with_errors / incomplete / unknown = do not use the results, surface errors, warnings, log_file.
  3. Discover ids before using them. mike_model_info (setups, scenarios), mike_results_list (quantities, element ids), mike_get_values (current values). Never invent an element id, table or column.
  4. Warm-up. Global peaks can sit at t≈0 (initial condition). Pass skip_hours to mike_results_summary / mike_results_compare / mike_results_flooding / mike_plot_profile and always report peak_time.
  5. Variants are copies, not scenarios. mike_set_scenario only activates scenarios that already exist in the model (prepared in the GUI). Table edits made "inside" a scenario are not isolated (see mike-compare), so a what-if edit goes on a separate model copy.
  6. License boundary. run / edit / scenario / import need MIKE+ + a checked-out license; results / plots / compare / rain / manifest do not. On a license failure ("license not available", demo mode, an edit that "did not persist", an import that "exceeds the demo limit") stop that branch and report it; do not retry in a loop, the seat is rate-limited and re-tries make it worse.
  7. Evidence only. Report a number only from a tool return; report a figure only when the tool returned ok:true and png; say "not run" for anything skipped.

4. Stop rules (stop the branch, report, ask)

  • mike_run status is incomplete / completed_with_errors / unknown.
  • Any license or "did not persist" error.
  • The requested element / quantity / table / column does not exist.
  • mike_results_compare returns aligned: interpolated and the two runs were meant to share a time base (different setups were compared).
  • The user's model would have to be edited in place (no copy possible).

5. QA gates before the final answer

Claim Evidence required
"the run completed" status in {ok, completed_with_warnings} and result_files non-empty
"peak X at t" from mike_results_summary/read/compare, with skip_hours stated
"N nodes flood" mike_results_flooding (outlets excluded by default; say so)
"the change increased/decreased ..." mike_results_compare deltas, signed, with units and %
"see figure" the PNG path returned by a mike_plot_* tool
"reproducible" manifest.json written, missing_files: []

6. Report template (one screen)

  1. Mode + what ran: model copy, simulation setup, scenario, elapsed.
  2. QA: run status, warnings count, anything skipped and why (license, missing data).
  3. Numbers: peaks with units + times, flooded nodes, deltas (for C).
  4. Figures: paths.
  5. Provenance: manifest path.
  6. Not done / next step.

7. Skill map

Skill Tools License
mike-model mike_model_info, mike_set_scenario yes
mike-params mike_get_values, mike_set_values yes
mike-runner mike_run yes
mike-import mike_import_swmm yes
mike-results mike_results_list / summary / read / flooding no
mike-compare mike_results_compare, mike_plot_compare no
mike-plot mike_plot_rain_flow / timeseries / network / compare / profile no
mike-rain mike_rain_to_dfs0 no
mike-audit mike_manifest_write no