ai-cae-run-manager

name	ai-cae-run-manager
description	Evidence-first run management for AI-assisted CAE and engineering automation. Use when Codex needs to plan, execute, audit, or report a simulation/CAD automation task with AI CAE Automation Lab, MCP tools, run folders, evidence scans, credibility grading, or report generation.

AI CAE Run Manager

Use this skill to make every engineering automation task traceable. Treat a task as incomplete until it has a run folder, input source, execution log, exported evidence, and a report.

Core Workflow

1. Check the environment before modeling:

ai-cae-toolbox env-check --json

If the command is not installed, run from the cloned repository:

python -m ai_cae_lab.toolbox env-check --json

2. Ask the toolbox for a bridge plan when a solver adapter is involved:

ai-cae-toolbox bridge-plan --solver <solver> --objective "<objective>"

3. Create a run directory before generating solver files:

ai-cae-toolbox create-run --solver <solver> --case <case-name> --objective "<objective>"

Use the generated run folder structure:

runs/<run-id>/
  run.json
  run.log
  inputs/
  scripts/
  outputs/
  exports/
  logs/

4. Generate a tiny smoke template when available:

ai-cae-toolbox write-smoke-template --solver <solver> --run-dir <run-dir> --case <case-name>

5. Put source artifacts in the run folder:

• modeling scripts and journals in scripts/
• input decks, geometry placeholders, and small public-safe examples in inputs/
• solver logs in logs/
• exported plots, CSV/JSON tables, reports, and screenshots in exports/
• native solver outputs in outputs/, while keeping large/private files out of git

6. After execution, scan evidence:

ai-cae-toolbox scan-evidence <run-dir>

7. Generate the report:

ai-cae-toolbox generate-report <run-dir>

MCP Tool Usage

If the ai-cae-automation-lab MCP server is available, prefer structured tools:

• env_check
• create_run_record
• solver_bridge_plan
• write_solver_smoke_template
• scan_run_evidence
• generate_run_report
• abaqus_run_no_gui_script
• abaqus_submit_input_deck
• fluent_run_journal_file
• ansys_run_workbench_journal_file
• comsol_compile_java_file
• comsol_run_compiled_java_class
• comsol_run_batch_file
• matlab_run_script_file
• openfoam_run_case_command
• paraview_run_pvpython_script
• pcschematic_check_installation

Use CLI commands when MCP is unavailable.

Validation Rules

• Never claim solver success without logs and exported evidence.
• Mark dry runs, failed runs, and visual previews honestly.
• Record exact commands, executable paths, and error messages in run.log.
• Do not publish private paths, licensed manuals, raw customer models, credentials, or large generated solver outputs.
• If a commercial solver is missing or unlicensed, produce a reproducible setup package and stop before claiming results.