aieng-cad-cae-copilot

name	aieng-cad-cae-copilot
description	MCP-first evidence workflow for AIENG CAD/CAE setup, solver preflight, approval-gated solver execution, and result extraction. Use when the user asks to inspect CAE readiness, patch setup, prepare/run CalculiX, or report stress/displacement evidence.

aieng-cad-cae-copilot

MCP-first CAE workflow discipline for agents operating the active AIENG Workbench. This skill teaches order of operations and claim discipline; it does not add solver capability.

Purpose

• Use package evidence and MCP tool returns as the source of truth.
• Prepare solver runs without inventing missing material, loads, or constraints.
• Keep cae.run_solver behind the approval boundary.
• Report only evidence-backed results and explicit limitations.
• For repeatable loops, follow the relevant entry in ../engineering_skill_contracts.json: cae-preflight, solver-run-orchestrate, design-target-review, or evidence-report-synthesize.

When to use

Use for CAE readiness inspection, setup patching, solver preflight/deck generation, solver execution, FRD metric extraction, field-region extraction, and refreshed result summaries. Do not use for new CAD authoring or schema/tool implementation.

Required workflow

1. Inspect state with aieng.agent_context { project_id } and, when needed, aieng.inspect_package or aieng.write_completeness_report.
2. If material, loads, or constraints are missing, ask the user or apply explicit setup with cae.apply_setup_patch.
3. Run cae.prepare_solver_run { project_id } before any solver execution. Use the recommended_next_calls list in the response to decide the next cae.* call.
4. If the input deck is missing but setup is sufficient, call cae.generate_solver_input { project_id }.
5. Call cae.run_solver { project_id } only after successful preflight and only through approval. If AIENG_MCP_BLOCK_APPROVAL_TOOLS=1 is active, report the server refusal and stop.
6. After a successful solver run, call cae.extract_solver_results, optionally cae.extract_field_regions, then postprocess.refresh_cae_summary.
7. Re-read context/results before reporting final numbers.

Credibility tiering

Every result-bearing output carries a single credibility stamp (the shared V&V-40 tier). Ordered low → high: critique_finding < surrogate_prediction < proxy_assembly_result < executed_solver_result. Read credibility.tier and report it with the numbers; never present a lower tier as if it were an executed-solver result. An output that claims a solver result without solver_executed: true is downgraded to unverified (rank 0) with a downgrade_reason — treat it as not solver-backed. production_ready is false unless explicitly certified.

Hard rules

• Never claim a solver ran unless cae.run_solver completed successfully and result artifacts exist.
• Surface the credibility tier of any result you report; do not upgrade a proxy/surrogate/critique result to "verified".
• Never claim convergence unless solver evidence supports it.
• Treat stale artifacts as historical until refreshed or rerun.
• FRD scalar extraction is not full field validation.
• AIENG package readiness is not physical certification.
• Respect all [APPROVAL REQUIRED] tools and user decisions.

Response contract

Report: current evidence state, setup changes, preflight result, approval status, solver/result artifacts, extracted metrics, stale evidence, and limitations.