Claude Skill

agentfem

Build, review, run, validate, migrate, or extend AgentFEM finite-element projects. Use for AgentFEM studies, meshes, materials, constraints, loads, solution steps, results, campaigns, scientific datasets, surrogate/PINN/neural-operator integration, verification, public API extens

LLM Mart · 0 points · 1 views 0 listing impressions 0 install-command copies
Download haoming-luo-agentfem-skills_agentfem-fc86db6.zip · 27 KB

Install

skills CLI npx skills add https://github.com/haoming-luo/agentfem/tree/main/skills/agentfem
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install haoming-luo-agentfem@llmmart
Git git clone https://github.com/haoming-luo/agentfem.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole haoming-luo/agentfem collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

AgentFEM

People and agents use the same readable case.py and installed CLI. A prepared project runs without AI. Choose the shortest path that satisfies the request.

Start with the user's task

  • Install: read installation.onboarding at https://haoming-luo.github.io/agentfem/agentfem.json. Reuse a working environment and the user's chosen channel. Finish with installed identity, doctor readiness and WSL workspace protection. Report version and environment location, then offer an optional example. Create or run one only when requested; an existing request to install and try a model is sufficient.
  • Run or edit a project: read its instructions, case.py and agentfem.toml; use check → run → inspect. Model edits follow the workflow reference below. check already includes migration findings.
  • View results: use agentfem show latest --json and the recorded artifacts. Do not run a new simulation just to answer a result question.
  • Diagnose or upgrade: inspect the specific failure or migration findings. Use doctor --json after environment/version changes or runtime failures.
  • Extend: read the relevant ownership, validation and extension references before implementing numerical functionality.

Reuse context already checked within the task. Read one relevant capability topic before requesting the full inventory. The installed API/options are authoritative when online documentation describes a newer version. For an uncertain keyword, query agentfem capabilities --api MODULE.CALLABLE --json; for a concrete Step use models.step_capability(...). Missing query support in an older install calls for its installed help, not an automatic upgrade.

Reference Routing

References are relative to this skill directory. Keep the whole directory when installing the skill; read only what the current task needs.

  1. references/workflow.md: constructing or changing a model.
  2. references/concepts.md: selecting physical objects or maturity levels.
  3. references/scientific_workflows.md: material history, contact, thermal transfer, Abaqus migration, restart, campaigns and specialized output. Search for the current topic; this is not an installation checklist.
  4. references/module_map.md: locating code and ownership.
  5. references/validation.md: changing scientific code or evaluating a numerical claim.
  6. references/extension_rules.md: adding a public helper, provider, material family or extension boundary.

Common decisions

  • Prefer Study → Model → materials/boundaries → model.step → Result. Keep physics in case.py and operational settings in agentfem.toml.
  • Use public APIs and provider option contracts; keep lower-level operators and FEniCSx available when the task needs them. Reuse framework state, rollback, output and restart instead of writing a project-owned solver.
  • Publish through Result and use its artifact paths; avoid case-owned connectivity or XDMF rewrites. verify checks provenance, while physical acceptance needs relevant reference, balance and convergence evidence.
  • On WSL, protect project storage once with agentfem workspace --protect. Recheck custody before replacing/removing a runtime, not before every solve.
  • If permissions or a private password entry are needed, give one concrete handoff. Resume from the completed stage instead of repeating installation.
  • Offer one relevant next step after completing a task. Suggestions are not permission to run extra benchmarks, tune a model, install integrations or publish data. Keep support/account actions voluntary and explicitly requested.
Files (agentfem)
  • agents
    • openai.yaml 296 B
      Loading…
  • references
    • concepts.md 8.4 KB
      Loading…
    • extension_rules.md 1.5 KB
      Loading…
    • module_map.md 5.5 KB
      Loading…
    • scientific_workflows.md 24.8 KB
      Loading…
    • validation.md 6.9 KB
      Loading…
    • workflow.md 8.7 KB
      Loading…
  • SKILL.md 3.8 KB
    ---
    name: agentfem
    description: Install, build, run, inspect or extend AgentFEM finite-element projects using the public workflow and declared capabilities. Use for AgentFEM modeling, results, migration and solver extensions.
    ---
    
    # AgentFEM
    
    People and agents use the same readable `case.py` and installed CLI. A prepared
    project runs without AI. Choose the shortest path that satisfies the request.
    
    ## Start with the user's task
    
    - **Install:** read `installation.onboarding` at
      https://haoming-luo.github.io/agentfem/agentfem.json. Reuse a working environment
      and the user's chosen channel. Finish with installed identity, doctor readiness
      and WSL workspace protection. Report version and environment location, then
      offer an optional example. Create or run one only when requested; an existing
      request to install and try a model is sufficient.
    - **Run or edit a project:** read its instructions, `case.py` and
      `agentfem.toml`; use `check → run → inspect`. Model edits follow the workflow
      reference below. `check` already includes migration findings.
    - **View results:** use `agentfem show latest --json` and the recorded artifacts.
      Do not run a new simulation just to answer a result question.
    - **Diagnose or upgrade:** inspect the specific failure or migration findings.
      Use `doctor --json` after environment/version changes or runtime failures.
    - **Extend:** read the relevant ownership, validation and extension references
      before implementing numerical functionality.
    
    Reuse context already checked within the task. Read one relevant capability
    topic before requesting the full inventory. The installed API/options are
    authoritative when online documentation describes a newer version.
    For an uncertain keyword, query `agentfem capabilities --api MODULE.CALLABLE --json`;
    for a concrete Step use `models.step_capability(...)`. Missing query
    support in an older install calls for its installed help, not an automatic upgrade.
    
    ## Reference Routing
    
    References are relative to this skill directory. Keep the whole directory when
    installing the skill; read only what the current task needs.
    
    1. `references/workflow.md`: constructing or changing a model.
    2. `references/concepts.md`: selecting physical objects or maturity levels.
    3. `references/scientific_workflows.md`: material history, contact, thermal
       transfer, Abaqus migration, restart, campaigns and specialized output.
       Search for the current topic; this is not an installation checklist.
    4. `references/module_map.md`: locating code and ownership.
    5. `references/validation.md`: changing scientific code or evaluating a
       numerical claim.
    6. `references/extension_rules.md`: adding a public helper, provider, material
       family or extension boundary.
    
    ## Common decisions
    
    - Prefer Study → Model → materials/boundaries → `model.step` → Result.
      Keep physics in `case.py` and operational settings in `agentfem.toml`.
    - Use public APIs and provider option contracts; keep lower-level operators and
      FEniCSx available when the task needs them. Reuse framework state, rollback,
      output and restart instead of writing a project-owned solver.
    - Publish through Result and use its artifact paths; avoid case-owned
      connectivity or XDMF rewrites. `verify` checks provenance, while physical
      acceptance needs relevant reference, balance and convergence evidence.
    - On WSL, protect project storage once with `agentfem workspace --protect`.
      Recheck custody before replacing/removing a runtime, not before every solve.
    - If permissions or a private password entry are needed, give one concrete
      handoff. Resume from the completed stage instead of repeating installation.
    - Offer one relevant next step after completing a task. Suggestions are not
      permission to run extra benchmarks, tune a model, install integrations or
      publish data. Keep support/account actions voluntary and explicitly requested.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.