Materials Simulation Skills

Give AI coding agents tested, source-backed expertise in numerical methods and computational materials science instead of guesses.

Stars
★ 94
Last updated
5d ago
License
Apache-2.0
Primary language
Python

At a glance

How it runs
Agent plugin / skillCLI
Works with
Universal · cross-platformCodex · Claude Code
Cost
Free software; you pay for model usage
Setup effort
Low · running in minutes
You'll need
Python 3.10-3.12pipNode.js (npx)Shell / CLINetwork accessLocal filesystem
Typical use
A CFD or heat-transfer researcher wants to know whether a timestep satisfies the CFL condition; the numerical-stability skill returns stability and a recommended maximum dt.
Not a fit if
  • Teams expecting skills to improve scientific answers: the README states this is not yet shown.
  • Users on a niche agent without Agent Skills standard support.
Website
skills.sh ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-15latest release v1.0.0 (2026-11-15)

What does this agent do, and when should you use it?

Materials Simulation Skills is a collection of 24 Agent Skills for computational materials science, spanning core numerical methods, simulation workflows, ontologies, HPC deployment, verification and validation, data management, robustness, and a meta skill-evaluator. Each skill bundles instructions with small, tested scripts that the agent loads only when a question needs it; for example, numerical-stability runs cfl_checker.py to check CFL conditions. Skills follow the open Agent Skills standard and are documented for 20+ tools including Claude Code, Codex, Cursor, Antigravity and GitHub Copilot. A pip package and mss CLI let you list, install, validate and evaluate skills, backed by a deterministic eval gate and CI across Python 3.10–3.12 on Linux, macOS and Windows. The README is explicit that there is no evidence yet that a live agent gives better scientific answers with these skills than without them; a measurement harness exists and the work is tracked in issue #13.

The repository ships skills as folders containing instruction docs plus runnable scripts. When asked a question, the agent selects the matching skill, runs its script (e.g. cfl_checker.py --dx 0.01 --dt 0.001 --velocity 2.0 --json), then explains the checked output. Everything also runs by hand: mss list lists skills, mss run numerical-stability cfl_checker -- --dx 0.01 --dt 0.001 --velocity 2.0 --json executes a script directly, mss bundles lists curated bundles, and mss install --agent claude --bundle verification-and-validation installs a bundle into a named agent. A skill-evaluator meta skill evaluates any skill across any agent CLI, and mss eval runs the deterministic eval gate that asserts documented numbers equal script output.

  1. A CFD or heat-transfer researcher wants to know whether a timestep satisfies the CFL condition; the numerical-stability skill returns stability and a recommended maximum dt.
  2. An engineer running a mesh convergence study wants to know if the mesh is in the asymptotic range; convergence-study checks stress values from h = 0.4, 0.2, 0.1.
  3. An HPC user needs a SLURM job script; the HPC Deployment skills generate sbatch scripts and help diagnose runtime problems.
  4. A materials simulation team preparing a paper or project archive uses the Data Management skill to package FAIR-compliant reproducible data.
  5. A team lead standardizing agent capability installs a curated bundle into Claude Code with mss install --agent claude --bundle ….
  6. A developer building or auditing skills uses skill-evaluator plus mss validate and mss eval to confirm deterministic output.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • npx skills add HeshamFS/materials-simulation-skills
  • pip install materials-simulation-skills

How do you install or deploy this agent?

The primary Agent Skills install is:

npx skills add HeshamFS/materials-simulation-skills

Other documented paths:

gh skill install HeshamFS/materials-simulation-skills --pin v1.0.0

As a Claude Code plugin:

/plugin marketplace add HeshamFS/materials-simulation-skills

Install the mss CLI via pip and add a curated bundle:

pip install materials-simulation-skills
mss bundles
mss install --agent claude --bundle verification-and-validation

Work on the repository itself:

git clone https://github.com/HeshamFS/materials-simulation-skills.git
cd materials-simulation-skills && pip install -e ".[dev]"

How do you use this agent?

After installation, describe the task in your own words and let the agent pick the skill:

Is dt=0.001 stable for advection with v=2.0 m/s and dx=0.01?

You can also name a skill directly:

Use convergence-study to check if my mesh is in the asymptotic range:
h = 0.4, 0.2, 0.1 gave stress = 98.5, 99.6, 99.9 MPa.

Verify script output yourself:

pip install materials-simulation-skills
mss list
mss run numerical-stability cfl_checker -- --dx 0.01 --dt 0.001 --velocity 2.0 --json

Run the eval gate and test suite:

mss eval
python -m pytest tests/

What are this agent's strengths and limitations?

Pros
  • Skills ship small tested scripts, and every documented number is asserted by eval cases (README reports 115 cases, 131 checks, 725 assertions passing in the latest run), so results can be re-checked by hand.
  • Built on the open Agent Skills standard and documented for 20+ tools including Claude Code, Codex, Cursor and GitHub Copilot, reducing single-agent lock-in.
  • 24 skills across 8 categories plus a machine-readable skills_index.json catalog for discovery and integration.
  • Rules cite their sources, such as ASME V&V20, the SchedMD sbatch spec and CMSO IRIs, supporting auditability.
Limitations
  • The README explicitly states it is not yet shown that a live agent gives better scientific answers with these skills; only the measurement harness exists, tracked in issue #13.
  • Requires a Python 3.10–3.12 environment; the npx install path additionally needs Node.js.
  • Coverage is bounded by the scripts, so simulation questions outside those checks still rely on the user's judgment.
  • The agent runtime needs API keys or a paid LLM service, adding per-use model costs.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Materials Simulation Skills This agent 56 · Major gaps Agent plugin / skillFree + model costs ★ 94 5d ago Python Codex · Claude Code
agnix Config Linter 79 · Good CLIFree ★ 491 today Rust Codex · Claude Code
J-Space Cognition Suite 72 · Some gaps Agent plugin / skillFree + model costs ★ 3k 2mo ago Python —
Harness AI Kit 64 · Some gaps CLIFree ★ 26 2mo ago Python Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
56/ 100 5-point scale 2.8 / 5
Trust 13/29
Reliability 8/14
Adaptability 12/18
Convention 12/18
Effectiveness 7/13
Verifiability 4/8
Why each dimension lost points
Trust13 / 29 · 2.2/5

least_privilege=2: docs/SECURITY.md tiers skills by tool surface (HIGH/MEDIUM/LOW) and CI sets read-only GITHUB_TOKEN, a solid privilege reduction. user_confirmation=1: scripts write files, generate SLURM scripts and run commands, yet no SKILL.md requires confirmation or dry-run before execution. data_flow_transparency and sensitive_data_handling=1: SECURITY.md mentions size caps, allow_pickle=False and shlex.quote, but sampled SKILL.md files show no concrete data-flow statement and no handling of sensitive data (HPC credentials, paths). external_effects and rollback=1: mss install has --force overwrite with no backup/rollback, and external side effects of generated scripts are undeclared. dependency_security=2: pip-audit, dependency-review and CodeQL exist, but dependencies are lower-bound only (numpy>=1.21) with no lockfile. source_attribution=2: standards fields cite ASME V&V20, Saad, Lele etc., and README acknowledges the Agent Skills standard.

Reliability8 / 14 · 2.9/5

self_consistency=2: SKILL.md CLI examples and JSON output fields broadly match CI smoke commands, though several sampled SKILL.md files are truncated so full cross-check is impossible. dependency_availability=1: requirements.txt lists only numpy while linear-solvers and others declare SciPy, which is only in the optional 'full' extra, so a plain install may lack dependencies. failure_messages=2: SKILL.md files carry Error Handling tables and the CLI catches OSError/KeyError/ValueError/RuntimeError returning code 2, but coverage is partial.

Adaptability12 / 18 · 3.3/5

audience_and_scenarios=2: README targets materials-science/numerical users with 24 skills and scenario examples. capability_boundaries=2: each SKILL.md has Goal, Requirements and Decision Guidance, though non-applicable cases are not systematically stated. trigger_precision=2: descriptions carry rich trigger phrases ('is my mesh fine enough'), but overlapping descriptions across skills risk mis-triggering. environment_fit=2: docs/INSTALL.md covers Claude Code, Codex, Copilot, Cursor, Antigravity and more, with Python 3.10-3.12 cross-platform CI.

Convention12 / 18 · 3.3/5

information_architecture=2: README, docs/SKILLS.md, skills_index.json and marketplace.json are well structured, though the sampled marketplace.json is truncated. install_notes=3: docs/INSTALL.md thoroughly covers npx, gh skill, pip mss, plugin marketplace and manual copy. naming_stability=2: skill names match directories and metadata has version fields, but per-skill versions differ (1.2.2/1.2.3). examples_and_faq=2: CLI examples and conversational workflows exist but no standalone FAQ. known_limitations=2: README honestly states it is not yet shown that a live agent gives better answers and links VALIDATION_SCOPE. license=3: full Apache-2.0 text consistent with pyproject. versioning_changelog=1: version numbers exist but no CHANGELOG file. maintenance_responsibility=1: publisher is unverified; only marketplace.json names Hesham Salama with no governance or maintenance commitment.

Effectiveness7 / 13 · 2.7/5

output_usability=2: scripts emit structured JSON with fields documented in SKILL.md, directly consumable by an agent. marginal_value=1: README admits it is not yet shown that skills improve agent scientific answers (issue #13), so incremental value over generic numerical libraries is unverified. cost_benefit=2: skills load on demand and scripts are lightweight (mostly stdlib), low install cost, though installing all 24 skills may add context overhead.

Verifiability4 / 8 · 2.5/5

claim_traceability=2: standards fields cite standards and literature item by item and README claims rules cite their source. cross_source_corroboration=1: README claims 115 cases/725 assertions passing, but the sampled packet lacks the docs/evaluations report, so it cannot be cross-checked. fact_inference_separation=2: README clearly separates 'tested' from 'not yet shown' and VALIDATION_SCOPE warns against treating passing checks as physical-simulation evidence.

Risks and how to mitigate them
  • Scripts write files, generate SLURM scripts and run commands, yet no SKILL.md requires user confirmation or dry-run before execution; review generated artifacts before use on real HPC systems.
  • Skills such as linear-solvers declare SciPy, but requirements.txt lists only numpy and SciPy is only in the optional 'full' extra, so a plain pip install may lack dependencies.
  • README admits it is not yet shown that skills improve agent scientific answers (issue #13); incremental value is unverified.
  • Publisher identity is unverified and maintenance responsibility/update path is unclear; dependencies are lower-bound only with no lockfile, limiting supply-chain reproducibility.
  • Several SKILL.md files and marketplace.json are truncated in the sample, so not all 24 skills were reviewed; unseen files are unknown, not missing.
Evidence confidence: Low Reviewed Nov 16, 2026 Reviewed revision a4bd93c4224f
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/materials-simulation-skills.svg)](https://followagents.com/en/agents/materials-simulation-skills)
<a href="https://followagents.com/en/agents/materials-simulation-skills"><img src="https://followagents.com/badge/materials-simulation-skills.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v1.0.0
    Release notes excerpt
    First tagged release. It makes the version-pinned install in the README work: ```bash gh skill install HeshamFS/materials-simulation-skills --pin v1.0.0 npx skills add HeshamFS/materials-simulation-skills ``` What's in it 24 skills across 8 categories: core numerical methods (8), simulation workflow (7), ontology (3), HPC deployment (2), verification & validation (1), data management (1), robustness (1) and the skill-evaluator meta skill. Full catalog: [docs/SKILLS.md](https://github.com/HeshamFS/materials-simulation-skills/blob/v1.0.0/docs/SKILLS.md). Checks at this tag …
    View on GitHub ↗

FAQ

Will these skills make my agent's scientific conclusions more accurate?
The README explicitly says this is not yet shown. The skills provide tested scripts and deterministic outputs that prevent numerical and documentation drift, but they do not guarantee physically correct simulations; see docs/VALIDATION_SCOPE.md for the remaining limits.
Which agent tools are supported?
The skills follow the Agent Skills standard; the README lists Claude Code, Codex, Cursor, Antigravity, GitHub Copilot and 20+ tools. Per-agent steps are in docs/INSTALL.md.
What runtime do the skills need?
Scripts run on Python 3.10–3.12; the npx skills add install path also needs Node.js. CI tests Linux, macOS and Windows.
How do documented numbers stay consistent with the scripts?
Each eval case runs its script and asserts the exact output; mss eval is the deterministic gate, and the latest run is recorded in docs/evaluations/2026-10-20.md.
Is Materials Simulation Skills free?
The software is free, but Materials Simulation Skills calls an AI model, and you pay that model provider for usage (or run a local model).
Is Materials Simulation Skills open source?
Yes. The source code is on GitHub (HeshamFS/materials-simulation-skills) under the Apache-2.0 licence.
Is Materials Simulation Skills safe to use?
FollowAgents' FARS review (Nov 16, 2026) scored Materials Simulation Skills 13/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is Materials Simulation Skills still maintained?
Yes. The latest commit on GitHub is from 2026-11-15, within the last 30 days. The latest release is v1.0.0 (2026-11-15).
Who makes Materials Simulation Skills?
Materials Simulation Skills is published on GitHub by the account HeshamFS. FollowAgents has not verified which person or company is behind that account.
What is Materials Simulation Skills's official website?
The project's GitHub repository lists https://www.skills.sh/heshamfs/materials-simulation-skills as its website. That link is set by the repository owner; the source code itself is on GitHub (HeshamFS/materials-simulation-skills).
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts