Portable AI skill packages for manufacturing analysis, documentation and qualified review. The library combines sector-neutral methods with Canadian jurisdiction overlays and professional role workflows.
Use it to structure production planning, quality analysis, maintenance reviews and evidence handoffs. Each skill defines its required inputs, method, outputs and review boundaries.
Development status: This is a public development library. A stable v1 release is deferred pending independent model evaluation and qualified source review. See the release-candidate audit for the full assessment.
- 159 distinct skills across 20 families, covering production, quality, reliability, safety and other manufacturing functions.
- 18 professional skillsets with 38 workflows that compose relevant skills for a role.
- Canadian jurisdiction overlays kept separate from generic manufacturing methods.
- Repository validators and synthetic examples for checking contracts, calculations and expected routing.
Start with the Wiki, browse the skill catalogue, or choose a professional skillset.
Sector specializations are planned; no sector requirement packages are implemented.
You need Git and Python. Validation runs in CI with Python 3.14 on Linux and Windows and uses only the Python standard library.
git clone https://github.com/jeremylongworth-source/AgentManufacturing.git
cd AgentManufacturing
python scripts/validate-all.pyTo find the skills for a production-planning workflow:
python scripts/resolve-skillset.py production-planner horizon-planThe resolver returns REFERENCES_RESOLVED with paths to the relevant SKILL.md files. Read those instructions and their references, then gather the required evidence before applying the methods in your chosen agent host.
The resolver lists references; it does not execute skills or assess the supplied evidence. This workflow provides no automatic agent installer, and host compatibility must be verified separately.
Continue with the worked example or troubleshooting guide.
These skills support analysis and qualified handoffs. They do not authorize machine operation, isolation or restart, safety-circuit changes, engineering signoff, product release or legal conclusions. Do not use model output to bypass safeguards or fabricate manufacturing records.
Establish current sources, jurisdiction, units, populations and review authority for each conclusion. Generic calculations and successful reference resolution do not establish sector or legal applicability. Read the safety boundaries and source rules.
The repository includes 243 expected routing cases and assisted, self-reviewed evaluations. These are not independent measurements of model reliability. Baseline comparisons, repeated model trials and multi-turn robustness tests remain unperformed. The validation guide explains what each check establishes.
| Location | Purpose |
|---|---|
| skills/ | Atomic instructions, references and host metadata |
| skillsets/index.json | Professional roles and workflow manifests |
| composition contract | Canonical skill paths and evidence-reuse rules |
| specializations/registry.json | Planned sector coverage |
| domain contract | Architecture and jurisdiction separation |
| tests/ | Expected scenarios, synthetic fixtures and evaluation evidence |
| docs/wiki/ | Maintained Wiki source |
There are 161 package directories for 159 distinct skills because two historical duplicates are preserved. Use the canonical paths in the composition contract. Some YAML files intentionally use JSON syntax for dependency-free parsing.
Read CONTRIBUTING.md for proposing changes, validation and evidence requirements. Agent-assisted contributors should also read AGENTS.md.
For sensitive vulnerabilities, follow SECURITY.md. For community concerns, follow CODE_OF_CONDUCT.md. Do not include sensitive reports, credentials or confidential plant records in public issues or pull requests. No support response time is guaranteed.
The initial development roadmap is complete for its documented scope. Remaining work includes independent evaluation, qualified source and standards review, merge-gate policy, and host and support scope. See the roadmap, audit follow-ups and changelog.
AgentManufacturing is licensed under MIT. Third-party references retain their own rights; citation does not grant permission to reproduce protected standards or other source material.
You can support the maintainer through GitHub Sponsors.