Skip to content

anthony-maio/synthesis-skills

Synthesis Skills

Canonical curated skill repository for Synthesis.

This repo is the public review and distribution layer for skill-first self-extension. It keeps the registry model simple:

  1. Skills live as normal directories under skills/.
  2. Metadata is derived from SKILL.md and package structure.
  3. catalog/skills.json is generated for machine search.
  4. STSS scans every changed skill in CI, and optional attestations are tracked in the catalog.
  5. Pull requests are the curation and trust-promotion path.

Repository Layout

synthesis-skills/
├── skills/                  # Curated skill packages
├── catalog/                 # Generated machine-readable search metadata
├── scripts/                 # Validation and catalog generation
├── tests/                   # Registry automation tests
└── .github/                 # CI and contribution templates

Skill Package Rules

  • Each skill lives at skills/<skill-name>/.
  • Each skill must include SKILL.md.
  • Each skill must include PROVENANCE.json.
  • Each skill must include REGISTRY.json.
  • Each skill may include attestation.json from STSS.
  • Optional directories are scripts/, assets/, references/, and agents/.
  • Skill front matter is intentionally small: name and description.
  • description must start with Use when.

Local Workflow

python scripts/validate_repo.py
python scripts/build_catalog.py
pytest -q

To run the STSS gate locally before a PR:

npm install -g stss@0.4.3
python scripts/run_stss_gate.py --base origin/main --head HEAD

To verify that the checked-in catalog is current:

python scripts/build_catalog.py --check

Governance Model

  • Lifecycle stage is tracked separately from trust level.

  • The canonical registry is optimized for one primary skill per capability family.

  • Drafts and challengers may exist, but canonical duplicates should not.

  • Family-level catalog data tracks the current canon plus any challengers under review.

  • UNTRUSTED: local drafts in agent environments, not curated here yet

  • PROBATION: proposed in an open PR or being prepared for review

  • TRUSTED: merged curated skills in this repository

  • VERIFIED: reserved for system or explicitly elevated skills

This repository primarily stores curated TRUSTED skills plus the automation needed to review new submissions.

STSS Tracking

  • CI runs stss scan against every changed skill directory in a PR or push.
  • attestation.json is optional, but when present CI also runs stss verify.
  • catalog/skills.json records whether a skill has an attestation plus its signing and policy metadata.
  • Verification keys can be supplied through STSS_PUBLIC_KEYS_JSON or STSS_PUBLIC_KEY in GitHub Actions secrets.

Provenance vs Attestation

  • PROVENANCE.json records who authored the skill, where it came from, and what license applies.
  • attestation.json records that the exact package snapshot in this registry was scanned or signed by the Synthesis registry process.
  • A registry attestation does not imply original authorship or upstream endorsement.
  • Mirrored and adapted external skills may carry registry attestations as long as their provenance remains explicit.

Submission Policy

  • Today, STSS scanning is required in CI for every changed skill package.
  • Today, signing is optional. If attestation.json is present, CI verifies it.
  • Structural validation, catalog freshness, tests, provenance review, and human curation remain required for merge.
  • Agent-generated skills should be treated as draft proposals until they clear review, even if they include an attestation.
  • REGISTRY.json is the source of truth for lifecycle stage, trust level, and capability family.
  • Challenger submissions must explain how they relate to the current canon and why they should exist.

Submission Path

  1. Open a PR that adds or updates a skill under skills/.
  2. Run validation and regenerate the catalog.
  3. Include PROVENANCE.json, intended host agents, and any helper scripts in the PR.
  4. Let CI and review decide whether the skill should merge.

More detail lives in docs/submitting-skills.md.

Seeding

If you are bootstrapping the first curated set, use docs/seeding.md. The short version is to start with a small founding set of high-signal skills, keep provenance explicit, and treat the first 10 to 20 merged skills as the quality bar for the registry.

About

Canonical curated skill repository and automation for Synthesis

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors