Publishing & registering skills

Publishing & registering skills

Share skills by registering them locally and, optionally, submitting them to a remote hub API.

:::warning No publish command There is no kazma hub publish command. Use register for local install from a path, and submit only when a hub API is configured and reachable. :::

Prerequisites

• A directory with valid skill_manifest.yaml
• Optional: tools.py (native tools shape) or entry-point module
• KAZMA_SECRET if you will sign the skill

Recommended flow (local)

1. Validate

kazma hub validate ./my-skill
# optional JSON:
kazma hub validate ./my-skill --json

2. Sign (integrity)

kazma hub sign ./my-skill
# or: kazma hub sign ./my-skill --secret "$KAZMA_SECRET"

Writes checksum + HMAC into the manifest (see Skills, MCP & Tools).

3. Register locally

kazma hub register ./my-skill

4. Verify

kazma hub search "my-skill"
kazma hub list
kazma hub info <skill-id>

Optional: remote certification submit

If you operate a hub API (hub_url / CLI default):

kazma hub submit ./my-skill --source-url https://github.com/you/my-skill
kazma hub status <submission_id>

There is no --level flag on submit. Local readiness:

kazma hub check-certification ./my-skill

Versioning

Use semantic versioning in the manifest:

• Patch (1.0.1): bug fixes
• Minor (1.1.0): backward-compatible features
• Major (2.0.0): breaking changes

Best practices

1. Keep tools small and HITL-aware (danger tools must match real capabilities).
2. Prefer the native tools: map used by built-in skills under kazma-skills/.
3. Document env vars and permissions in the skill README.
4. Run validate + check-certification before sharing.