Skills-Registry Skill

Goal

Interoperate with the universal agentskills.io SKILL.md ecosystem — both publishing GABBE's skills out to registries and drawing vetted external skills in. Because GABBE's 171+ skills already conform to the open SKILL.md standard, this is mostly packaging + a validate/import path, not new authoring. Every import is treated as untrusted supply-chain input and gated by GABBE_AUTONOMY + budget. This is the bidirectional bridge update-scan.skill uses when it scans registries for "new/better skills."

The optional gabbe CLI (gabbe registry publish|add) and scripts/registry_export.py / scripts/registry_import.py automate this, but the steps below also work by hand (Python-independent): the agent can fetch a SKILL.md, validate it, and place it itself.

Registries

Configured in project/gabbe.config.json → registries. Common targets:

• Agent Garden / github.com/google/skills — Google's official skills repo.
• skills.sh (Vercel), Agensi (curated, security-scanned), agent-skills-hub, VoltAgent/awesome-agent-skills, agentskill.sh.

Publish (export GABBE skills out)

1. Emit the agentskills.io-standard <slug>/SKILL.md tree (scripts/registry_export.py, or by hand from agents/skills/).
2. Produce a manifest.json (name/version/per-skill sha256) and an optional A2A-style agent-card.json (skills-as-capabilities for federated discovery — ties to coordination/agent-interop.skill).
3. The maintainer performs the actual publish (holds the registry/repo credentials). GABBE produces the publish-ready artifact + a CI job; never auto-publishes.