smlflg/hai-skill-spine

HAI Skill Spine

hai-skill-spine is the local source of truth for Samuel's cross-harness agent skill system. It inventories and deduplicates upstream-owned skills, exposes one public Router/Operator skill, and deterministically projects adapters for:

• Codex
• Claude Code
• Hermes
• OpenCode
• Pi
• Cursor Agent
• Gemini
• HAI Agent

main is the only hand-edited source branch. Harness branches are generated artifacts and must never be edited directly.

Safety model

• Discovery and routing are read-only.
• Operator commands are read-only unless sync --apply is explicitly invoked.
• Apply requires an exact target and accepted diff.
• Each harness has a durable lock and a recovery journal.
• Secrets, credentials, telemetry, caches, runtime state, raw transcripts, and audio are excluded.
• Unknown or prohibited licenses block remote release.
• Commit, remote creation/push, and live harness activation are separate Samuel gates.

Local development

PYTHONPATH=src python3 -m unittest discover -s tests -v
PYTHONPATH=src python3 -m hai_skill_spine.cli --help

The package uses only Python's standard library at runtime. PyYAML is optional: the checked-in registry is deliberately a JSON-compatible YAML document.

See docs/architecture.md for source-truth boundaries and docs/human-gates.md for the operations intentionally left for Samuel.