knowledge-activation

Reference implementation of the Knowledge Activation thesis as a Claude Code plugin.

Gives agents institutional memory through a self-improving skill ecosystem: every missing skill becomes a gap report, every gap report becomes a contribution opportunity, every merged skill ships to all operators on the next plugin update.

Core principle: the agent is the only orchestrator. No daemon, no server, no queue.

Install

# User scope (default) — available across all your projects
claude plugins install knowledge-activation

# Project scope — scoped to this repository
claude plugins install knowledge-activation --scope project

Bootstrap

After installing, run once to wire the plugin into your project:

invoke bootstrapping-agent-directives

This writes activation directives into your CLAUDE.md and registers two hooks in settings.json.

Structure

• skills/meta/ — 6 meta-skills + 2 reference stubs (8 files total). The plugin machinery.
• skills/org/ — Governance team skills (infra, security, compliance). Fill with your org's standards.
• skills/teams/ — Team-owned subfolders. Teams are fully autonomous here.
• hooks/ — Two shell scripts: version check (PreToolUse) and gap log flush (Stop).
• validators/ — 24-criterion quality gate for all skills.
• templates/ — Blank skill, gap report, and scorecard templates.

Contributing

See CONTRIBUTING.md. Gap reports filed automatically — every session that hits a missing skill feeds the flywheel.

Private Forks

See PRIVATE-FORKS.md for enterprise deployments with internal-only content.