canonic-skill

Manage project vision and requirements via Canonic. Human-led design, AI-led implementation.

Features

• init: Bootstrap project. Check Canonic, set workspace, create vision.md.
• Handoff: Edit in Canonic → agent waits → agent resumes with chosen action.
• Markdown is Gospel: Requirements in markdown are absolute truth. AI must follow.
• Anti-Pattern Guard: AI modifying docs is discouraged. Prefer human-in-the-loop via Canonic.

Usage

1. Initialize Project

/canonic init

• Checks Canonic install.
• Asks for workspace path (outside code repo recommended).
• Synthesizes vision.md from context or user input.
• Opens in Canonic for review.

2. Review & Handoff

/canonic review path/to/doc.md

• Opens doc in Canonic.
• Agent waits.
• User edits → clicks "Implement this" (or other action).
• Agent resumes with updated content.

Philosophy

• Human Design: Use Canonic to define what to build.
• AI Implement: Agent reads reviewed markdown as source of truth (Gospel).
• No Desync: If requirements change in Canonic, agent updates internal state.
• Anti-Pattern: AI editing docs = bad. Human review = good.

Install

Gemini CLI:

gemini extensions install https://github.com/Canonical-AI/canonic-skill --auto-update

Claude Code:

npx skills add Canonical-AI/canonic-skill --local

Other:

See INSTALL.md for more options.

Requirements

• Canonic (macOS, Windows, or Linux).
• Agent with shell + HTTP capability.

How it works

Canonic runs a local HTTP server and writes a discovery lockfile at ~/.canonic/api.lock. The skill reads the lockfile, connects to the server, opens your file for review, and waits for you to send it back via the action picker in the app.

See the Canonic agent API docs for the full protocol.