Using it from an AI agent
Claude Code skills, rules files, and whether to use the CLI or MCP.
comitia's commands are built for AI to use, not humans. The output reads like clean prose, so an AI can understand it directly.
Teaching your AI about comitia — rules files
comitia init -g writes a short note into your AI's rules file.
| Tool | File |
|---|---|
| Claude Code | ~/.claude/CLAUDE.md |
| Codex | ~/.codex/AGENTS.md |
| Gemini CLI | ~/.gemini/GEMINI.md |
The block lives between <!-- comitia:start --> and <!-- comitia:end -->, and running it again replaces it in place — you never end up with two copies. Run it without -g to write only to the current folder's CLAUDE.md/AGENTS.md.
Claude Code skills
| Skill | What it does |
|---|---|
/comitia:read | Look up and answer from Slack, meeting notes, and the wiki. Read only |
/comitia:ingest | Add a source, find related documents, and leave a revised summary |
/comitia:slack | Post to Slack — through the Comitia bot. Always shows you the channel and text before posting, and asks |
/comitia:sync | Pull Slack/meeting notes right now instead of waiting for the hourly sync. The "sync status" line at the top of list/search shows how fresh things are |
These also trigger from plain language — "what did we say on Slack," "add this meeting note to the wiki," "organize this and share it on Slack." Skills only load their body when called, so the standing cost is just one description line.
When a new version ships:
npm i -g comitia && comitia skills
Rules live on the server — comitia rules
How the wiki should be shaped (axes, document format, what not to touch) differs by company, so it lives on the server, not in the package. Your AI reads comitia rules before writing anything. To change it, create or edit _rules.md in the Comitia wiki — from then on, that's the source of truth.
CLI vs. MCP
| CLI | MCP | |
|---|---|---|
| Setup | One comitia init -g | A config file per app |
| Token cost per conversation | Only when called, tens of tokens | Tool descriptions are always loaded, thousands |
| Where | Any AI with a terminal | Apps that support MCP |
Both talk to the same server. If your AI has a terminal, use the CLI. For apps without one (like a plain Claude desktop conversation), use MCP — issue a token from the connection tokens screen and pass it as Authorization: Bearer <token>.
Things that work well to say to your AI
- "Summarize the recent Slack discussion about Comitia" — starts from raw sources (
recent/raw) - "Where do things stand with the Acme project?" — starts from the wiki (
read) - "Upload this meeting note and organize it" — follows the
/comitia:ingestflow - "Summarize what we need to discuss now and share it in #project_0to1" —
/comitia:slackdrafts it and posts once you approve (requires sources & organize permission, and the bot must already be in that channel) - Ask for sources (source ids, document paths) and it'll attach them