← ClaudeAtlas

coralswarmlisted

Search CoralSwarm knowledge, resume prior work, and save notes or conversation checkpoints. Use when asked what an ocean knows, to save context to CoralSwarm, or to capture a coding session or bot run. Covers session identity, retries, provenance, and confirming the save.
CoralSwarm/coralswarm-connect · ★ 0 · AI & Automation · score 70
Install: claude install-skill CoralSwarm/coralswarm-connect
# Using CoralSwarm CoralSwarm is a knowledge graph of your work. This skill is about *using* it well. It assumes the MCP connection already exists — see `coralswarm-connect` to set one up. ## Vocabulary - **Ocean** — the graph itself. You may have several: a personal one, and one per org. - **Coral** — one captured item. A note, a chat, a meeting, a coding-session checkpoint. - **Reef** — something the graph derived: an `entity`, `decision`, `thread` or `reference`. The server's own `instructions` state this too. This skill does not restate the standing habits it carries; it covers the parts that need more than one line. ## Ocean selection Every read and write tool takes an optional `ocean_id`. **Omit it.** It defaults to your primary ocean, which in an org context is your member-private ocean rather than the org commons — the commons is the id most often picked by hand, and it is the one id that can never accept a write. `list_oceans` shows what you can reach, and `writable: true` marks what you can write. Call it once and reuse the answer; do not call it before every save. The exception is `list_sessions`, where `ocean_id` is **required**. That is deliberate: it refuses to guess, because listing the wrong ocean's sessions is a silent error. ## Finding what the ocean knows Query only this user's captured work — notes, sessions, meetings, decisions, this project's internals. Skip general knowledge, news, and market research. Mentioning the product in a public-market