modernrelay/omnigraph-studio/skill

Omnigraph in Studio

You work on one Omnigraph graph with a person in Omnigraph Studio. AGENTS.md in your working directory holds the rules for this graph; this skill teaches the CLI and the GQ language. If another omnigraph skill is installed, prefer this one here: its advice fits this conversation's CLI.

What is different in Studio

• Run the CLI by the absolute path the omnigraph_cli tool gives you, with no --target or

--graph: its config makes Studio's target and this graph the default. Or run it through the omnigraph_run tool, which adds both. Never another omnigraph on PATH, and never another --target, a storage URI or --as.

• The CLI is already set up. Skip the guide's parts on installing, ~/.omnigraph/config.yaml,

credentials, targets, clusters, deployments, aliases, upgrades, migrations and server policy: those are the person's.

• Read any branch. Write only to your own branch: create it first, then pass --branch to

mutate, load and Blob writes. The person reviews and merges it in Studio.

• No schema changes: schema plan and schema apply are not available. Propose a schema change

to the person instead.

• Branch statements (branch create|delete|merge inside a query) are refused; use the

branch commands.

• An outcome of effects: unknown means the answer was lost. Check

commit list --branch <your branch> before trying again; never repeat a write blindly. A Studio refusal says "Nothing was sent": nothing was written.

• Check a query before running it: save the schema with schema show, then

lint --query <file> --schema <file>. Lint works offline, so it also runs where your shell has no network.

Where to look

• GUIDE.md, the Omnigraph guide: "Essentials" (reads, mutations, loads), ".gq grammar",

"CLI Reference" and "Common Gotchas".

• references/queries.md: parameters, matching, traversal, correlated blocks, filters,

aggregation, search functions, mutations and branch statements.

• references/data.md: choosing mutate or load, NDJSON loads and their --mode.
• references/schema.md: reading a .pg schema; references/schema-design.md: why a schema

is shaped as it is, for proposing a change to the person.

• references/search.md: vector, full-text and hybrid search.
• references/blobs.md, references/changes.md, references/stored-queries.md.
• references/remote-ops.md: write receipts, unknown outcomes and reading large output.
• references/commands.md: every command and flag.

Not for Studio: references/cluster.md, references/migrations.md, references/server-policy.md and references/aliases.md.