topoteretes Cognee Code Graph
Use this skill when the user asks Antigravity to build a Cognee memory of a repository, index code, or answer structural code questions (who calls X, what breaks if X changes, how does A reach B) from memory.
Requires a Cognee server >= 1.5.3. Indexing runs the enola code-graph pipeline server-side and makes no LLM or embedding calls — it is fast, deterministic, and token-free.
Rules
- Server-first: the scripts below POST to the running Cognee server. There is deliberately no CLI fallback for code indexing/search — an unreachable server means "not done", never "index some other way".
- One narrow dataset per repository (default
codebase-<repo-name>-<digest>); do not mix repositories into one dataset. The digest is the indexed path, so two checkouts sharing a basename stay in separate graphs.--codesearches resolve the name from the current checkout, so it rarely needs to be typed. - Never index directories holding secrets you would not ingest as documents;
the pipeline skips
.env/dotfiles/binaries itself, but a git URL to a private repo still lands its code in the target dataset.
Index a repository
"${COGNEE_ANTIGRAVITY_PLUGIN_ROOT:-$HOME/.gemini/config/plugins/cognee}"/scripts/cognee-index-repo.sh <repo-path-or-git-url> [--dataset <name>] [--index-vectors] [--wait <seconds>]
- Local path (e.g.
.): works when the Cognee server shares this filesystem — the default local plugin server does. Cloud servers reject local paths; pass a git URL instead. - Git URL: the server shallow-clones it; freshness follows pushed commits (local edits are invisible to it).
--index-vectors: also embeds the code facts so semantic/hybrid search can see them (needs an embedding provider). Without it, code knowledge is reachable ONLY through code search below.- The submit returns quickly (background pipeline);
--wait 60pollspipeline=code_graph_pipelineuntil queryable.
Indexing writes enola's snapshot to <repo>/.enola/ (untracked) — add it to
.gitignore or global excludes. The plugin's change detection ignores it.
Indexing is also the freshness opt-in: for local-path repos the plugin records a git fingerprint and automatically re-submits the repo in the background when a turn changed the working tree. Re-running the command on an unchanged repo is skipped server-side (content hashes), so it is always safe.
Automatic indexing
Opening the agent inside a git repository indexes it automatically in the background at session start — no setup step. What happens depends on the server, because indexing a repo means the code has to be readable by it:
| Situation | Behavior |
|---|---|
| Local server (default), new repo | Indexed automatically — the server reads the working tree in place, so the code never leaves the machine |
| Cloud/remote server, new repo | Not indexed automatically; run the index command with a git URL, or set COGNEE_CODE_AUTOINDEX=always |
| Already-indexed repo | Always refreshed if the tree changed since the last index — regardless of the setting above |
| Not a git repo / no code files / very large repo (>3000 source files) | Skipped; index it explicitly if you want it |
Set COGNEE_CODE_AUTOINDEX=off to disable automatic indexing of new repos
(explicitly indexed repos keep refreshing). Explicit indexing has no size cap.
What the graph reflects (freshness)
The graph tracks different things depending on where the Cognee server runs. Both are normal, expected behavior — but the results look identical, so know which one applies:
| Server | Graph reflects | Stays current via |
|---|---|---|
| Local (default) | The working tree, uncommitted and untracked changes included | Automatic re-index after any turn that changed a file |
| Cloud / remote | The last pushed commit — the server clones the repo and cannot read this machine's disk | Pushing, then re-indexing |
On a cloud server, local edits are invisible to the graph until they are pushed. Nothing
errors and nothing looks unusual: a question about a symbol you just renamed locally is
answered from the pushed state, confidently. So when the answer matters and the work is
in progress, either push first or say so in the answer. {"operation": "delta"} shows
what the last index actually changed and is the fastest way to check what the graph
currently knows.
Local-path indexing therefore requires a server that shares this filesystem; cloud servers reject local paths and need a git URL.
Query the code graph
"${COGNEE_ANTIGRAVITY_PLUGIN_ROOT:-$HOME/.gemini/config/plugins/cognee}"/scripts/cognee-search.sh "<seed>" 10 --code [--code-query '<json>']
Without --code-query, the query text is a seed name (exact/suffix/substring
match). With it, pick one exact operation:
| Operation | Answers | Example |
|---|---|---|
query_facts |
filtered listing | {"operation":"query_facts","kind":"route","limit":50} — all API endpoints |
explore |
neighborhood of one node | {"operation":"explore","name":"UserService","max_depth":1} |
traverse |
follow edges from seeds | {"operation":"traverse","start":"main","direction":"forward","max_depth":3} |
find_path |
how A reaches B | {"operation":"find_path","source":"AuthMiddleware","target":"Database"} |
impact_analysis |
what breaks if X changes | {"operation":"impact_analysis","targets":["process_payment"]} |
delta |
what the last index changed | {"operation":"delta"} |
An ambiguous seed returns an error listing the candidates — retry with an
exact id or a repo filter. A seed that doesn't resolve returns empty (not
an error): the graph has no such symbol.
When to use which search
- Structural question naming a symbol/file →
--code. Exact, instant, no tokens. - Conceptual question naming nothing ("how does auth work here?") →
regular
cognee-search.sh(hybrid/graph). Graph-only code is invisible there unless indexed with--index-vectorsor the repo's docs were ingested as documents. - Chain them: hybrid discovers the name,
--codegives the exact structure around it. - Verify freshness: after edits,
{"operation":"delta"}shows what the last re-index added/updated/removed.
Treat results as a map, not ground truth — verify important claims against the actual files before editing code.
Per-file ingestion (no repo index)
To store a single code file in normal memory under its real filename (routes down the zero-LLM code path server-side, no cross-file edges):
"${COGNEE_ANTIGRAVITY_PLUGIN_ROOT:-$HOME/.gemini/config/plugins/cognee}"/scripts/cognee-remember.sh --file src/payments.py --node-set project_docs
Prefer the repo index above when callers/imports across files matter.
Automatic recall
Once a repo is indexed from its checkout, the per-prompt memory recall adds a
code lane automatically when the prompt mentions an identifier-shaped token —
the facts appear in the injected context as === Code graph facts ===.