codebase - SKILL.md Agent Skill

name: codebase description: Use when ingesting, cognifying, or querying a codebase with Cognee CLI from Codex.

Cognee CLI Codebase Workflows

Use this skill when the user asks Codex to build a Cognee memory of a repository, index code, query implementation details, or create a code-aware knowledge graph.

Rules

Use uv run cognee-cli ...; do not use MCP.
Start by defining the codebase scope and dataset name.
Never ingest .env, private keys, credentials, local database files, virtualenvs, dependency caches, or generated build output.
Prefer focused ingestion over sending an entire large repository at once.
Use rg --files first to inspect candidate paths.

Scope

Create a dataset name that is stable and readable, such as:

codebase-cognee
codebase-frontend
codebase-api

Inspect candidate files:

rg --files

Common exclusions include:

.git/
.venv/
node_modules/
dist/
build/
.next/
coverage/
.env
*.sqlite
*.db
*.key
*.pem

Ingest

For a focused set of source paths:

uv run cognee-cli add <path-1> <path-2> <path-3> -d <dataset-name>
uv run cognee-cli cognify -d <dataset-name> --background

For small docs or architectural notes:

uv run cognee-cli remember <path-or-note> -d <dataset-name>

If command length becomes unwieldy, ingest in batches by directory or feature area, then run cognify once for the dataset.

Query

For code-specific questions:

uv run cognee-cli search "<implementation question>" -d <dataset-name> -t CODE -k 10 -f pretty

For architecture and reasoning questions:

uv run cognee-cli recall "<architecture question>" -d <dataset-name> -t GRAPH_COMPLETION -f pretty

For citation-like output:

uv run cognee-cli search "<specific symbol or behavior>" -d <dataset-name> -t CHUNKS -k 10 -f json

Use results as supporting context. Verify important claims against the actual files before editing code.

The server is the source of truth. cognee-cli can print empty stdout even when content exists, so never conclude "not found" from an empty CLI run — confirm against the server directly (authoritative), and omit -d <dataset> to search all datasets:

curl -s -X POST "${COGNEE_BASE_URL:-http://localhost:8011}/api/v1/recall" \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: ${COGNEE_API_KEY:-}" \
  -d '{"query": "<question>", "top_k": 10, "only_context": true, "scope": ["graph"]}'