embed

# Embed Skill Embeds a text question using `gemini-embedding-001` (Vertex AI, project `mozdata`) and returns the resulting vector as a JSON array. ## Usage ```bash python "${CLAUDE_PLUGIN_ROOT:?set by the plugin system; if empty, invoke this via the Skill tool}"/skills/embed/scripts/embed.py --question "<your question>" ``` `` Output is a JSON array of 3072 floats printed to stdout: ```json [0.012, -0.034, 0.091, ...] ``` ## Steps 1. Run the script with the user's question as `--question`. 2. Capture stdout — that is the embedding vector. 3. Return or pass the vector as needed (e.g. to the `vector-search` skill). ## Authentication This skill connects to Vertex AI using a service account — the `cloud-platform` scope is enforced on the impersonated token in code. Just log in: ```bash gcloud auth application-default login ``` Your authenticated Google Cloud credentials are only used to create a new and temporary access token on behalf of the service account defined in `SERVICE_ACCOUNT`, which requires the `roles/iam.serviceAccountTokenCreator` role on that service account. Your login no longer needs `--scopes`. ## Troubleshooting | Symptom | Fix | |---------|-----| | `GCP authentication required` / `API error 401` | Re-run `gcloud auth application-default login`; confirm you have `roles/iam.serviceAccountTokenCreator` on the service account in `SERVICE_ACCOUNT` | | `Missing dependency` | `pip install google-auth requests` | | `Unexpected response from Vertex AI` | Confirm Vertex AI is enabled in project `mozdata` and the embedding model is available |