The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's database setup script generates a full PostgreSQL connection URI using a password retrieved from the local environment configuration.
Evidence: scripts/setup_database.py utilizes env.get("SUPABASE_DB_PASSWORD") to construct the connection string required for direct database access via psycopg2.
[DATA_EXFILTRATION]: The lead search pipeline transmits existing LinkedIn URLs from the Supabase database to the CrustData API to ensure only net-new leads are returned.
Evidence: scripts/prospect_search.py fetches existing profile URLs and includes them in the exclude_profiles request body sent to api.crustdata.com.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays professional profile data (e.g., headlines and skills) from the CrustData API, which could potentially contain adversarial instructions designed to influence the agent.
Ingestion points: Professional profile data is fetched from the CrustData search endpoint in scripts/prospect_search.py.
Boundary markers: The profile data is presented to the user and stored in the database without the use of explicit boundary delimiters.
Capability inventory: The skill has the ability to write to the local filesystem (CSV) and the remote Supabase database.
Sanitization: Profile strings are extracted and processed without specific sanitization filters for instruction-like patterns.

crustdata-supabase