The Agent Skills Directory

[COMMAND_EXECUTION]: The query command in scripts/supabase.sh allows the execution of arbitrary SQL strings. When used with the SUPABASE_SERVICE_KEY, which bypasses Row Level Security (RLS), the agent gains full administrative control over the database schema and data.
[EXTERNAL_DOWNLOADS]: The vector-search command fetches embeddings from OpenAI's official API (api.openai.com). This is a well-known service used for legitimate vector operations.
[PROMPT_INJECTION]: The skill processes data retrieved from database tables (via select or query) and passes it back to the agent without sanitization or boundary markers, creating an indirect prompt injection surface.
Ingestion points: Data returned from Supabase REST and RPC endpoints in scripts/supabase.sh.
Boundary markers: None identified in the script or instructions.
Capability inventory: The skill can execute arbitrary SQL (cmd_query), perform network requests via curl, and call generic database functions (cmd_rpc).
Sanitization: The script uses jq to properly encode outgoing JSON payloads and SQL strings, but it does not sanitize or validate incoming data from the database before presenting it to the agent.

supabase