The Agent Skills Directory

[COMMAND_EXECUTION]: The execute_sql tool in SKILL.md allows for the execution of arbitrary PostgreSQL queries against a project's database, providing full data access and modification capabilities.\n- [COMMAND_EXECUTION]: The deploy_edge_function tool enables the deployment of arbitrary code to Supabase Edge Functions, allowing for serverless logic execution.\n- [CREDENTIALS_UNSAFE]: The get_publishable_keys tool provides the ability to retrieve sensitive API keys, including service role keys, for a Supabase project.\n- [PROMPT_INJECTION]: The skill facilitates the processing of data from potentially untrusted external sources, creating a surface for indirect prompt injection.\n
Ingestion points: Tools such as get_logs, execute_sql, and search_docs ingest data from Supabase environments and documentation which could contain malicious instructions.\n
Boundary markers: The skill documentation does not specify the use of delimiters or 'ignore' instructions for processed data to prevent the agent from obeying instructions embedded in tool outputs.\n
Capability inventory: The skill possesses significant capabilities, including execute_sql for database modification and deploy_edge_function for code deployment, which could be abused if an injection is successful.\n
Sanitization: There is no mention of data validation or sanitization before the ingested data is returned to the agent's context.

managing-supabase