The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The skill explicitly permits and provides templates for executing powerful commands via Bash and mcp__plugin_supabase_supabase__execute_sql. Specifically, it suggests using pg_terminate_backend to kill database processes and GRANT to modify database permissions.
PROMPT_INJECTION (HIGH): High vulnerability to Indirect Prompt Injection. The skill instructs the agent to ingest untrusted data (system logs) to determine 'Immediate Actions' and 'Root Cause Fixes'. An attacker who can influence logs—for instance, by triggering specific errors with crafted input—could inject instructions that the agent might interpret as necessary remediation steps.
Evidence for Category 8 (Indirect Prompt Injection):
Ingestion points: System logs accessed via mcp__plugin_supabase_supabase__get_logs, file reading via Read, and output processing via Grep.
Boundary markers: Absent. The skill provides no instructions for the agent to distinguish between legitimate log data and embedded instructions.
Capability inventory: Full access to Bash and mcp__plugin_supabase_supabase__execute_sql (SQL write/admin capabilities).
Sanitization: None. There is no requirement for the agent to sanitize or validate data extracted from logs before using it to form shell or SQL commands.

error-patterns