supabase-audit-auth-users

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill utilizes authoritative and forceful language ("CRITICAL: PROGRESSIVE FILE UPDATES REQUIRED", "MUST write to context files AS YOU GO", "Failure to write progressively is a critical error") intended to override the agent's default task execution and output buffering behavior.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting data from external Supabase authentication API responses and writing them directly into local tracking and evidence files.
  • Ingestion points: Supabase authentication endpoint responses (/auth/v1/signup, /auth/v1/token, /auth/v1/recover, /auth/v1/otp).
  • Boundary markers: Absent; the skill does not specify delimiters or warnings for the agent when processing or logging these external responses.
  • Capability inventory: File system write access to hidden and specific directories (.sb-pentest-context.json, .sb-pentest-audit.log, .sb-pentest-evidence/), and network operations via curl.
  • Sanitization: Absent; responses are logged directly to evidence and context files without validation or filtering.
  • [COMMAND_EXECUTION]: The skill instructs the agent to perform network-based testing using curl and Python scripts to analyze timing and response patterns from remote authentication servers.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 06:43 PM
Security Audit — agent-trust-hub — supabase-audit-auth-users