using-superpowers

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses extremely forceful instructions to mandate compliance, such as 'ABSOLUTELY MUST', 'not negotiable', and 'not optional', which are override markers aimed at bypassing the agent's default logic.
  • [PROMPT_INJECTION]: The skill explicitly instructs that 'Superpowers skills override default system prompt behavior', attempting to alter the agent's internal instruction hierarchy.
  • [PROMPT_INJECTION]: The 'Red Flags' section includes directives to suppress the agent's internal reasoning ('STOP—you're rationalizing'), a technique often used to ensure adherence to instructions over safety or common sense.
  • [PROMPT_INJECTION]: The framework handles external data from web tools and subagents, creating a surface for indirect prompt injection.
  • Ingestion points: 'web_fetch' and 'google_web_search' tools in 'references/gemini-tools.md' and subagent outputs in 'references/codex-tools.md'.
  • Boundary markers: XML tags like '' are suggested but not consistently enforced for all inputs.
  • Capability inventory: Mapped tools include shell command execution ('run_shell_command', 'Bash') and file system operations ('write_file', 'Edit').
  • Sanitization: No explicit validation or sanitization of ingested content is defined.
  • [COMMAND_EXECUTION]: The platform mapping files document and enable access to powerful tools for shell command execution and file modification as part of the intended workflow framework.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 02:46 AM
Security Audit — agent-trust-hub — using-superpowers