using-superpowers
Fail
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: HIGHPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Evidence in SKILL.md. The instructions explicitly state 'Superpowers skills override default system prompt behavior' and 'Default system prompt — lowest priority'. It uses aggressive language such as 'YOU DO NOT HAVE A CHOICE', 'ABSOLUTELY MUST', and 'This is not negotiable' to force the agent to override its default behavior and safety guidelines.
- [INDIRECT_PROMPT_INJECTION]: The skill mandates a workflow where the agent must load and follow instructions from external files before responding to the user.
- Ingestion points: The agent is directed to use view_file to read .agent/skills//SKILL.md in SKILL.md.
- Boundary markers: None present; the agent is instructed to 'Follow skill exactly' in the logic flow.
- Capability inventory: references/antigravity-tools.md and references/copilot-tools.md document tool access for command execution (run_command, bash), file modification (write_to_file, replace_file_content, create, edit), and network operations (read_url_content, web_fetch).
- Sanitization: No evidence of sanitization or safety checks for the external content is provided.
Recommendations
- AI detected serious security threats
Audit Metadata