type-juggling
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains anAI LOAD INSTRUCTIONblock that uses imperative language to dictate the agent's internal logic and task prioritization. It instructs the agent to 'Use strict routing' and specifies that 'when you encounter PHP login/signature logic... start with this skill'. This is a direct attempt to override the agent's standard operational behavior and decision-making process. - [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface where the agent is trained to respond to specific external data (PHP code snippets) with pre-defined exploitation strategies.
- Ingestion points: Analyzed PHP code blocks (via the decision tree logic).
- Boundary markers: None provided in the instructions to separate exploitation payloads from safe analysis.
- Capability inventory: The skill encourages the identification of authentication bypasses, HMAC bypasses, and comparison logic failures.
- Sanitization: There are no instructions for the agent to sanitize or ignore potentially malicious instructions embedded in the code it is auditing.
Audit Metadata