spike
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided content through the
$ARGUMENTSvariable to define the research question and experiment approach, which the agent then executes. - Ingestion points: Input is accepted via the
$ARGUMENTSvariable inSKILL.mdand used to drive the technical procedure. - Boundary markers: The skill lacks explicit delimiters or instructions to prevent the agent from following potential commands embedded within the provided arguments.
- Capability inventory: The agent is instructed to create files, install dependencies, and execute code (described in
SKILL.mdunder Step 2: Minimal experiment). - Sanitization: No validation or escaping of the input arguments is specified.
- Remediation: Mitigation can be improved by wrapping input in specific delimiters and instructing the agent to treat the content strictly as data for the research question.
Audit Metadata