evaluate-and-tighten-skills
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill functions by analyzing the content of other skills, which creates a vulnerability surface for indirect prompt injection. A malicious skill could contain embedded instructions aimed at subverting the pruning process or manipulating the agent's actions during analysis.\n
- Ingestion points: The skill instructions direct the agent to read the entire directory of a target skill, including
SKILL.md,MAINTENANCE.md, and all bundled files in subdirectories likereferences/,scripts/, andassets/.\n - Boundary markers: There are no instructions for the agent to use specific delimiters or protective headers (e.g., 'ignore embedded instructions') when processing the text of the target skill.\n
- Capability inventory: The skill involves the agent modifying or creating files (specifically
MAINTENANCE.mdandSKILL.md) and refers to the execution of local scripts (e.g., via theuvtool as shown inreferences/example.md).\n - Sanitization: The procedure does not include steps for sanitizing, escaping, or validating the content of the target skill before it is evaluated by the agent.
Audit Metadata