author-style-guide
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill directly incorporates user-provided
$ARGUMENTSinto the primary instruction line, creating a standard surface for direct prompt injection. - [PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface. The skill ingests data from
$ARGUMENTSand{{CONFIG_PATH}}to generate 'Reference Prompt Blocks' meant for subsequent image generation tasks. This allows instructions hidden in these sources to influence the agent's future behavior. - Ingestion points: User-provided descriptions in
$ARGUMENTSand brand identity data from{{CONFIG_PATH}}(SKILL.md). - Boundary markers: Absent. The skill does not use delimiters or instructions to prevent the agent from obeying instructions embedded in the style guide content.
- Capability inventory: File read/write operations (SKILL.md) and image generation using the
@google/genailibrary. - Sanitization: None. Data is extracted and formatted into the style guide without validation or escaping.
Audit Metadata