write-as-me
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user-supplied data and file content directly into instructions without sanitization or boundary markers.
- Ingestion points: Arbitrary user input is captured via
$ARGUMENTSinSKILL.md, and the skill specifically referencesAUTHOR_VOICE.mdas a source document. - Boundary markers: The prompt lacks delimiters (such as XML tags or triple quotes) to separate the instruction "Topic/prompt" from the user-provided data, increasing the risk of the model following instructions embedded within the input.
- Capability inventory: The skill delegates tasks to the
voice-writeragent; no direct system commands, file system writes, or network requests are explicitly defined in this skill file. - Sanitization: There is no evidence of filtering, escaping, or length-limiting applied to the input variables before they are processed by the agent.
Audit Metadata