capture
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data and writes it to a persistent file, creating a potential surface for indirect prompt injection.
- Ingestion points: User-supplied text captured via the
$ARGUMENTSvariable inSKILL.md. - Boundary markers: The item is formatted as a markdown list entry (
- [ ] <item>), which provides minimal separation between user content and the file structure. - Capability inventory: The skill utilizes the
ReadandEdittools to modify the target file. - Sanitization: The skill performs URL cleaning to remove tracking parameters but does not implement validation or escaping to prevent malicious instructions from being stored in the captured text.
Audit Metadata