cold-reader
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which is interpolated into a sub-agent prompt, creating a vulnerability to indirect prompt injection.
- Ingestion points: The user's draft content is interpolated into the
{{TEXT}}placeholder inreferences/reader-prompt.md. - Boundary markers: The prompt uses a header (
## 本文) to indicate the user text but lacks robust delimiters or explicit instructions to the sub-agent to ignore embedded commands within that text. - Capability inventory: The skill executes a local Python script (
scripts/restate_check.py) for word reuse analysis and invokes a sub-agent to review the text. The script only uses standard libraries and performs regex-based counting, lacking network or unauthorized filesystem write capabilities. - Sanitization: There is no evidence of input validation or sanitization performed on the user text before it is presented to the sub-agent.
Audit Metadata