global-reviewer
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted survey drafts, creating a potential surface for indirect injection where malicious input could influence report content. 1. Ingestion points: output/DRAFT.md, outline/*.yml, and citations/ref.bib via the scripts/run.py script. 2. Boundary markers: Absent; the script reads raw file content for metadata extraction without specific delimiters. 3. Capability inventory: Limited to workspace file read/write operations; no network access or shell execution of ingested data. 4. Sanitization: A dedicated _sanitize function in scripts/run.py redacts TODO/TBD placeholders and scaffold comments before writing the final report.
- [SAFE]: The skill is designed for survey auditing and operates purely on local workspace files. Analysis of the scripts and instructions found no evidence of exfiltration, obfuscation, or unauthorized privilege escalation. The role prompts and workflow are consistent with the stated purpose of improving draft coherence.
Audit Metadata