scientific-writing-draft

Pass

Audited by Gen Agent Trust Hub on Jul 17, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns, such as prompt injection, obfuscation, or unauthorized network activity, were detected in the skill instructions or style guide.
  • [DATA_EXFILTRATION]: The skill reads project files (story.md, outline.md, references.md) and writes to draft.md. These operations are essential for the drafting process and are performed within the local workspace environment.
  • [COMMAND_EXECUTION]: The reference to 'dispatching a subagent' is a high-level task-routing instruction for the AI and does not represent a request to execute arbitrary shell commands or external scripts.
  • [PROMPT_INJECTION]: The skill processes untrusted user data and has file-writing capabilities, which represents an indirect prompt injection surface. (1) Ingestion points: Reads story.md, outline.md, and references.md. (2) Boundary markers: Instructions require all generated prose to trace back to specific outline bullets, with explicit 'Red flags' to prevent deviation. (3) Capability inventory: Writes generated draft content to draft.md. (4) Sanitization: No explicit text sanitization is documented.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 17, 2026, 05:35 PM
Security Audit — agent-trust-hub — scientific-writing-draft