osprey-appraise

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, which is a common vector for indirect prompt injection attacks.
  • Ingestion points: In Phase 1A, the skill is instructed to 'Read whatever format — PDF, markdown, email' for client-provided audits, as well as ingesting Raven case files and Hawk survey reports. In Phase 1B, it is instructed to perform a 'Quick Scope Survey' scanning for secrets in code and auth patterns.
  • Boundary markers: The instructions do not specify any boundary markers or delimiters (e.g., XML tags or special tokens) to separate untrusted client data from the agent's core instructions.
  • Capability inventory: The skill's primary capabilities involve reading files and writing markdown files (Phase 5C: quotes/[client-name]-[date]-security-remediation.md). It does not appear to use subprocess calls, network operations, or other high-risk tools.
  • Sanitization: There are no instructions for sanitizing or escaping the content of external audits before they are processed by the agent to generate a proposal.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 01:31 AM