valuation-red-team

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and audit valuation artifacts produced by other agents, creating an attack surface for indirect prompt injection via the reviewed data or narratives.
  • Ingestion points: Processes multiple files including classification.json, cost-of-capital.json, forecast.json, dcf-result.json, and narrative.md.
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when reading potentially untrusted data from these artifacts.
  • Capability inventory: The skill executes local Python scripts (validate.py, dcf.py) and generates finding artifacts (challenge.json, challenge.md).
  • Sanitization: There is no mention of sanitizing or validating inputs for embedded instructions or malicious patterns.
  • [COMMAND_EXECUTION]: The skill utilizes the shell to execute local Python scripts (validate.py and dcf.py) located within the platform's internal <skills> environment. These are used for mechanical consistency checks and implied expectation analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 02:36 PM
Security Audit — agent-trust-hub — valuation-red-team