revise-spec-from-review

Pass

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions and associated evaluation fixtures focus exclusively on structured text manipulation of local markdown files. There are no network requests, shell command executions, or external script downloads defined.
  • [SAFE]: The evaluation data (e.g., 'evals/sources/browser-boundary-observed.txt') contains sanitized examples with explicit notes that proprietary information and live credentials have been excluded.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to ingest and act upon untrusted review findings.
  • Ingestion points: The agent reads from a 'named spec folder' and a 'supplied adversarial-spec-review' (e.g., 'review.txt').
  • Boundary markers: Absent. The instructions do not mandate specific delimiters for external content, though they provide logical constraints for the agent's persona.
  • Capability inventory: The agent has file-system write access to the targeted spec folder to modify '.md' files.
  • Sanitization: Absent. The skill does not explicitly sanitize or validate the content of the review findings before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 12, 2026, 07:11 AM
Security Audit — agent-trust-hub — revise-spec-from-review