skills/openai/plugins/security-scan/Gen Agent Trust Hub

security-scan

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection Surface: The skill is designed to ingest and analyze untrusted data, such as security advisories and external URLs, which is inherent to its primary role as a security scanner.
  • Ingestion points: Untrusted content enters the context through user-provided security context in SKILL.md and vulnerability advisory text (CVE/GHSA) in references/scan-artifacts-and-ledger.md.
  • Boundary markers: The skill explicitly directs the agent to treat this data as "untrusted analysis data," restricts link following to authorized actions only, and requires that workers remain offline during source review.
  • Capability inventory: The workflow utilizes tool calls and local Python scripts (e.g., generate_rank_input.py and finalize_scan_contract.py) to manage the scan inventory and subagent dispatch for parallel analysis.
  • Sanitization: Instructions specifically warn against treating input as shell syntax, which serves as a safeguard when data is passed to the internal analysis scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 08:05 PM
Security Audit — agent-trust-hub — security-scan