review-feature

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull request descriptions and issue bodies to define the feature's scope and requirements. This creates a surface for indirect prompt injection where a malicious PR or issue description could contain instructions intended to subvert the agent's findings or influence downstream tasks.\n
  • Ingestion points: Pull request bodies retrieved via gh pr view and issue descriptions retrieved via gh issue view (SKILL.md, Phase 0).\n
  • Boundary markers: The skill outputs findings using a structured markdown format, but it lacks explicit isolation or delimiters for untrusted text interpolated during the audit process.\n
  • Capability inventory: The skill is restricted to read-only operations and does not have access to Edit or Write tools (SKILL.md, Operating Rules).\n
  • Sanitization: Findings are validated against a specific markdown schema (finding-format.md), but there is no semantic sanitization of the untrusted input to detect or neutralize instruction-overriding patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 08:12 AM
Security Audit — agent-trust-hub — review-feature