drift-analysis

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes external and potentially untrusted project data to generate synthesized reports.
  • Ingestion points: Data is collected from GitHub issues, pull requests, and documentation files like README.md and PLAN.md.
  • Boundary markers: The templates and analysis instructions do not use delimiters or instructions to ignore embedded commands within the processed data.
  • Capability inventory: The skill uses ingested data to create prioritized reconstruction plans and drift reports, which could be manipulated by content within the source files.
  • Sanitization: No sanitization or validation logic is specified for the ingested content prior to LLM analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 03:48 PM