docfactory-backlog

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION] (HIGH): The skill is vulnerable to Indirect Prompt Injection by processing untrusted project documentation into executable tasks for downstream agents.\n
  • Ingestion points: Reads from multiple markdown files including project brief, PRD, UI/UX spec, and architecture files (e.g., 02-prd.md, 04-tech-architecture.md).\n
  • Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands within the ingested source documents.\n
  • Capability inventory: The skill generates 06-backlog.md which contains "Verification" steps. The documentation explicitly states these are intended for an IDE agent to execute with "high confidence."\n
  • Sanitization: Absent. The skill performs no validation or escaping of the content derived from ingested documents before interpolating it into executable fields.\n- [COMMAND_EXECUTION] (LOW): The skill includes and encourages the use of a local Python script for validation.\n
  • Evidence: scripts/validate_docfactory_backlog.py is used to verify the structure of the output. While benign (using standard libraries for regex and file reading), it represents a command execution surface within the skill's operational workflow.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 09:10 AM