post-merge-voice-gate

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it analyzes output/DRAFT.md, which contains content generated in previous automated steps. If this draft content contains malicious instructions, the agent may be influenced by them during analysis.
  • Ingestion points: The skill reads and processes output/DRAFT.md and outline/transitions.md within its analysis workflow.
  • Boundary markers: The role prompt in SKILL.md does not specify clear delimiters (such as XML tags or unique markers) to separate the instructions from the data being analyzed, which is a common best practice for preventing injection.
  • Capability inventory: The skill has capabilities to read files (read_text) and write reports to the local file system (atomic_write_text, write_quality_report).
  • Sanitization: The Python script provides a layer of deterministic sanitization by using regular expressions to match specific patterns, but the LLM-based role prompt does not incorporate validation or sanitization of its input.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 01:59 PM