spec-reviewer

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted technical specifications that could contain instructions to manipulate the review outcome.\n
  • Ingestion points: Specification content is ingested from user messages, local file paths, or conversation history (SKILL.md, Step 1). Additional data is retrieved from Jira tickets using the brianna skill (SKILL.md, Step 2).\n
  • Boundary markers: The instructions do not define clear delimiters or "ignore" instructions for the ingested specification content, potentially allowing embedded commands to influence the agent's behavior.\n
  • Capability inventory: The skill utilizes the brianna skill for data retrieval and references local architectural guides. It does not possess direct command execution or network exfiltration capabilities.\n
  • Sanitization: The skill mitigates the impact of potential injections by explicitly forbidding the generation of code snippets and requiring a highly structured output format (APPROVED or NEEDS WORK).
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:26 AM