spec-engine

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data such as user requests, tickets, research, and Azure DevOps (ADO) handoffs, creating a surface for indirect prompt injection.
  • Ingestion points: Untrusted data enters the agent context through user input and normalized ado-gateway handoff JSON as described in SKILL.md and rules/handoff-consumption.md.
  • Boundary markers: The skill employs a structured output contract and specific rules to isolate external review comments from core requirements, which helps reduce the risk of the agent accidentally following instructions embedded in the data.
  • Capability inventory: The skill's environment is restricted to the read_file tool, and it lacks network access or file-writing permissions, significantly limiting the potential impact of an injection attack.
  • Sanitization: Mitigation is provided through rules/requirement-quality.md, which bans vague or untestable language, and a validation script (scripts/validate-output.sh) that allows the agent to verify its own output against quality and safety rules.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 05:46 PM