oma-pm

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user requirements and project context to generate planning artifacts, which creates a surface for indirect prompt injection attacks where malicious data could influence task decomposition.
  • Ingestion points: SKILL.md defines inputs including "User request", "product goal", and "Existing codebase context".
  • Boundary markers: The provided instructions do not specify explicit delimiters or structural markers to wrap untrusted user content.
  • Capability inventory: The skill uses file system analysis tools (get_symbols_overview, search_for_pattern) and writes results to the local .agents/results/ directory.
  • Sanitization: No evidence of input sanitization or validation logic is present in the skill resources.
  • [NO_CODE]: The skill as submitted consists entirely of Markdown guidance and JSON templates. The absence of executable scripts (Python, JavaScript, or shell) eliminates common vectors for code-based attacks such as remote code execution or system-level persistence.
  • [SAFE]: The skill incorporates security-positive guardrails, such as requiring security and testing considerations for every decomposed task and defining a 'scope' property to restrict downstream agent file system access. It also provides a structured error recovery playbook to handle ambiguous requirements safely.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 04:52 PM