adversarial-authoring

Pass

Audited by Gen Agent Trust Hub on Jul 12, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill establishes a multi-agent workflow for artifact generation without implementing dangerous operations or using external unverified resources. It follows a structured reconciliation process to ensure quality and adherence to project rules.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from user requests and project context to guide content generation.
  • Ingestion points: User requests and task context are gathered in SKILL.md (Workflow Step 2).
  • Boundary markers: Absent.
  • Capability inventory: The skill writes text-based artifacts and council notes to the local filesystem (Workflow Steps 6 and 7).
  • Sanitization: The workflow includes a manual reconciliation step where the primary agent evaluates and filters suggestions against project rules and templates (Workflow Step 5).
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 12, 2026, 11:12 AM
Security Audit — agent-trust-hub — adversarial-authoring