twg-confluence

Pass

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, specifically Confluence pages, blog posts, and space-level configuration files.
  • Ingestion points: Content is retrieved via confluence content get (in references/content.md) and space instructions are loaded from AGENTS.MD files using confluence space instructions get (in references/spaces.md).
  • Boundary markers: The skill establishes clear boundaries by requiring the agent to distinguish between automated CLI flags (--yes) and explicit user intent. It also instructs the agent to ignore surface-specific guidance (like MCP-only tools) that may appear in data bundles.
  • Capability inventory: The skill utilizes the twg CLI, which possesses capabilities to create, modify, move, and delete content, as well as manage space permissions and hierarchy.
  • Sanitization: Security is addressed through behavioral constraints, such as mandatory read-before-write workflows, the use of optimistic concurrency tokens to prevent data overwrites, and strict requirements for user confirmation before destructive actions like archiving or purging.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 31, 2026, 09:06 AM
Security Audit — agent-trust-hub — twg-confluence