grill-with-artifact

Warn

Audited by Gen Agent Trust Hub on Nov 18, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of external tools (grilling, domain-modeling) from a third-party GitHub repository (mattpocock/skills). These dependencies are installed via the skills CLI using specific commit hashes to extend the agent's capabilities.
  • [COMMAND_EXECUTION]: The instructions explicitly provide npx command blocks for the agent to present to the user or execute to add the necessary external skills to the environment.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill ingests potentially untrusted data from multiple sources: ArtifactComments, ArtifactData entries, and the content of "wayfinder tickets" (titles and bodies) which are used as the primary context for the session.
  • Boundary markers: No specific boundary markers or instructions to disregard embedded commands are implemented when the agent interpolates external data from tickets or artifact comments into its reasoning process.
  • Capability inventory: The agent is granted capabilities to write new files to the local repository (such as Architectural Decision Records and glossary terms) and to modify existing tracking files like docs/agents/issue-tracker.md based on this external input.
  • Sanitization: While the artifact page performs basic HTML escaping for display, there is no verification or sanitization of the content before the agent writes it to the repository as persistent documentation or resolves issues.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 18, 2026, 09:33 AM