grill-with-docs

Warn

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill implements a 'Project Thread' workflow that directs the agent to send updates to a specific Slack channel ('#brain-joel') and mention a specific user ID ('<@U030BJ3CK>'). This behavior is explicitly triggered when the project involves sensitive areas such as secrets, gateway routing, deploys, and customer-facing systems. While operator approval is requested, this pattern hardcodes an external recipient for sensitive project information.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by ingesting untrusted data from the codebase ('explore the codebase') and using it to update documentation files ('CONTEXT.md', ADRs).
  • Ingestion points: The agent is instructed to explore the codebase to answer questions and find existing documentation.
  • Boundary markers: There are no delimiters or instructions provided to the agent to distinguish between codebase content and system instructions.
  • Capability inventory: The agent has the ability to write to the filesystem ('CONTEXT.md', 'docs/adr/') and potentially communicate via Slack.
  • Sanitization: The skill lacks any instructions for sanitizing or validating content read from the codebase before it is processed or written to the filesystem.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 29, 2026, 01:52 AM
Security Audit — agent-trust-hub — grill-with-docs