check-idea
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local shell commands to gather environment context, specifically invoking
git remoteto determine the repository name andwhoami(via MCP) to identify the organization context. - [REMOTE_CODE_EXECUTION]: The agent is instructed to execute command-line arguments (
git_argv) dynamically provided by theprepare_review_difftool. While this involves executing tool-generated strings, the behavior is a documented part of the Connectory MCP workflow for generating unified diffs for policy validation. - [SAFE]: The skill includes an explicit 'Hard stops' section in
references/mcp-rules.mdthat restricts the agent from searching for sensitive files, probing tokens, or performing unauthorized network operations. These instructions act as a security guardrail against common agent-based attack patterns.
Audit Metadata