agentic-coding
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
using-git-worktreesskill instructs the agent to execute shell commands for git operations and project environment setup, includinggit worktree add,npm install,pip install, andcargo buildbased on project configuration files.\n- [EXTERNAL_DOWNLOADS]: During workspace setup, the skill facilitates the download of project dependencies from well-known package registries via standard tools like npm, pip, and cargo.\n- [PROMPT_INJECTION]: The skill uses templates insubagent-driven-developmentandrequesting-code-reviewthat interpolate external plan text and code summaries into subagent prompts, creating a surface for indirect prompt injection.\n - Ingestion points: Task descriptions and requirements in
implementer-prompt.md,spec-reviewer-prompt.md, andcode-reviewer.md.\n - Boundary markers: Markdown headers are utilized to separate instructions from external task data.\n
- Capability inventory: The workflow involves code implementation, test execution, and git operations.\n
- Sanitization: There is no evidence of specific data sanitization for the interpolated content.
Audit Metadata