safehouse-worktrees

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to manage git worktrees and execute standard development tools including npm, cargo, pip, poetry, and go. These are performed within the local workspace context for project initialization and testing.\n- [INDIRECT_PROMPT_INJECTION]: Configuration is ingested from the project file AGENTS.md. The skill mitigates potential injection risks by validating the extracted directory paths against a specific whitelist of patterns (.worktrees, worktrees, or ~/.config/gangsta/worktrees/*) within a shell case statement before use.\n- [EXTERNAL_DOWNLOADS]: The skill fetches project dependencies from well-known registries (such as npm and PyPI). It explicitly highlights supply chain risks and recommends using lockfile-bound installation methods to ensure environment integrity.\n- [DATA_EXFILTRATION]: No exfiltration patterns were detected. The skill includes a mandatory safety step using git check-ignore to ensure worktree directories are excluded from version control, preventing the accidental exposure of workspace artifacts.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 07:46 AM