worktree
Warn
Audited by Gen Agent Trust Hub on Apr 10, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill identifies, lists, and copies sensitive environment configuration files (matching the
.env*pattern) when setting up new worktrees. - Evidence: Step 1 (
infocommand) detectsenvFilesand Step 2 asks the user to select files to be copied to the new worktree directory. - [COMMAND_EXECUTION]: The skill executes a local Node.js script (
.claude/scripts/worktree.cjs) via shell commands that incorporate variables derived from user input, such as project names and feature descriptions. - Evidence: Step 4 shows the construction of execution strings like
node .claude/scripts/worktree.cjs create "<PROJECT>" "<SLUG>" --prefix <TYPE> --env "<FILES>". - [PROMPT_INJECTION]: The skill contains instructional blocks that use imperative language to mandate specific task-planning behaviors and override default agent operation modes.
- Evidence: Frontmatter-adjacent 'IMPORTANT' blocks and 'IMPORTANT Task Planning Notes' at the end of the file dictate how the agent must break down tasks and verify work.
Audit Metadata