The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs various repository operations including deleting git branches, removing worktrees, and staging all changes. It also identifies and executes arbitrary project-specific scripts (such as linting or type-checking) found in the repository configuration.
[EXTERNAL_DOWNLOADS]: The rebase-worktrees workflow automatically detects lockfiles and runs the corresponding package manager (npm, pnpm, yarn, or bun) to install project dependencies, which involves fetching code from external package registries.
[CREDENTIALS_UNSAFE]: The skill reads and compares .env files to synchronize environment variable keys between different git worktrees. This process is restricted to local management and includes a user confirmation step before updating keys.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data from git commit messages, logs, and source code during its workflows. Ingestion points: Git history, commit author names, and file contents. Boundary markers: Generally absent, although heredocs are used in specific shell commands to mitigate injection risks. Capability inventory: File system access, git repository manipulation, and execution of shell commands and package managers. Sanitization: External data is processed without explicit validation or escaping steps.

lgnh