The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands for workspace management and validation.
Evidence includes Git operations (git worktree add, git check-ignore, git rev-parse), directory operations (ls, cd, basename), and running project-specific test runners (npm test, cargo test, pytest, go test).
[EXTERNAL_DOWNLOADS]: The skill triggers the download of external dependencies through standard package managers based on files found in the repository.
Evidence: Execution of npm install, pip install, poetry install, and go mod download when relevant configuration files are detected.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests data from untrusted repository files to drive high-capability actions.
Ingestion points: The skill reads package.json, requirements.txt, Cargo.toml, go.mod, and pyproject.toml to determine setup steps. It also searches GEMINI.md for directory preferences using grep.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands in the ingested files are present.
Capability inventory: The skill can execute package installations and run arbitrary test suites, which effectively allows code execution defined within the repository files.
Sanitization: No sanitization or validation of the contents of the configuration files is performed before they trigger command execution.

using-git-worktrees