The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads from a local file to determine configuration, which creates a potential surface for indirect prompt injection if the project file is compromised.
Ingestion points: The CLAUDE.md file is searched for directory preferences using grep_search in the SKILL.md file.
Boundary markers: No specific boundary markers or instructions to ignore embedded instructions are used when processing the content of CLAUDE.md.
Capability inventory: The skill can execute various shell commands including git, npm, cargo, pip, poetry, and go. It also has file-write capabilities for modifying .gitignore.
Sanitization: The skill does not perform explicit sanitization of the strings retrieved from CLAUDE.md before using them in directory paths.
[EXTERNAL_DOWNLOADS]: The skill automatically fetches project dependencies from well-known registries during the setup phase.
Evidence: The skill executes npm install, pip install -r requirements.txt, poetry install, go mod download, and cargo build based on the detection of standard project manifest files. These operations download packages from official services such as NPM, PyPI, and Crates.io.
[COMMAND_EXECUTION]: The skill executes several system and build commands as part of its core functionality.
Evidence: Uses git worktree add, git rev-parse, and git check-ignore for workspace management. It also runs project tests using npm test, cargo test, pytest, and go test to verify the baseline state.
[DYNAMIC_EXECUTION]: The skill constructs file paths and commands dynamically based on environment variables and project configuration.
Evidence: The worktree path is dynamically assembled in SKILL.md using the $LOCATION, $project, and $BRANCH_NAME variables before being passed to the git worktree add command.

using-git-worktrees