The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill reads and interprets instructions from untrusted external files within the project directory.
Ingestion points: The skill reads directory preferences from CLAUDE.md and identifies build/test requirements from package.json, requirements.txt, Cargo.toml, pyproject.toml, and go.mod.
Boundary markers: No boundary markers or isolation techniques are used when processing these files.
Capability inventory: The skill possesses extensive execution capabilities including shell command execution, file modification (.gitignore), and network-enabled package managers.
Sanitization: No sanitization is performed on the data ingested from the project files before it influences agent behavior or command execution.
[Command Execution] (HIGH): The skill automatically executes setup and test suites which can run arbitrary code defined in the repository.
Evidence: Steps 3 and 4 in SKILL.md explicitly instruct the agent to run npm install, npm test, cargo build, cargo test, pip install, poetry install, and go test based on file detection.
[External Downloads] (MEDIUM): The skill invokes package managers that fetch and execute third-party code from public registries (npm, PyPI, etc.).
Evidence: Commands like npm install and pip install -r requirements.txt download external dependencies at runtime.

using-git-worktrees