The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to clone external repositories (via URL or local path) and execute their internal build and test gates (e.g., npm run build, cargo test, make). This execution occurs in a "scratch clone" but still runs code defined by the target repository on the agent's host environment.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands for repository analysis, including git, awk, grep, and npx. It constructs shell commands to extract dependency maps and validate commit messages against the repository's own rules.
[DYNAMIC_EXECUTION]: During the verification phase, the skill dynamically detects execution commands from repository manifests (like package.json, Cargo.toml, or Makefile) and runs them. This runtime identification and execution of scripts from an external source is a significant attack surface.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository under analysis, including file names, commit messages (git log), and configuration files. This data is used to establish the navigation map and message conventions, creating a surface where malicious content in the repository could attempt to influence the agent's behavior.
Ingestion points: Filesystem inventory (git ls-files), commit history (git log), and configuration files within the analyzed repository.
Boundary markers: None explicitly defined for isolating the analyzed content from the agent's instructions.
Capability inventory: Ability to clone repositories, execute build/test scripts, and write plan files to the local filesystem.
Sanitization: None specified for the content extracted from the repository before it is used to inform message drafting or planning.

awesome-commit-plan