The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/inspect_repository.py script executes system git commands to discover repository state. It employs defensive programming by using subprocess.Popen with argument lists instead of shell strings, preventing command injection. Furthermore, it explicitly sanitizes the process environment by clearing sensitive Git configuration variables (e.g., GIT_CONFIG_PARAMETERS, GIT_DIR) and setting GIT_TERMINAL_PROMPT=0 to prevent credential prompts or configuration hijacking.
[INDIRECT_PROMPT_INJECTION]: The skill instructions and scripts ingest untrusted data from the project environment, such as Git branch names, commit messages, and documentation files (CONTRIBUTING.md, AGENTS.md). The skill mitigates potential injection attacks by implementing strict JSON schema validation for workflow definitions, using regex constraints for branch patterns (e.g., forbidding regex lookarounds and backreferences), and instructing the agent to clearly separate verified facts from suggestions.
[DYNAMIC_EXECUTION]: The skill identifies an optional gitflow.py plugin as an external dependency. While it describes how to execute this plugin from a host-provided path, the instructions explicitly prohibit the agent from guessing paths or executing commands without specific user authorization for write operations. The included inspect_repository.py script is self-contained and uses only the Python standard library.

git-workflow