The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script scripts/inspect_repository.py that executes Git commands to discover repository metadata.
Execution Safety: The script uses subprocess.Popen with a list of arguments rather than a shell string, preventing shell injection vulnerabilities. The arguments passed within the discovery functions are hardcoded strings (e.g., rev-parse, symbolic-ref, for-each-ref).
Environment Sanitization: The script implements a security-conscious environment by explicitly removing sensitive Git variables (e.g., GIT_CONFIG_PARAMETERS, GIT_DIR, GIT_INDEX_FILE, GIT_CONFIG_KEY_*) and setting GIT_TERMINAL_PROMPT='0' to prevent credential leakage or unauthorized interactive prompts.
Resource Management: It includes robust timeout handling (20s default) and process group termination (SIGKILL) to ensure that Git operations do not hang or consume excessive system resources.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local Git repositories, such as branch names, status, and configuration files, which could potentially contain malicious instructions.
Mitigation Through Structure: The skill mitigates this surface by using a Python script to parse Git output into structured JSON, which helps prevent the agent from misinterpreting repository content as direct instructions.
Instructional Guardrails: Main instructions in SKILL.md require the agent to clearly distinguish between "facts," "rules," and "user authorization," and to never assume that presence of a configuration file grants permission for destructive actions.

git-sync