The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses child_process.execFileSync to execute standard git commands (e.g., git log, git rev-parse, git show). Analysis shows these commands are safely constructed with an array of arguments, preventing shell injection. Input like branch names is validated against control characters and unsafe ref sequences (e.g., .., @{, \\).
[DATA_EXPOSURE_&_EXFILTRATION]: The skill instructions explicitly mandate using local Git history only and forbid querying external services. All network-related tools are absent from the allowed-tools or script implementation. The script strictly processes local repository data.
[PRIVILEGE_ESCALATION]: The skill operates within the user's current environment and does not attempt to use sudo or modify system-level configurations. It enforces that its output remains outside the .claude configuration directory to prevent internal state pollution.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were detected. All scripts are local to the skill package, and dependencies are limited to standard Node.js built-ins (fs, path, crypto, child_process).
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it reads git log output and commit patches (via git show). However, it includes specific instructions for the agent to treat these as evidence-based estimates and provides clear boundary markers in the generated work-packets/*.md files. The risk is mitigated by the analytical nature of the task and the lack of high-privilege write capabilities for the ingested data.

git-developer-performance