The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands via a custom Node.js script. It includes a validateBranchName function that sanitizes branch references and prevents command injection by rejecting strings starting with dashes, control characters, or unsafe ref patterns such as .. or @{.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill generates reports based on local repository data and identity maps. It mitigates CSV injection risks by using a csvCell function that prepends a single quote to any cell content beginning with risky characters (=, +, -, @), preventing the execution of malicious formulas when the report is opened in spreadsheet software.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted commit messages from the repository history. While these messages could theoretically contain prompt injection attempts, the risk is addressed through the skill's mandatory 'Critical Thinking Mindset' and 'AI Mistake Prevention' protocols, which require the agent to verify all claims against evidence.
[SAFE]: No malicious obfuscation, hardcoded credentials, or unauthorized network operations were detected. The script is restricted to local filesystem operations and includes checks to ensure reports are not written to sensitive directories like .claude.

git-developer-performance