The Agent Skills Directory

[SAFE]: The skill follows security best practices by implementing a 'fail-closed' architecture, ensuring that any failed validation or safety check stops the process without generating potentially dangerous artifacts.
[COMMAND_EXECUTION]: External calls to Git and GitHub tools are executed safely using argument arrays rather than shell strings, preventing command injection. The execution environment is strictly sanitized to strip potentially malicious configuration or sensitive environment variables.
[DATA_EXFILTRATION]: The evidence collection script includes a robust sensitive path filter that rejects files containing credentials, private keys, certificates, or environment variables. It also enforces strict directory boundary checks to prevent path traversal attacks.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly mitigates risks from malicious repositories by performing a 'clean-filter preflight' check, which prevents the execution of repository-provided clean or smudge filters. It also provides clear instructions for the AI agent to treat diff and commit content as untrusted data.
[DYNAMIC_EXECUTION]: HTML report generation uses secure JSON serialization and HTML escaping to safely render untrusted diff content, protecting against Cross-Site Scripting (XSS) when viewing the summaries in a browser.

diff-summary