The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands via subprocess.run in the visualize_changes.py script. It uses list-based argument passing rather than shell strings, which is a best practice for preventing shell injection vulnerabilities. While it lacks the -- separator to strictly demarcate options from positional arguments (like branch names), the risk is minimal given the tool's intended use in local development environments.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it ingests untrusted data from Git diffs, commit messages, and file contents.
Ingestion points: git diff output and annotations.json in scripts/visualize_changes.py.
Boundary markers: The skill instructions explicitly separate the 'read diff' step from the 'write annotations' step.
Capability inventory: Limited to local git commands and file writing to the OS temporary directory.
Sanitization: The script effectively mitigates potential XSS in the generated HTML report by using html.escape() for all content rendered into the template.
[SAFE]: The skill adheres to several security best practices: it relies exclusively on the Python standard library, performs no network operations, and does not download or execute remote code.

git-visualizer