The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/common.py to execute git commands for repository analysis, including diff, log, show, and rev-parse. These operations are performed with argument lists and without the shell=True flag, minimizing the risk of shell injection.
[INDIRECT_PROMPT_INJECTION]: The agent is tasked with interpreting repository diffs and pull request descriptions to author review summaries. This creates a surface area where an adversary could embed instructions or misleading code patterns in a pull request to influence the AI's summary or verdict.
Ingestion points: Git diff output and PR descriptions are ingested during analysis in SKILL.md and scripts/analyze_diff.py.
Boundary markers: The skill uses  comments for document structure, though these do not serve as hardened security boundaries.
Capability inventory: The skill possesses capabilities for local file reading, file writing (HTML report generation), and Git command execution.
Sanitization: The skill implements HTML escaping for code snippets and file paths in its scripts and report template. Furthermore, scripts/lint_fragments.py explicitly checks for and blocks executable <script> blocks in report fragments to prevent arbitrary JavaScript execution.
[EXTERNAL_DOWNLOADS]: The generated HTML reports (assets/template.html) reference external resources from well-known services, including mermaid.min.js via jsDelivr and typography from Google Fonts.

pr-visualization