The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/collect_diff_evidence.py script executes git and gh binaries using the subprocess.Popen function. While it implements defensive measures like cleaning environment variables and avoiding shell interpolation, the execution of external binaries based on user-supplied parameters (like commit SHAs or branch names) is a high-privilege operation.\n- [COMMAND_EXECUTION]: The scripts/generate_summary_report.py script uses subprocess.run to invoke system openers (e.g., open, xdg-open) to display generated HTML reports to the user.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize external, untrusted data including Git diffs and Pull Request metadata. An attacker could embed malicious instructions within commit messages or PR descriptions to influence the AI's analysis or the resulting report artifacts. Ingestion points: Captured output from git diff and gh pr view in scripts/collect_diff_evidence.py. Boundary markers: The workflow documentation (references/diff-summary-workflow.md) includes explicit warnings to treat all evidence as inert data. Capability inventory: Subprocess execution for git and gh, and file system write access to create report artifacts. Sanitization: scripts/generate_summary_report.py applies HTML entity escaping for report content and JSON serialization for data embedded within script tags.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates HTML reports containing interactive JavaScript from Markdown content using scripts/generate_summary_report.py. It also manages command execution through the subprocess module to interact with the local development environment.

diff-summary-md