The Agent Skills Directory

Command Execution via GitHub CLI:
The skill executes shell commands using the gh binary via subprocess.run in scripts/fetch_comments.py to retrieve PR metadata and comments.
The execution uses structured lists rather than passing a raw shell string, mitigating common shell injection vulnerabilities.
Indirect Prompt Injection Surface:
Ingestion points: Pull request conversation comments, reviews, and threads are fetched from external repositories via GraphQL queries and supplied directly to the agent context.
Boundary markers: No specific boundary isolation or filtering instructions are placed around the fetched comment strings to differentiate them from core instruction prompts.
Capability inventory: The skill possesses capability to modify code files based on input as highlighted by the instructions to 'Apply fixes for the selected comments'.
Sanitization: External content parsed from GitHub is passed into stdout as raw text data without sanitization, exposing a potential attack surface if an external contributor inserts malicious system-override instructions into a pull request comment.

gh-address-comments