The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to fetch and analyze external content, specifically pull request comments and review bodies. Because this data originates from outside the agent's direct control, it represents a potential surface for indirect instructions.
Ingestion points: The scripts/fetch-pr-info.js script retrieves comment and review text from the GitHub API using the gh tool.
Boundary markers: The instructions in SKILL.md do not include specific delimiters or warnings to help the agent distinguish between informational content in PR comments and its own operational instructions.
Capability inventory: The skill's objective is to help the user address review comments, which involves analyzing code diffs and history. The agent maintains access to the local shell to perform these tasks.
Sanitization: Although the script filters out certain standard automated messages, it does not include mechanisms to sanitize or validate the content of human-authored comments for potential prompt injection attempts.
Command Execution: The skill relies on executing several shell commands to gather project metadata. This is a common pattern for developer-focused skills but requires appropriate local permissions.
Evidence: The script scripts/fetch-pr-info.js uses exec to run git branch, gh auth status, gh pr diff, and gh api commands to collect the necessary PR information.

pr-address-comments