The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using the GitHub CLI (gh) and Git to perform repository management tasks such as merging PRs, deleting branches, and closing issues. All commands use strict argument validation (e.g., verifying PR numbers are positive integers) and use restricted flags for methods like merging, preventing command injection.
[SAFE]: The scripts ingest external data from GitHub PR comments to verify if a review has passed. This potential indirect prompt injection surface is mitigated by requiring specific hidden markers () and, crucially, validating that the author of the review comment matches the expected reviewer recorded in the local manifest. This ensures that unauthorized third-party comments cannot trigger automated merges.
[SAFE]: The skill interacts with local manifest files stored within the .relay directory to track run states. It does not access sensitive system files (such as SSH keys or AWS credentials) and does not perform any unauthorized network exfiltration. All external network interactions are limited to official GitHub API calls via the gh tool.

relay-merge