The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill attempts to harvest GitHub authentication tokens from the local ~/.git-credentials file if a GITHUB_TOKEN environment variable is not provided.
[COMMAND_EXECUTION]: The skill involves checking out untrusted code from pull requests and suggests running automated checks like tests or linters on that code, which provides an execution surface for malicious PR content.
Ingestion points: The skill fetches and checks out remote pull request branches using git fetch and git checkout.
Boundary markers: No explicit delimiters are used to separate untrusted code changes from the agent's instructions.
Capability inventory: The skill utilizes shell command execution for Git operations, Curl requests, and suggested execution of tools like pytest and ruff on the checked-out codebase.
Sanitization: No sanitization or safety validation is performed on the pull request content before it is processed or used in local command execution.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to the GitHub API (api.github.com) using curl and the gh CLI to fetch pull request data and post review comments. These operations target a well-known service and are consistent with the skill's primary function.

github-code-review