The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates primarily through the gh CLI. It implements a significant security control via scripts/checked_gh.py, which validates the active GitHub login and target host before any write operation is permitted. This effectively prevents unauthorized actions or accidental execution under the wrong account context.
[DATA_EXPOSURE]: Documentation across the skill emphasizes the protection of GitHub tokens and SSH keys. It provides specific patterns for using ephemeral Docker configurations for GHCR publishing and explicitly warns against printing or logging credentials.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted data from GitHub (such as PR bodies and issue comments), it incorporates strict 'read-only preflight' and 'independent readback' requirements. These instructions ensure the agent verifies the current state and intended outcome before reacting to external data, minimizing the risk of following malicious instructions embedded in processed content.
[DYNAMIC_EXECUTION]: The inclusion of checked_gh.py involves the execution of sub-processes. This script is designed as a security hardening tool that sanitizes command arguments to prevent flag injection and ensures a clean environment for sensitive operations.

github-ops