The Agent Skills Directory

[SAFE]: The skill implements strong boundary controls, requiring explicit user authorization before performing any external action. It specifically prohibits creating PRs from protected branches like main or master and prevents the creation of duplicate PRs.
[SAFE]: The skill includes a pre-publication safety check using a Python script (scripts/check-pr-content.py). This script scans for a wide range of sensitive data patterns, including:
Hardcoded credentials (AWS keys, GitHub tokens, Slack tokens)
Private SSH keys
Local environment artifacts (home directories, internal IP addresses, local hostnames)
Internal planning files and documents
[COMMAND_EXECUTION]: The skill utilizes standard development tools (git, gh CLI) to perform its intended functions. These tools are used following best practices, such as using specific flags (--repo, --base, --head) to avoid ambiguity and preventing implicit network operations during dry runs.
[DATA_EXPOSURE]: The skill follows a "least privilege" approach to data publication. It instructs the agent to only include information necessary for the reviewer and provides a strict checklist to exclude internal context, session identifiers, and agent reasoning from the final pull request body.

create-pr