This repository contains Agent Skills: Markdown instruction files (SKILL.md) plus a small number of shell helper scripts under scripts/. The SKILL.md files are not executable code — they are prompts read by an AI coding agent (Claude, or another agent that supports the Agent Skills / AGENTS.md convention). Installing a skill does not run any code by itself; the risk surface is what the agent does when it later follows the instructions in that file, using whatever tools and permissions the agent already has (shell, git, gh, file edits, etc.).

In short: treat a SKILL.md the same way you would treat a code review checklist or a runbook handed to a new engineer — advisory text that a sufficiently privileged actor (here, an AI agent) can choose to act on. It is not a binary, a package with a build step, or something that executes on npm install / pip install.

• Everything under scripts/ (install-user.sh, list-skills.sh, link-skills.sh, test-install-user.sh, validate-plugin.py) is real, executable shell/Python. Review any script before running it, as you would any script from the internet. install-user.sh writes/backs up files under $HOME (or --target-home) and supports --dry-run to preview changes without writing anything. test-install-user.sh and validate-plugin.py are dev/CI-only checks with no network or destructive action.
• skills/engineering/subagent-pipeline/SKILL.md instructs the agent to run gh CLI commands (gh pr review, gh api .../pulls/.../comments, gh pr merge) against GitHub on the agent's behalf, as part of an implement -> review -> fix -> merge workflow. These are authenticated GitHub operations, not arbitrary remote code execution, but they do mean an agent following this skill can open PRs, post review comments, and merge to a repository it has access to.
• skills/engineering/ai-agent-pr-metadata/SKILL.md instructs the agent to edit PR templates and post or wrap GitHub PR comments/reviews with AI agent, model, and run metadata. This uses authenticated gh operations and should not add AI attribution to commit messages.
• No SKILL.md in this repository instructs fetching remote URLs (curl, wget, HTTP fetch) as part of its documented workflow. If you find one that does in a future revision, please report it (see below) so this file can be updated.

Please report security concerns by opening a GitHub issue in this repository. Use the security label if available. For concerns you would prefer not to disclose publicly before a fix lands, mention that in the issue title (e.g. "SECURITY: ...") and keep exploit details out of the initial issue body — a maintainer will follow up on how to share specifics.

This is a solo-maintained, community-facing skills repository with no dedicated security contact address; GitHub issues are the fastest and most reliable way to reach the maintainer.

This policy covers the contents of this repository only: the SKILL.md files, AGENTS.md, CLAUDE.md, AI_ENGINEERING_WORKFLOW.md, system-level/core.md, and scripts/. It does not cover the security posture of any AI agent or CLI tool that installs or runs these skills (e.g. Claude Code, other Agent Skills clients) — report issues with those tools to their own maintainers.