This repository ships markdown skills that instruct an AI coding agent, and shell helper scripts those skills can invoke. The realistic risk surface here is:

• A skill's instructions causing an agent to run a destructive or unsafe shell command.
• A shell script under plugins/*/skills/*/scripts/ doing something unsafe (unsanitized input, unsafe temp-file handling, an unpinned/untrusted download).
• Supply-chain risk in bin/, package.json, or a GitHub Actions workflow.
• A manifest (.claude-plugin/marketplace.json, .agents/plugins/marketplace.json) or plugin.json pointing installers at something other than what it claims to.

This repo does not run a server, does not handle user credentials, and does not process end-user data — most classic web-app vulnerability classes don't apply.

Please do not open a public GitHub issue for a security report.

Private vulnerability reporting is enabled on this repository. Use it to report:

1. Go to the repository's Security and quality tab.
2. Click Report a vulnerability to open the advisory form.
3. Fill in a title and description of the issue — a proof-of-concept skill invocation or script input that demonstrates the problem is the most useful thing you can include. Affected version/commit and suggested severity are optional but help triage.
4. Submit. This opens a private draft security advisory visible only to maintainers and repository admins — it does not create a public issue or notify anyone else.

You'll get an acknowledgement, and can continue the conversation privately on the advisory thread, including collaborating on a fix via the temporary private fork GitHub offers from that screen if one is warranted.

If private vulnerability reporting isn't available for any reason, open a regular issue asking a maintainer to reach out for a private channel — don't include exploit details in that issue.

This package doesn't maintain long-lived version branches; security fixes land on the latest published version. If you're on an older tagged release, please update before reporting to confirm the issue still reproduces.

We'll credit reporters (unless you ask not to be) once a fix is published. There's no bug bounty — this is an open-source marketplace, not a funded program.