Claude’s official security plugin
Three lines before something goes wrong: agents sweep the repo, three more try to knock each finding down, two must agree.
2.9K saves on X, as of Posted
/plugin install claude-security@claude-plugins-official
/reload-plugins
/claude-security scan the whole repo at highNext Switch to Opus 5.5 with /model, then send them in Claude Code.
I’m begging you: if you write code, run Claude’s official security plugin.
It sends a team of agents through your whole repo. Every finding goes to three more agents whose only job is to knock it down, and it only makes the report if two of them agree.
On a big repo it can take tens of minutes to a few hours. It might find nothing, and it burns usage either way. Still beats saying “if only we’d run it” after something goes wrong.
It really is three lines.
It asks once whether to go ahead. Say yes. Anything marked CRITICAL or HIGH in the report, fix it.
If the install fails, run /plugin marketplace add anthropics/claude-plugins-official first. If it says the Workflow tool is missing, turn on Dynamic workflows in /config and restart. You’ll need Python 3.9 or later. Keep Claude Code open while it runs, and don’t switch branches.
When it’s done, run /claude-security again and pick Suggest patches. Each finding becomes a patch file, and you decide which ones to apply.
Share this tip
