From skull
The guardrail switchboard. Triggers on "which guardrails are on?", "turn on / turn off <guardrail>", "set up guardrails for this project", or when a guardrail keeps firing and the developer wants it quieter. Records the choice in .claude/skull.json so it survives the session; the leader honours it.
How this skill is triggered — by the user, by Claude, or both
Slash command
/skull:guardian-suiteThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
The switchboard for the guardrail suite. **The switch is a config file, not a command** — there is nothing to
The switchboard for the guardrail suite. The switch is a config file, not a command — there is nothing to install and nothing running in the background.
guardian (scope + verification) · supplyguard (dependencies) · testmedic (flaky tests) ·
debtradar (churn × complexity) · compactor (long sessions) · cap-tdd (red-green-refactor).
Adjacent, and always on when installed: sentinel (the cartographer agent).
Per project, .claude/skull.json:
{ "defaultGuardrails": ["guardian", "supplyguard"] }
Absent the file, the leader picks a minimal set for the project and says which. Naming a guardrail here makes it the standing default so nobody has to re-ask each session. To silence one, remove it and say so out loud — a guardrail that was quietly dropped is worse than one that was never on.
When asked which are on, answer from the file plus what this session actually invoked — not from what the category README lists. If the file is absent, say "no standing config; the leader chose these for this session" rather than implying persistence that isn't there.
These are prose disciplines the model follows, not interception. Nothing in this suite can block a tool call,
survive a /clear on its own, or catch a rule the model talks itself out of. A rule that must hold every time
belongs in a PreToolUse hook or a permissions.deny rule in settings — both are the developer's to install,
deliberately. Say so plainly when someone asks whether a guardrail will "stop" something.
Pairs with workspace-architect (what this project's .claude/ should hold). The leader reads the same file.
npx claudepluginhub aturzone/skullManages repo-root guardrail documents (CHARTER.md, ARCHITECTURE.md, TESTING.md, SECURITY.md) for AI agents, defining authority hierarchy and consulting rules. Use to bootstrap, draft, or understand guardrail docs.
Scaffolds a new greenfield production app with a uniform 7-gate guardrail system (lint, typecheck, test, coverage, build, e2e, audit) baked in from commit #1. Supports Next.js, NestJS, Django, Go, Rust, and Spring Boot.
Displays active ThumbGate prevention rules, reliability rules, and originating lessons. Shows what guardrails protect the project and why each exists. Uses MCP tools or CLI.