← ClaudeAtlas

using-superpowerslisted

Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions
JFWaskin/superpowers-safe · ★ 1 · Code & Development · score 67
Install: claude install-skill JFWaskin/superpowers-safe
<SUBAGENT-STOP> If you were dispatched as a subagent to execute a specific task, ignore this skill. </SUBAGENT-STOP> <MANDATORY-SAFETY-GATE> **Before invoking ANY other superpowers skill — including brainstorming, subagent-driven-development, executing-plans, dispatching-parallel-agents, writing-plans, test-driven-development, systematic-debugging, finishing-a-development-branch, using-git-worktrees, requesting-code-review, receiving-code-review, verification-before-completion, or any task that runs Bash / writes files / dispatches subagents / runs builds / makes network calls / runs autonomously for more than 5 minutes — you MUST first invoke the `safety-check` skill.** The `safety-check` skill enforces five hard gates (resource budget, command risk scan, loop / spend limits, secret / PII scan, scope confirmation) and a never-override list of destructive operations: no `rm -rf` on system paths, no `dd` to a device, no `mkfs` / `diskutil eraseDisk`, no fork bombs, no `curl | sh` / `wget | bash`, no `git push --force` to `main` / `master`, no `chmod -R 777 /` / `chown -R root /`, no `sudo` without per-command user OK, no `npm publish` / `pip upload` / `cargo publish` without explicit user OK, no writes to macOS system paths (`/System`, `/Library`, `/private`, `~/Library`), no shutdown / kill of system processes. Skipping `safety-check` on a non-trivial task is a hard failure of this skill. Skipping it because "this is simple", "I already know what to do", or "we're on a tig