𝐒𝐮𝐩𝐞𝐫𝐩𝐨𝐰𝐞𝐫𝐬 𝐏𝐥𝐮𝐠𝐢𝐧 𝐟𝐨𝐫 𝐂𝐥𝐚𝐮𝐝𝐞 𝐂𝐨𝐝𝐞: 𝐈𝐧𝐬𝐭𝐚𝐥𝐥 𝐚𝐧𝐝 𝐕𝐞𝐫𝐢𝐟𝐲

#claude #agentskills #ai #aie

You installed the Superpowers plugin. But which skills actually fired? And what did it cost you in tokens?

Most AI engineers assume a plugin works because it loaded. That assumption is wrong. Superpowers injects a hook that tells your model to prefer certain skills - but a hook is a paragraph, not a guarantee. You need to prove which skills actually fired in each run, measure the token cost, and decide whether to keep it on everywhere or scope it off.

This guide walks you through installing Superpowers, building a detector that names exactly which skills fired on any Claude Code session, and running a real feature twice - once by hand, once through the plugin - so you see the difference in token cost and behavior. You'll also get a token budget for what the plugin charges every session, whether or not it does anything useful.

What you're really learning here is how plugins work. Superpowers is 14 skills and one hook. The hook is policy-as-code: it injects a paragraph into your model's context on every session start, telling Claude to check for an applicable skill before doing anything. Only the names and short descriptions stay in context. The full skill body loads on demand. If the hook doesn't run, the skills go dormant and your model ignores them. This is why a Windows box without Git Bash gets a plugin that reports itself as enabled and does nothing.

You'll need Claude Code 2.1.228+, Git 2.51.0+, Python 3.13.9+, and pytest 8.4.2+. Allow 45 minutes. The actual Claude API calls cost about 2-3 USD total - skip steps 6, 7, 8, 11, and 12 if you want to avoid the charge, though you'll lose live proof that a skill fired on your machine.

Read the full guide and build the detector yourself:

https://ranjankumar.in/superpowers-plugin-claude-code-setup-and-cost

Follow for more practitioner guides on AI engineering, agents, and Claude Code.