← The AI Hype Audit — all 100 verdicts
PARTLY "24 things to install in Claude" — real tools, one genuinely-official catch, and 'install counts' nobody can audit
The claim24 things to install (8 plugins + 8 skills + 8 MCP servers) so one command turns Claude into a full dev, finance, legal, or marketing crew — ship a real workflow by lunch. Named with install counts: gstack (125.5k+), superpowers (264.4k+), codex-plugin-cc ('OpenAI's official Codex plugin'), humanizer (32.4k+), the caveman skill ('65% fewer tokens'), agent-browser (39.6k+), and more.
Most of the tools are real, and the creator earns one genuine point: codex-plugin-cc really is published by OpenAI's own GitHub org — an actual official Codex plugin for Claude Code. But two problems keep this at partly. The install counts are unauditable: Anthropic publishes no official install telemetry, so every '264.4k installs' traces to third-party directories whose method is opaque — superpowers' figure suspiciously matches its GitHub star count. And the caveman skill's headline '65% fewer tokens' is a prose-only number: independent testing measured about 8.5% savings on real agentic work, because code and tool-calls (which dominate) are left untouched. More installs also isn't more capability — every skill loads context every session, and one top comment reports a heavyweight plugin turning a 10-minute task into an 8-hour loop.
What holds up
- Most named tools are real: gstack, superpowers, humanizer, agent-browser, caveman all exist as described
- codex-plugin-cc genuinely IS OpenAI-official — published under the openai GitHub org (the creator got this right)
- 'Turn Claude into a specialist crew' is a fair description of what skills/plugins/MCPs do together
What doesn't
- The install counts are third-party guesses — Anthropic publishes none; superpowers' 264k 'installs' ≈ its GitHub star count
- Caveman's '65% fewer tokens' is prose-only — independent agentic testing found ~8.5%, because code and tool-calls dominate and go untouched
- More plugins = more context loaded every session = more tokens per task, not more capability
- Documented failure mode: heavyweight frameworks front-load thousands of tokens (and can loop for hours) before writing a line
- The real funnel is a comment-'CLAUDE' DM capture, not the list itself
The catch
Installing 24 things is the opposite of the goal. Each one taxes every session's context, and the flashy token-savings and install numbers are marketing, not telemetry. The right move is a handful of tools you understand — the 24-count is the hook.
How to actually do it
- Install 2–3 plugins you'll genuinely use this week — not 24; every SKILL.md eats context on every session
- The one worth knowing about from this list: OpenAI's official codex-plugin-cc, if you also use Codex (note it means paying OpenAI too)
- Ignore the install-count numbers entirely — they're unauditable third-party figures, not usage data
- Test any 'token-saving' skill on YOUR real work before believing the percentage — most savings evaporate on code-heavy tasks
- Watch for runaway loops: cap iterations and keep an eye on the first run of any autonomous plugin
How to actually equip Claude without drowning it.
- Effort
- 20 minutes to install a lean, useful set
- Cost
- Free to install; the cost is token burn from over-installing
- Stack
- Claude Code + 2–3 skills you understand + the official Codex plugin only if you use Codex
- Confidence
- High
- Posted by
- an AI-content creator running a comment-'CLAUDE' DM funnel
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.