← The AI Hype Audit — all 645 verdicts

REAL

Holds up: piling on Claude Code skills really can cost you correct picks, and Anthropic's own docs explain why, though the reel's exact scores are its own unaudited test

Facebook · Nov 2026

The claim'145 Claude skills loaded. But can Claude actually pick the right one?' An animated test in Claude Code (Opus 5.5) shows a skill picked correctly 21 of 28 times when loaded alone versus 14 of 28 with all 145 loaded, with only about 100 of the 145 reaching the model as bare names. Lesson: clear descriptions, fewer loaded skills, name the skill in the prompt, or use a skill router.

The mechanism in this reel is documented by Anthropic. Claude Code puts a listing of skill names and descriptions into context so Claude can choose one. Its docs say the listing always contains every name, but with many skills installed it drops descriptions to fit a character budget of about 1% of the model's context window, starting with the skills you use least. A skill that arrives as a bare name loses the keywords Claude matches on, which is exactly what the reel's 'names only, about 100 of 145' panel shows. The docs also confirm the fixes the reel recommends: invoke a skill directly with /skill-name, write specific descriptions with the key use case first, trim unused skills (there is a /skill-doctor command for that), mark low-priority skills name-only, or raise the budget with the skillListingBudgetFraction setting. What we cannot verify is the scoreboard itself (21 of 28 alone versus 14 of 28 with all 145, 56 runs). It is an animated graphic from the poster's own test with no published data, and results will vary with which skills you have. The 'skill router' idea is reasonable but is the poster's pattern, not a built-in feature. The caption sells nothing; it is a save-and-comment engagement post.

What holds up

  • Read Anthropic's Claude Code skills docs (code.claude.com/docs/en/skills), section 'Skill descriptions are cut short': descriptions are dropped to fit a budget of about 1% of context, least-used first.
  • Confirmed /skill-name direct invocation, the 1,536-character per-skill cap, skillListingBudgetFraction and SLASH_COMMAND_TOOL_CHAR_BUDGET in the same docs.
  • Viewed the reel's contact sheet: an animated scoreboard (Opus 5.5, Claude Code 2.1.312, 56 runs) with no raw data or repo link.
  • No independent replication of the exact scores found.

What doesn’t

  • Exact pick rates are the poster's own unpublished test.
  • The 'skill router' is a do-it-yourself pattern, not a Claude Code feature.
  • Save-this and comment engagement hooks.

The catch

The lesson is right and backed by the vendor's docs. The precise scores are illustrative, so test your own setup rather than quoting them.

How to actually do it

  • Run /doctor and /context in Claude Code to see how big your skills listing is and whether it is over budget.
  • Run /skill-doctor and disable skills you never use; set rarely used ones to name-only in skillOverrides.
  • Rewrite each important skill's description to lead with its main use case and the words people actually type.
  • For must-run skills, start the message with /skill-name instead of hoping Claude picks it.

More skills is not smarter. Fewer, better-described skills, and naming the one you want, is the documented route to reliable picks.

Confidence
Medium
Posted by
an AI tips page (about 8,000 plays); reel sent to Buddy Sun Nov 15, 2026, 12:21 AM

See the original claim →

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.