AI Compass / Quick reads
A skill has to prove itself against “without”
That a skill gets loaded doesn't mean it helps. A with-and-without comparison shows it in an hour.
In depth: Does a skill still help? – measuring skills
A skill that helped an older model may now only cost a new one tokens. For this, AI LABS presents Caliper: it writes tests from a prompt and an expected result, runs them several times with and without the skill and shows whether the effort pays off (14 Sept). Meanwhile the Claude Code docs describe the same method as standard: the same tasks in fresh sessions with and without the skill. The official skill-creator plugin compares hit rate, time and tokens in the process (docs checked). It also works by hand: two separate sessions and a learning.md in the skill folder (9 July).
Tool and alternative
skill-creator is free and from Anthropic: /plugin install skill-creator@claude-plugins-official. Caliper is free (MIT, externally checked) and installs according to the README with npx skills@latest add edonadei/caliper. Its skills are now called /grill-skill and /evaluate-skill, in the video still “grill” and “evaluate”. The example in the README (80 to 100%, 30% fewer tokens) is the developer’s own claim.
Sources
- AI LABS, “Insane GitHub Repos That 10x Your Codex And Claude Code Setup”, 14 Sept 2026 (10:19–11:23)
- AI LABS, “All The Types Of Claude Loops Explained In 13 Minutes”, 9 July 2026 (6:00–7:00)
- edonadei/caliper and Claude Code docs Skills, retrieved 17 Oct 2026
Advertising in the source: Manufact (14 Sept), MiniMax with affiliate link (9 July, approx. 3:46–4:50), each plus paid community. “10x” in the title is an exaggeration.
Want to try this with your team? That’s what the AI Lab is for. See the AI Lab