Gemini Skills — Google's Official Skill Repo with Documented 28%→96% Lift

Google's official gemini-skills repository, with a documented case study of a single skill lifting Gemini 3.1 Pro's API-code pass rate from 28% to 96%. The most-cited public win for the skills format.

August 27, 2026
skillsgeminigoogleagentskills-iogoogle-gemini
Gemini Skills — Google's Official Skill Repo with Documented 28%→96% Lift

Gemini Skills

google-gemini/gemini-skills is Google's official skills repo. The most-cited public win for the format: a published Google blog case study showed a single skill lifted Gemini 3.1 Pro's API-code pass rate from 28% to 96% — a 3.4× improvement from packaging the right procedural knowledge into a SKILL.md. It's the cleanest demonstration that skills are not just routing convenience, they measurably move model performance on real tasks.

When to use

  • You work in the Gemini ecosystem and want first-party Google-maintained skills
  • You want a reference for how a large model provider structures skill catalogs
  • The 28%→96% case study is a useful framing when pitching skills internally

When not to use

  • You're not on Gemini — the skills are cross-compatible (agentskills.io) but the case study and tuning are Gemini-specific
  • You want broad domain coverage — this repo is narrower than Anthropic Official Skills or Superpowers

How it works

The repo follows the standard SKILL.md + scripts/ + references/ pattern. Skills are activated by trigger descriptions in the frontmatter and load progressively. The published case study used a skill that bundled API schema documentation, common error → fix mappings, and a verification step — the same pattern works across any model's API coding tasks.

Install

npx skills add google-gemini/gemini-skills

Full SKILL.md

Full source at the canonical repo: github.com/google-gemini/gemini-skills. Embedded copy omitted at launch; the canonical source tracks the latest version.

Verify it yourself

Quick static check — confirm any scripts are doing only the operations their parent SKILL.md claims (API calls, schema parsing). The "official" badge from Google does not exempt the repo from the standard static scan.

Two-model review — paste each SKILL.md you intend to use into two different models with the Vetting Guide prompt. Confirm both return LOW or MEDIUM with no injection class hit.

Review transparency

FieldValue
Last reviewed2026-08-28 (launch baseline)
Reviewerssite baseline (static scan only)
Source repogithub.com/google-gemini/gemini-skills
LicenseApache-2.0
Re-vet triggerupstream change; >30 days since review