S
Supabase skills
SkillThe supabase and supabase-postgres-best-practices skills loaded in the eval. The change is listed per model, including zeros and drops.
Supabase · supabase.com/evals · github.com/supabase/evals · Agent Skills
Change from the baseline
Supabase Evals · Framework eval · 2026-10-10 · 23 tasks × 3 runs
Supabase writes the tasks and grades them. This measures a model together with Supabase, not one framework against another. Six models and 23 tasks. The vendor authors the tasks and grades them.
| # | Model | Baseline | With skills | Change |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 high · Claude Code | 91.3 | 89.9 | −1.4 |
| 2 | Kimi K3 OpenCode | 84.1 | 87.0 | +2.9 |
| 3 | Grok 4.7 high · Grok | 78.3 | 89.9 | +11.6 |
| 4 | GPT 6 Sol medium · Codex | 78.3 | 78.3 | 0 |
| 5 | Claude Sonnet 5 high · Claude Code | 75.4 | 79.7 | +4.3 |
| 6 | GPT 6 Luna medium · Codex | 63.8 | 75.4 | +11.6 |
Skip it
Do not treat the biggest gain as the baseline leader. The baseline is on the Supabase library page.
Often used with
SupabaseLibraryPostgres, auth, and storage. The official eval asks whether a model can finish tasks on this backend.Next.js AGENTS.mdPromptThe bundled Next.js docs in the eval. Each model's change is listed on its own, including zeros.Convex AI rulesPromptThe agent rules in the Convex docs. With Guidelines scores are unverified, so this page does not list a change.