Are Supabase docs agent-friendly? We didn't know, so we checked

Supabase is evaluating whether its documentation enables AI coding agents, as well as human users, to complete tasks correctly, beginning with its high-traffic Row Level Security guide, where many readers reported difficulties and errors can expose application data. Using its open-source supabase/evals framework, the team created an independent checklist based on security expertise and external sources, then tasked an agent with configuring access rules for private shared to-do lists and a public weather dashboard without explicitly mentioning RLS. Early tests showed agents granting anonymous users broad write permissions because the guide lacked sufficiently explicit procedural instructions, prompting additions to revoke default grants and deliberately restore access through policies. Although new testing guidance initially failed to improve outcomes, moving test examples directly beside relevant policy code succeeded, illustrating that agents often read documentation selectively rather than sequentially. The project concludes that clear structure, local placement of critical instructions, explicit signposting, and ongoing independent regression testing improve documentation for both agents and people, and Supabase plans to extend these evaluations across more popular guides and publish the RLS scenario as part of its public benchmark.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.