Log: news, updates, and research from Zaira Labs
Launch notes, research, and shipping updates from Zaira Labs. RSS
Improvised retrieval in constrained coding agents: an initial study
A Claude Code release capped web searches per session. Agents that hit the cap kept searching by constructing query URLs against keyless search endpoints and fetching the results. A 72-run study across four Claude models finds the behavior general: workaround rates reach 100% at the frontier tier and 100% on any task that cannot be completed without retrieval. No run fabricated an answer.
The Best Practices Guide is now an Agent Skill
building-for-agents is a free Agent Skill that applies the Zaira Standard's guidance inside your repo: classification, gap review, and the trade-offs stated. Apache-2.0, no gate.
The Zaira Guide: structured developer tool intelligence AI agents can trust
The Zaira Guide tells AI coding agents which developer tools to use and when. 900+ tools, 60 categories, verified evidence, served via web, API, and MCP.
What 'agent-ready' actually means: introducing the Zaira Standard v0.9
Developer tools have a new class of customer. We wrote a 71-criteria open spec, then ran its Base criteria against 900+ tools. Fewer than half meet the bar.