From zero to an AI agent's push being blocked in about ten minutes. nimblegate's docs now live on the site: a short quick start, guides for tokens, agents and MCP, and a page for every check it runs.

Two AI agents, two branches, both green. They still collide at merge, because both edited the same file an hour ago. New guide: see overlapping branches while the agents are still working, and send it to whatever runs them as a webhook.

Two agents on one repo, each on its own branch. Both changed src/checkout/total.js, and normally you'd find out at merge time. nimblegate compares open branches on every push and shows the overlap while the agents are still working.

nimblegate 0.6.0: when a coding agent can't get the tests to pass, it sometimes takes a shortcut. Skips the failing test, tells CI to ignore the failure, turns the linter off. Six new checks flag those on every push. They warn, never block.

Nice surprise: a detailed write-up of nimblegate on tinyash.com. It explains why the check sits on a separate gateway instead of a local hook, so the agent's machine never holds the upstream token. Thanks for writing it up! 感谢 tinyash 写了这么详细的介绍！

Running several AI coding agents on one repo? nimblegate 0.5.0 now tells you when two open branches change the same files, while the agents are still working, not at merge time. Self-hosted git gateway, nothing phones home.

The Next Trend After Agent Looping Most agents today rely on simple looping. The emerging standard is this: Looping + Graphic Chart + Guardrail This keeps looping power while adding real safety. github.com/nimblegate/n... #AIAgents #AgentWorkflow

nimblegate is on show hn today i let AI agents push to my repos and got tired of checking every diff for basic mistakes like test credentials. so i built a self-hosted gate: agents push to it, it checks against your rules, relays clean ones and bounces mistakes back to fix.