Vibecoding / AI slop / Notes
I'm Rob.
I build the apps on m7kni.io with AI agents. This is where I write about my adventures in vibecoding AI slop, and share hints and tips from the things I learn along the way.
Start here
2 entry points- One agent context is not enoughWhy I run Claude Code, Codex and pi as separate homes in each of two contexts, and publish their rules, hooks and settings from one source repository.
- The agent loop protocolHow I run long agent loops on pi, Codex and Claude Code: a short frozen charter, a living queue from the board, lanes that hand back uncommitted candidates and a root that alone pushes.
Latest articles
3 articlesReviewing the loop with its own history
How I use recorded agent runs to review the loop harness, test changes and decide what to keep. The agents do the analysis; I start the reviews and authorise the changes.
The rules my agents run with
The global rules every Claude Code and Codex session starts with: how they are written, what stays out of them, what the main ones say and how they reach this site.
Running Claude Code with OpenAI models through LiteLLM
Claude Code can keep its tools and hooks while using OpenAI models. The difficult parts are effort routing, prompt caching, parallel-agent affinity and features the gateway cannot supply.
From the scratchpad
Recently updatedExact _redirects rules beat splats
At the Cloudflare edge a catch-all splat can beat a narrower splat written above it, so where an exact rule and a splat overlap, put the exact rule above it and never rely on the order of two splats.