I Cut My AGENTS.md Files by 50% Using an Eval
How I evaluated one global and four project AGENTS.md files, removed about half the instructions, and checked what still changed agent behavior.
I write, ship and sell software products. Indie. Writing code in Swift, TypeScript and Ruby. Based in Singapore, working remotely. This is my blog.
How I evaluated one global and four project AGENTS.md files, removed about half the instructions, and checked what still changed agent behavior.
Why I run small products on one Hetzner server with Kamal, and how Agent Control serializes deployments when coding agents are doing the work.
Why I built Agent Control, my macOS app and CLI for coordinating coding-agent sessions, queues, callbacks, loops, and cross-project work.
AI-built SaaS apps can work, but the expensive cleanup starts when nobody supervises the system design, data model, auth boundaries, and production workflows.
Cloudflare Tunnel changed local integration testing for me because it gave local services stable HTTPS names that behave like the real app boundary.
Webhooks feel hard locally because you are debugging two machines, a public URL, raw request bytes, signatures, retries, and your own handler at the same time.
How I debug browser security features that look like broken code: CORS, SameSite cookies, mixed content, CSP, iframe permissions, and isolation headers.
CORS errors usually mean the browser is enforcing the API contract your server declared. Fix the contract instead of turning the check off.
Browsers enforce useful security rules, but your backend still has to authenticate, authorize, validate, rate limit, and reject direct API calls.
How I decide whether a coding-agent workflow belongs in a skill, slash command, MCP server, or plain CLI script.