Guardrail OS — Full Kit

Your AI coding agent moves fast. This is how you stop it from breaking the work.If you run Claude Code, Cursor, or Codex, you've seen it: the agent merges on green tests that never touched the real code path. It patches a bug with a workaround. It refactors five files when you asked for one. Or it just… stops halfway through and says nothing.The model isn't the problem. The agent has no operating discipline — the kind a real engineering team enforces without thinking about it.Guardrail OS gives it one. It's a governance system, not a pile of prompts: three roles with a real authority split, quality gates enforced on every commit, and a rulebook designed to grow itself the first time something breaks.The three roles PM — sets scope, makes the rulings, and verifies the work instead of trusting the report. Supervisor — reviews every change and can reject it. The quality gate. Coder — implements, but can't merge its own work or expand scope. The split is the point. Discipline comes from the structure, not from hoping a prompt holds.Runs in any tool. The three roles can't message each other, so they coordinate through shared files in your repo. You run the PM in one window and the Supervisor — which spawns the Coder — in another, and they hand off through an append-only log. Two windows, one folder, no platform lock-in. Works with Claude Code, Cursor, Codex, or Cowork. Bootstrap prompts are included so you just paste and go.What's enforced A strict priority order: security > accuracy > simplicity > maintainability > performance > ship date, always last. Quality gates every commit — typecheck, tests, build, lint. No shortcuts. Size limits — small files, small commits, small PRs. A security checklist on every mutation — auth, anti-forgery, rate limit, validation, owner-scoping, audit log. Plan → approval → code. No code before a signed-off plan. An escalation protocol with a report-back cadence, so agents don't silently stall. Governance that compounds: a codify-on-first-occurrence mechanism turns each failure into a permanent rule. What's in the box The three role protocols (PM / Supervisor / Coder), fully generalized and stack-agnostic A file-based hub protocol + paste-to-start bootstrap prompts — run the team in any tool, no lock-in Starter hub files: handoff log, status board, decisions, and findings A one-file config to point the whole system at your stack, limits, and gate commands A worked reference adapter (Next.js + Supabase) — copy its shape for any stack Five templates: task assignment, PR review, escalation, incident/findings, ADR Three real case studies showing how each rule was born from an actual production failure README + setup guide Lifetime updates Stack-agnostic. The rules don't care what you build with. Works with any agent that reads markdown instructions.Try before you buy. The free core is on GitHub (MIT): github.com/mwasko23/guardrail-os-lite — this kit is the full three-role team, the tooling, and the case studies.Built the hard way. This system grew out of running a real three-agent team that shipped a production app. The rules are earned, not invented.Digital product — delivered instantly. All sales final.

Found on Devadex — the discovery index for independent software the big search engines bury. More from gumroad.

Report this listing