A file-driven operating system for making OpenClaw generate production-grade code with less cleanup debt.
This pack is built for builders using OpenClaw across:
- React and Next.js apps
- APIs and backend jobs
- agent workflows and runtime-heavy systems
AI coding often looks fast at first and expensive later.
The goal of this pack is to force performance, architecture, and verification into the process before implementation starts.
Instead of asking OpenClaw to "write optimized code", you:
- define budgets
- define architecture constraints
- externalize rules into durable files
- verify each implementation loop
Official OpenClaw pattern:
AGENTS.mdTOOLS.mdstyle workspace notes- file-driven workspace behavior
- context inspection and trimming
- ACP handoff to external coding harnesses
Custom framework in this repo:
ENGINEERING_PRINCIPLES.mdPERFORMANCE_SPEC.template.mdAPI_PERFORMANCE_SPEC.template.mdAGENT_RUNTIME_SPEC.template.md- prompts and verification scripts
- Copy the relevant spec template and tailor it to your feature or system
- Update
AGENTS.mdand your localTOOLS.md - Ask OpenClaw to read:
AGENTS.mdENGINEERING_PRINCIPLES.md- the relevant spec file
TOOLS.mdif environment details matter
- Have it produce a blueprint first
- Implement in small loops
- Run verification
- Write durable lessons back into your workspace
Use:
PERFORMANCE_SPEC.template.mdREACT_PERFORMANCE_RULES.mdprompts/build-performance-first-feature.mdprompts/optimize-react-render-path.md
Use:
API_PERFORMANCE_SPEC.template.mdprompts/build-performance-first-api.md
Use:
AGENT_RUNTIME_SPEC.template.mdprompts/audit-openclaw-agent-runtime.md
Read:
README.quickstart.mddocs/quickstart-matrix.md
This repo is safe to share publicly if you keep it sanitized.
Do not publish your real:
MEMORY.mdmemory/USER.md- internal hosts
- SSH targets
- API keys
- production URLs
- customer data