Framework Version: 2.2
Level 5: Agent-First Engineering — March 2026 From intent to deployed product in one day. Every time. Mechanical enforcement. Parallel execution. Self-verifying agents.
- Claude Code — the agentic coding engine
- AionUI — orchestration GUI for parallel agent sessions
- Node.js 20+ — runtime
- Git — version control
- Coolify or Vercel — deployment platform
git clone https://github.com/stringztechnologies/stringz-framework.git
cd stringz-framework
./setup.shThat's it. Open Claude Code in any project and the workflow activates automatically.
To update later: ./update.sh
Windows users: use .\setup.ps1 instead of ./setup.sh
mkdir my-project && cd my-project && claude
Claude detects no CLAUDE.md, runs First Contact when you ask it to build something, and flows into the 6-phase workflow.
Read WORKFLOW.md — The complete 6-phase agent-first framework.
| Phase | Deliverable | Agents Used | Time |
|---|---|---|---|
| 1. Specify | SPEC.md | None (human + Claude chat) | 30-60min |
| 2. Architect | CLAUDE.md + schema + REVIEW.md + lint | repo-scorer | 30-60min |
| 3. Implement | Working code (wave commits) | architecture-enforcer, test-writer, decision-recorder | 2-8hrs |
| 4. Verify + Deploy | Live URL, zero P0s | visual-verifier, security-auditor + Comet QA | 1-2hrs |
| 5. Brand | Brand-aligned UI | brand-aligner + Comet comparison | 30-60min |
| 6. Deliver | Client says yes | repo-scorer, onboarding-writer, maintenance-scanner | 30-60min |
| Level | Name | Description |
|---|---|---|
| 1 | Vibe Coding | Prompt and pray. No structure, no persistence. |
| 2 | Structured Prompting | CLAUDE.md, SPEC.md. Context persists across sessions. |
| 3 | Workflow Engineering | 6-phase process. Templates. Skills. Wave-checkpoint rhythm. |
| 4 | Multi-Agent | Parallel execution. Builder + Tester + Scout. Subagent delegation. |
| 5 | Agent-First Engineering | Mechanical enforcement. Self-verifying agents. ADRs. Repo scoring. Full autonomy with guardrails. |
Copy these into every new project:
templates/SPEC.md.template— Interview-driven specificationtemplates/CLAUDE.md.template— Project identity + conventionstemplates/KNOWLEDGE.md.template— Accumulated learning journaltemplates/TASKS.md.template— Wave-based task trackertemplates/REVIEW.md.template— Review-specific rules (separate from build rules)templates/PREFLIGHT.md.template— Pre-implementation environment & project checklisttemplates/STRINGZ.md.template— Personal context — tells every Claude session who you are and how you worktemplates/ONBOARDING.md.template— Onboard existing projects to the Stringz Workflow
Reusable domain knowledge that compounds across projects:
| Skill | Domain | Use When |
|---|---|---|
supabase-nextjs-fullstack |
Full-stack | Next.js + Supabase project |
multi-currency-ledger |
Finance | Payments in multiple currencies |
property-management-core |
Real Estate | Buildings, units, tenants, leases |
notification-queue |
Messaging | Scheduled reminders, multi-channel |
mobile-first-dashboard |
UI/UX | Phone-first admin dashboard |
fullstack-agent-squad |
Dev Process | Agent orchestration patterns |
cloudflare-site-crawler |
Design | Extract design tokens from websites |
Install in any project by copying to .claude/agents/:
| Agent | Purpose | Phase |
|---|---|---|
architecture-enforcer |
Mechanical rule checking against CLAUDE.md | 3 (each wave) |
test-writer |
Generate tests in parallel with builder | 3 (parallel) |
code-reviewer |
Post-implementation quality review | 3 (checkpoint) |
decision-recorder |
Auto-generate Architecture Decision Records | 3 (checkpoint) |
visual-verifier |
Boot app, check routes, report errors | 4 (pre-deploy) |
security-auditor |
Security vulnerability scan | 4 (pre-deploy) |
brand-aligner |
Generate brand alignment prompts | 5 |
repo-scorer |
7-metric legibility scorecard (70-point scale) | 2 + 6 |
maintenance-scanner |
Tech debt detection | 6 + weekly |
onboarding-writer |
Generate setup guides and docs | 6 (delivery) |
qa-prompt-generator |
Auto-fill QA prompts from CLAUDE.md | 4 (pre-QA) |
git clone https://github.com/stringztechnologies/stringz-framework.git
cd stringz-framework
./setup.shThe script checks prerequisites, copies STRINGZ.md to ~/.claude/, symlinks all skills and agents, and prints a summary. Edit ~/.claude/STRINGZ.md with your personal details after running.
# macOS
brew install aionui
# Windows/Linux — download from GitHub releases
# https://github.com/iOfficeAI/AionUi/releasesAionUI auto-detects Claude Code and provides a visual workspace for running parallel agent sessions.
Templates are copied automatically when Claude runs First Contact on a new project. To copy manually:
cp stringz-framework/templates/CLAUDE.md.template ./CLAUDE.md
cp stringz-framework/templates/KNOWLEDGE.md.template ./KNOWLEDGE.md
cp stringz-framework/templates/TASKS.md.template ./TASKS.md
cp stringz-framework/templates/REVIEW.md.template ./REVIEW.mdSee CONTRIBUTING.md for how to add skills, agents, and templates. Team members run ./update.sh after merges to pick up new additions.
You are not a coder. You are a context engineer. Your job is to design environments where agents produce reliable output.
Stringz Technologies — Addis Ababa & Beyond — 2026