Plugin
nagisanzenin-engram
Evidence-based learning engine — first-principles curricula, free-recall verification with receipts, FSRS-scheduled memory, and explorable artifacts. Learn anything; keep it.
3 skills3 agents1 hook+2+11.5k
@nagisanzeninAgent Orchestration
Plugin
production-grade
14 AI agents, one install, idea to production.
Plugin Signal65/100Solid
maintained71
capable59
documented75
adopted48
Install
$ npx -y skills add nagisanzenin/production-grade --agent claude-code
Stats
181
Stars
50
Forks
Maintained
Maintenance
JavaScript
Language
2mo ago
Last commit
7mo ago
Created
2h ago
Added
What's inside
Skill(14 skills)
Badges
README
Production Grade Plugin
/plugin marketplace add nagisanzenin/claude-code-plugins
/plugin install production-grade@nagisanzenin
New in v5.6 — one repo, four platforms. The same 14 skills now install natively on Claude Code, OpenAI Codex, OpenCode, and Pi — one shared core, thin per-platform glue, honest per-platform status. Support matrix below.
New in v5.5 — The Loop Engine. Agents stop claiming "done" and start proving it. Every stage now loops against a check it can't argue with — typecheck, tests, or the running app clicking its own buttons — until it converges. How it works.
Built With This Plugin
Built something with this plugin? Open a PR to add your project here.
| Project | Live | Description |
|---|---|---|
| PingBase | pingbasez.vercel.app | Free uptime monitoring — get emailed when your website goes down. GitHub OAuth, Stripe billing, Turso DB. |
| LLM Matrix Arena | llm-matrix.vercel.app | Browse and compare LLM models across N dimensions. Community-driven voting from real developers — not benchmarks, real opinions. |
| SkyClaw | github.com/nagisanzenin/skyclaw | Cloud-native Rust AI agent runtime. Telegram-native — deploy one binary, paste your API key, and control your server through chat. |
Release Timeline
2026-08-19 v5.6 ●━━━ Omni-platform — native installs on Codex, OpenCode, Pi; platform-adaptation contract; pi payload-harness 14/14
│
2026-07-02 v5.5.2 ●━━━ Loop-engine robustness — resilient report-artifact writes + orchestrator recovery; deep-smoke requires an adversarial input
│
2026-07-02 v5.5.1 ●━━━ Loop-engine hardening — trustworthy BUILD-exit deep-smoke gate, worktree-safe scoped oracle-gate, brownfield toolchain adoption
│
2026-07-02 v5.5 ●━━━ The Loop Engine — oracle-driven iteration: edit-gate hook, TDD pair, functional drive replaces hand-testing
│
2026-03-07 v5.4 ●━━━ Harmonization — mode-aware autonomy, cross-session enforcement, agent skill loading
│
2026-03-07 v5.3 ●━━━ Worktree isolation, self-healing gates, cost dashboard
│
2026-03-07 v5.2 ●━━━ Frontend overhaul — functional-first, design polish, 4 visual style presets
│
2026-03-07 v5.1 ●━━━ Boundary Safety — 6 patterns for system boundary bugs, from real deployment
│
2026-03-06 v5.0 ●━━━ Verified & Resilient — receipt enforcement, re-anchoring, adversarial review
│
2026-03-06 v4.4 ●━━━ Freshness protocol — agents WebSearch to verify volatile data before implementing
│
2026-03-06 v4.3 ●━━━ Visual identity, pipeline dashboard, gate ceremonies
│
2026-03-06 v4.2 ●━━━ Adaptive routing, 10 execution modes, everyday SWE work
│
2026-03-05 v4.1 ●━━━ Engagement modes, scale-driven architecture, adaptive interviews
│
2026-03-04 v4.0 ●━━━ Two-wave parallelism, internal skill agents, dynamic task generation
│
2026-03-04 v3.3 ●━━━ Brownfield-safe — works on existing codebases
│
2026-03-03 v3.2 ●━━━ Auto-update, MECE intent routing, protocol crash fix
│
2026-03-02 v3.1 ●━━━ Polymath co-pilot — the 14th skill
│
2026-03-01 v3.0 ●━━━ Full rewrite — Teams/TaskList, 7 parallel points, shared protocols
│
2026-02-28 v2.0 ●━━━ 13 bundled skills, unified workspace, prescriptive UX
│
2026-02-24 v1.0 ●━━━ Initial release — autonomous DEFINE>BUILD>HARDEN>SHIP>SUSTAIN
The Pipeline
YOU ──→ "Build a SaaS for ..."
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ DEFINE │
│ T1 Product Manager ─── BRD, user stories, acceptance criteria│
│ T2 Solution Architect ─ ADRs, API contracts, data models │
│ │
│ ┌─────────────┐ ┌──────────────┐ │
│ │ GATE 1 │ │ GATE 2 │ │
│ │ Requirements│ │ Architecture │ │
│ └─────────────┘ └──────────────┘ │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ BUILD + ANALYZE (Wave A — parallel) │
│ │
│ Backend ──── N agents (1 per service) QA ──── test plan │
│ Frontend ─── N agents (1 per page) Security ── STRIDE │
│ DevOps ──── Dockerfiles + CI skeleton Review ── checklist │
│ SRE ───── SLOs │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ HARDEN (Wave B — parallel against code) │
│ │
│ QA ─────── unit / integration / e2e / performance tests │
│ Security ── code audit + dependency scan (4 parallel phases) │
│ Review ──── arch / quality / performance (adversarial) │
│ DevOps ──── build + push containers │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ SHIP │
│ │
│ DevOps ── IaC + CI/CD ┐ │
│ Remediation ───────────┘ parallel ┌──────────────┐ │
│ SRE ── chaos + capacity ┐ │ GATE 3 │ │
│ Data Scientist ─────────┘ parallel │ Production │ │
│ │ Readiness │ │
│ └──────────────┘ │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ SUSTAIN │
│ │
│ Technical Writer ── API ref + dev guides (parallel) │
│ Skill Maker ─────── 3-5 project-specific reusable skills │
│ Compound Learning ── pipeline insights for next run │
└─────────────────────────────────────────────────────────────────┘
│
▼
DONE ── receipts verified, agents cleaned up
3 gates. 2 waves. 10+ parallel execution points. ~3x faster than sequential. Every stage loops against an executable oracle until it converges — The Loop Engine.
The Loop Engine
New in v5.5 — the reason the agents stop lying about "done."
Software is never one pass. It's write, run, fail, fix, run again, until a check passes. Through v5.4 the pipeline was mostly feed-forward: each agent did its job, handed the artifact down, and correctness leaned on human review at the gates. v5.5 makes iteration the main path, on a single rule:
A task is done when a check it cannot argue with says so — not when the agent claims it.
That check is an oracle: a compiler, a type checker, a test suite, or the running app clicking its own buttons. No oracle, no loop.
┌──────────────┐
│ PRODUCE │ agent writes / fixes
└──────┬───────┘
▼
┌──────────────┐ red ┌────────────────┐
│ ORACLE │ ───────▶ │ DELTA BACK │
│ test·type·app│ │ failing output │
└──────┬───────┘ │ only → re-loop │
green │ └────────┬───────┘
▼ │
┌──────────────┐ no progress ×2 │
│ CONVERGED │ ◀─────────────────┘
└──────────────┘ plateau → escalate strategy, not effort
What it adds, concretely:
- Oracle bootstrap. Before any parallel wave, the pipeline writes two check scripts for your project: a fast one (typecheck + lint, under 15s) and a full one (tests + build + boot smoke).
- Enforced by a hook, not a prompt. A
PostToolUsehook runs the fast oracle after every file edit. Break something and the error lands back in the agent's lap immediately, not at the end. - Separated duties. QA writes failing tests first and owns the
tests/folder. A coding agent that weakens a test to go green gets flagged with a Critical finding. No grading their own homework. - Functional drive. Before the final gate, an agent boots the real app and drives it — every button, every form, every link. A control that renders but does nothing is a Critical bug. This is the hand-testing replacement.
- Convergence, not retries. Each loop tracks a number (failing tests, open findings) and stops when it hits zero or stops improving, shows the trend, then escalates. No retrying the same fix five times.
Fully backward compatible: a pipeline that never needs to loop runs exactly as before. Full design rationale in docs/LOOPS.md.
10 Execution Modes
Not just full builds. The orchestrator reads your request and routes automatically.
┌──────────────────────────────────────────────────────────────┐
│ │
│ Full Build ████████████████████████ all 14 agents │
│ Feature █████████████ PM+Arch+Eng+QA │
│ Harden ████████ Sec+QA+Review │
│ Ship ██████ DevOps+SRE+DS │
│ Architect ████ Solution Architect │
│ Test ███ QA Engineer │
│ Review ███ Code Reviewer │
│ Document ███ Technical Writer │
│ Optimize █████ SWE+Data Scientist │
│ Explore ███ Polymath │
│ │
└──────────────────────────────────────────────────────────────┘
File tree
More plugins by @nagisanzenin.
View @nagisanzenin’s profile@nagisanzeninFAQ
production-grade is a Claude Code plugin with 14 hand-picked skills, indexed on Flowy in the Agent Orchestration category. Install it with the command on its page. It includes code-reviewer, data-scientist, devops. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
- What is production-grade?
- 14 AI agents, one install, idea to production.
- How do I install production-grade?
- Run these in Claude Code: npx -y skills add nagisanzenin/production-grade --agent claude-code. Then prompt normally.
- Does production-grade auto-invoke its skills?
- Not yet. It is indexed on Flowy as a plain plugin. Request auto-invocation on its page and Flowy will route its skills for you as you prompt.
- Is production-grade free?
- Yes. Flowy is free and open source, with nothing gated. You can read every skill in full before you install.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it