P2P
Health Gecti
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 17 GitHub stars
Code Basarisiz
- rm -rf — Recursive force deletion command in editions/8.4.7-C/plugin/settings.example.json
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Meta-prompt system that writes prompts: raw idea in — a polished, model-specific prompt out. Cross-model (Claude, Gemini, GPT, DeepSeek, Grok, Qwen, Kimi, GLM), 8-agent QUORUM council, SCOPE.HELM, live model specs. v8.4.7, four editions, Claude Code / Cowork plugin.
P2P — Prompt-to-Prompt
English · Русский
A meta-prompt that writes prompts — and gets the job done. Stream of consciousness in → a polished, model-specific prompt out. The project's goal: eliminate classical prompt engineering for the everyday user.
You shouldn't need to know what Chain-of-Thought is or how to escape XML against prompt-injection. You just describe the task — "I want an expense-tracker app, data from Excel, dark theme" — and P2P handles decomposition, routing, agent selection and hallucination defense under the hood.
What is P2P
P2P (Prompt-to-Prompt) is a modular orchestration system loaded into an LLM that turns it into a prompt-engineering expert. RAG architecture (BASE / LIVE / ON-DEMAND), an 8-agent QUORUM council, the SCOPE.HELM engine for large tasks, and auto-updating Live Specs.
Philosophy: constraints, not pressure. Empirical, not aesthetic.
The project evolved from a single text prompt (v1) into a meta-prompt OS (v8 NEXUS) — see the evolution history.
⚡ Interactive Architecture Map
Full system map: modules, agents, commands, cross-references, data flow — in one D3.js interactive visualization with tabs, filters, animation and RU/EN switch. See how the 8 QUORUM agents, 11 commands and 6 ON-DEMAND modules connect.
🧭 Choose your edition
One architecture, four entry points for different hosts and form factors. Unsure? Pick the one tuned for your main model.
| Edition | For whom | Host | Start |
|---|---|---|---|
| 🟦 8.4.7-C (Claude, внутр. 8C.3) | You work in Claude (Code / Cowork / Projects) | Claude only | ~7K |
| 🟥 8.4.7-H (внутр. 8H.3) | You want maximum / use Grok | 8 hosts (native Grok) | ~60K |
| 🟩 8.4.7-N (внутр. 8N.3) | Your model isn't "native" | any of 8 | ~60K |
| 🟦 8.4.7-L (внутр. 8L.3) | Token economy / context limits / newcomers | universal | ~18K |
📊 Full comparison — editions/COMPARISON.md · 📖 naming guide — NAMING.md.
🚀 Quick start
Option A — plugin (Claude Code / Cowork)
/plugin marketplace add https://github.com/sanic732/P2P-4PDA-edition
/plugin install p2p-v8c3@p2p
Check: /p2p (main menu) · /p2p-teacher (interactive course). Update: /plugin update p2p-v8c3@p2p.
Option B — Chat / Projects / API (any host)
Load the chosen edition's .md files into Project Knowledge (or system prompt) and type start / /p2p. For Gemini you can use NotebookLM (token savings). Details in each edition's INSTALL and docs/INSTALL_GUIDE.md.
Launch triggers: /start · start · старт · /p2p · /menu. Menu not showing? Type full ui menu.
👥 The 8 QUORUM agents
| Agent | Role | When |
|---|---|---|
| 🟣 IRIS | Strategist & Cartographer | Task map, hidden dependencies, the right questions |
| 🟢 TECTON | System Architect | Prompt structure, code architecture, decision trees |
| 🟡 AXIOM | Logician & Verifier | Red-teaming, logic gaps, Confidence Score |
| 🟠 VECTOR | Optimization & Security | Prompt-injection defense, sanitization |
| 🟤 DATOS | Data Analyst | Fact-checking, empirical verification (X Firehose on Grok) |
| ⚫ ANON | Code Specialist / Security | Production-ready code, Stop Conditions (security in 8C.3) |
| 🔵 ARCHITECTON | Integrator | Resolves inter-agent conflicts, UI/UX |
| ☀️ HELIOS | Final Synthesizer | Merges the 7-agent chorus into a clean result |
On Grok (high) agents run natively in parallel (Heavy-16, 5-7× faster); on other hosts a simulated QUORUM is used.
📡 Live Specs
Model prices/quotas/bugs are updated separately (~every 1-2 weeks) from a dedicated Gist (live_specs.md, latest). At startup the system checks web-fetch capability and runs in online-update mode or from an embedded snapshot. Mechanics — in editions/COMPARISON.md.
📚 Documentation
| Section | Inside |
|---|---|
| NAMING.md | Names C/H/N/L/A/G, versions, statuses |
| FAQ.en.md | FAQ: install, hosts, tokens, troubleshooting |
| editions/COMPARISON.md | 4-edition comparison + Live Specs mechanics |
| CHANGELOG.md | Version history (v1 → v8) |
| legacy/HISTORY.md | Project evolution narrative |
| docs/ | Architecture, techniques, PILOT modes, mindmap |
🔬 Scientific Sources & attributions
Integrated ON-DEMAND techniques (RAPTOR, LongRAG, Self-Consistency, MCTS, LLMLingua, OPRO…) are prompting patterns inspired by published research; no third-party code is vendored, the project is MIT. /p2p-karpathy and Template M are inspired by Andrej Karpathy's philosophy. P2P's own mechanisms (QUORUM, SCOPE.HELM, PILOT, ATLAS…) are original. Full list with sources — NOTICE and docs/TECHNIQUES_v8C3.md.
Help & feedback
- Won't launch →
docs/INSTALL_GUIDE.mdorFAQ.en.md - Not sure how to use it →
/p2p-teacherafter install - Bug report / suggestion → Issues or the 4PDA thread
License: MIT (fork & modify; keep NOTICE). Author: sanic732 · 4PDA: Prompt to Prompt 8 NEXUS
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi