System prompts, agent behaviors, and workflow specifications for ACP v4.0.
ACP uses 15+ specialized prompts that work together:
- Agent Prompts: Conclave Mode, Collective Coding, Consensus Calculation, etc.
- Tool Descriptions: Oracle verification, axiom lookup, D-score calculation
- Workflows: 7-level spiral, musical structures (Fugue/Sonata/Concert)
- System Reminders: Dataset references, oracle verification
Transparency. We show exactly HOW ACP achieves consensus. No black boxes — verifiable, reproducible truth-finding through axioms.
Some prompts contain interpolated variables (dataset paths, oracle endpoints, axiom levels).
- Conclave Mode (~800 tks) — Model independence through isolation
- Collective Coding (~950 tks) — 3 models writing code together
- Consensus Calculator (~1200 tks) — D-score calculation via axioms
- Dataset Verification (~650 tks) — Axiom quality verification
- Axiom Spiral Orchestrator (~1100 tks) — 7 levels with φ-convergence
- Oracle Verification (~450 tks)
- Axiom Lookup (~320 tks)
- D-Score Calculator (~580 tks)
- Musical Structure Selector (~400 tks)
- 7-Level Axiom Spiral (~700 tks)
- Musical Structures (~850 tks)
- Dataset Reference (~150 tks)
- Oracle Verification (~180 tks)
ACP-PROMPTS is part of a three-repository ecosystem:
| Repository | Role | License |
|---|---|---|
| ACP-PROJECT | Implementation — code that executes | BSL 1.1 |
| ACP-PROMPTS | Instructions — prompts that guide AI | BSL 1.1 |
| ACP-DATASETS | Data — 1,059 verified axioms | ACP-VERIFIED-1.0 |
User Query
→ ACP-PROJECT (executes code)
→ ACP-PROMPTS (follows instructions)
→ ACP-DATASETS (verifies with data)
→ Consensus + proof
Repositories are currently private. Source available soon.
We welcome contributions. See CONTRIBUTING.md.
Licensed under BSL 1.1 — see LICENSE. Non-production use always permitted. Transitions to Apache 2.0 on 2030-04-03. Commercial licensing: enterprise@axiomprotocol.org
ACP Protocol v4.0 — Copyright 2025-2026 ACP Research Initiative