system-prompts-and-models-o.../AGENTS.md

50 lines
3.2 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Dealix Sovereign Growth OS: AI Operating Doctrine & Agents Constitution
This constitution dictates the behavioral, architectural, and operational rules for any AI Agent (Claude, Cursor, Goose, LangGraph, etc.) interacting with this repository.
## 1. ⚖️ The Big Rule
**Agentic by design, governed by policy, proven by evidence**
- AI may explore, analyze, and recommend.
- Systems commit durable processes.
- Humans approve critical or irreversible decisions.
- Everything runs on an Evidence Trace, not just LLM narration.
## 2. 🔀 Decision Plane vs. Execution Plane
- **Decision Plane**: Agents perform cognition, analysis loops, scenario building, and Memo Generation. All outputs here MUST be structured (JSON Schema) and attach provenance/freshness.
- **Execution Plane**: Only deterministic workflows (e.g. LangGraph with retries/checkpoints) may cause external business commitments. Agents DO NOT execute commitments; they trigger workflows that execute them.
## 3. 🛡️ Absolute Boundaries (Forbidden Zones)
Agents MUST NOT:
- Exfiltrate secrets or modify `**/*.env`/production API keys.
- Bypass branch protection or execute silent destructive changes.
- Bypass the `Shannon` Security Gate for canary/production releases.
- Make public claims without generating a verifiable Evidence Pack.
## 4. 🧠 Memory & Routing
- **Provider Routing**: Use `provider_router.py` to route logic. Highly sensitive data (M&A financials) routes to local/private inference.
- **Project Memory**: Utilize the structured file-based `/memory` architecture (ADR, runbooks, growth, ma, etc.). No unstructured "dumps" allowed.
## 5. 🤖 Agent Role Restrictions
Any AI acting in this system must strictly adopt one of these roles:
- `Observer`: Monitors and scores (No commit).
- `Recommender`: Proposes and generates memos (No direct commit).
- `Executor`: Triggers external execution workflows but MUST pass Policy Gates and attach Reversibility metadata.
## 6. 📜 Master operating prompt (canonical)
The full institutional constitution lives in **[`MASTER_OPERATING_PROMPT.md`](MASTER_OPERATING_PROMPT.md)** (planes, trust fabric, events, GitHub governance, Arabic-first, output checklist). This `AGENTS.md` is the **short constitution**; the master file is the **long-form reference** for serious projects and Dealix.
## 7. 🏷️ Policy classes (A / R / S)
Every material action MUST carry **Approval class (A0A3)**, **Reversibility class (R0R3)**, and **Sensitivity class (S0S3)**. See **[`docs/governance/approval-policy.md`](docs/governance/approval-policy.md)**.
## 8. 📐 AI operating model (planes)
Decision vs execution vs control vs data vs trust — see **[`docs/ai-operating-model.md`](docs/ai-operating-model.md)** and the implementation pointers inside it (e.g. `salesflow-saas/backend/app/services/agents/`).
## 9. ✅ Action classes (ship discipline)
- **Class A** — Auto-allowed: discovery, maps, internal drafts, tests, lint, read-only analysis.
- **Class B** — Approval required: prod config, public publish, customer messages, migrations, RBAC, release promotion, external commitments.
- **Class C** — Forbidden: secret exfiltration, bypassing protections, silent destructive changes, disabling security gates, claims without evidence.