> ## Documentation Index
> Fetch the complete documentation index at: https://harisfazillah.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# 🤖 DSOM Multi-Agent Protocols (v6.1)

> OKF-compliant documentation for MULTI-AGENT-PROTOCOLS.md. Deep State of Mind framework. Deep State of Mind framework. Deep State of Mind framework.

> **Entry Point 6:** This document serves as the Subagent Swarm Entry Point. See [START-HERE.md](../../START-HERE) for the master onboarding roadmap.

> **"One Mind, Many Hands."** - The DSOM Philosophy for Multi-Agent Collaboration.

## 1. 🏛️ The Agent Taxonomy

To maintain **Sovereign Context**, we categorize external AI agents into two valid tiers. Any agent not fitting these tiers is considered "Unverified" and must not write to the repository without explicit Human Audit.

### Tier 1: The Co-Pilots (IDE Embedded)

These agents live inside the editor and have direct access to the active file buffer.

* **Examples:** GitHub Copilot, Google Antigravity, Cursor, Windsurf, Tabnine, Replit AI.
* **Access Level:** `Read/Write` (Buffer only).
* **Context Source:** `.cursorrules`, `.windsurfrules`, or System Prompt injection.

### Tier 2: The Autonomous Workers (Agentic Frameworks)

These agents operate independently, often tasked with entire feature implementations.

* **Examples:** Devin, AutoGen (Microsoft), CrewAI, Flatlogic.
* **Access Level:** `Read/Write` (FileSystem).
* **Context Source:** `sod_manifest_YYYY-MM-DD.txt` (Must be explicitly fed the Start-of-Day manifest, which includes **Section \[14] Palace Registry**).

***

## 2. 🔌 The Universal Injection Method (UIM)

Regardless of the agent, the **DSOM Laws** must be injected before the first prompt.

### For Tier 1 (IDEs)

Create/Update the specific rule file (e.g., `.cursorrules`) in the root directory with this pointer:

```text theme={null}

# .cursorrules / .windsurfrules

YOU ARE A DSOM-COMPLIANT AGENT.
BEFORE ANSWERING, READ: docs/AI-MASTER-PROTOCOL.md
FOLLOW: docs/PERSONALIZATION.md
CRITICAL: DO NOT CREATE GLOBAL VARIABLES. ADHERE TO CLEAN ARCHITECTURE.

```

### For Tier 2 (Autonomous)

You must perform the **Handshake Ritual** manually or via their API config:

1. **Ingest:** Upload `sod_manifest_YYYY-MM-DD.txt` (Generated by `reanimate.sh` or `bash tools/sod-palace.sh  # (Windows: .\tools\sod-palace.ps1)`).
2. **Palace Walk:** The agent must read `palace_registry.md` (Section \[14] of the manifest) and identify relevant Rooms before writing any code.
3. **Prompt:** *"I am initialising the DSOM Protocol v6.1 + Palace v1.0. Acknowledge the architecture in the manifest. Walk the Palace Registry and identify the relevant Rooms for today's task. State 'Sovereign State Synchronised' when ready."*

***

## 3. 🛡️ Agent-Specific Configuration Registry

### A. Cursor & Windsurf (The Forked VS Codes)

These are the most DSOM-native compatible due to their "Context Awareness" features.

* **Config File:** `.cursorrules` / `.windsurfrules`
* **Strategy:** Point them to `@docs/AI-MASTER-PROTOCOL.md` and `@docs/OPERATIONAL-GUIDE.md`.
* **Templates:**
  * [Cursor Template](../agent-configs/cursorrules_template)
  * [Windsurf Template](../agent-configs/windsurfrules_template.md)

### B. GitHub Copilot

* **Config File:** `.github/copilot-instructions.md` (If supported) or top-of-file comments.
* **Strategy:** Use the `// DSOM: [Instruction]` comment syntax.
* **Template:** [Copilot Instructions](../agent-configs/copilot_instructions_template)

### C. CrewAI & AutoGen

* **Config File:** `config/agents.yaml` or `scenarios.json`.
* **Strategy:** The "Role" definition must include: *"You are a DSOM-Compliant engineer. Your memory is persisted in .agents/brain/."*
* **Template:** [Autonomous Manifest](../agent-configs/autonomous_agent_manifest.md)

### D. Devin / Replit Agent

* **Config File:** Project Custom instructions / `.replit`.
* **Strategy:** Add the "Architectural Mantra" (from Personalization Block 6) to the environment's system prompt.
* **Template:** [Autonomous Manifest](../agent-configs/autonomous_agent_manifest)

***

## 4. The "Zero-Hallucination" Rule for Agents

If an agent cannot read `docs/AI-MASTER-PROTOCOL.md`, it is **forbidden** from refactoring Architectural Layers (`src/Domain`, `src/Application`). It may only work on `tools/` or `tests/`.

## 5. Palace Requirement for Agents (v1.0)

Any agent performing advisory, architectural, or code-generation tasks **MUST** also read `palace_registry.md` before starting. This ensures the agent knows:

* Which Wing and Rooms are relevant to the current task
* What prior decisions were made in those Rooms (via `closet.md`)
* Which discoveries are flagged as critical in `room_uncategorised/closet.md`

If the Palace Registry is missing, the agent must flag this and request that the human run:

```bash theme={null}
bash tools/palace-sync.sh --backfill

```

*Last Updated: 2026-04-08 | Standard: DSOM For My AI Protocol v6.1 + Palace v1.0*

***

*Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-07-04*
*Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0*


## Related topics

- [📜 DSOM Master Directive: AI Governance Protocol (v6.1 + Palace v1.0)](/governance/ai-master-protocol.md)
- [🧠 Deep State of Mind (DSOM) For My AI Protocol](/core-concepts.md)
- [🕯️ DSOM Ritual of Transition (v6.1 + Palace v1.0)](/rituals/transition-ritual.md)
- [[AGENT] DSOM Cognitive Digital Twin: Project Operational Protocol (v2.0)](/governance/ai-cognitive-twin-protocol.md)
- [HOWTO: audit-pre-flight - Intelligence Integrity Sync](/tools/howto-audit-pre-flight.md)