Skip to content

πŸ“¦ Agent Plugins 1.0.0 Specification & DSOM Integration

Entry Point 20: This document defines the packaging, distribution, and runtime interoperability architecture for Agent Plugins 1.0.0 within the Deep State of Mind (DSOM) framework.


🎯 Executive Summary & Context

Agent Plugins 1.0.0 is an open, vendor-neutral directory specification governed by core maintainers from Google (DeepMind / Cloud), Amazon, Microsoft, OpenAI, Cursor, and Vercel. It eliminates packaging friction across disparate AI coding agents and IDEs by standardising how Agent Skills and Model Context Protocol (MCP) servers are bundled, discovered, and executed in a single portable directory.

Under the DSOM protocol, Agent Plugins 1.0.0 serves as the universal distribution and packaging standard, seamlessly bridging DSOM's Sovereign Brain (.agents/brain/), Skills repository (.agents/skills/), and FastMCP Server (tools/mcp/server.py) across Google Antigravity, Google Jules, Claude Code, Cursor, and enterprise agent environments.


πŸ›οΈ The 5-Layer Agent Ecosystem Architecture

flowchart TD
    Find["1. Find It: Agentic Resource Discovery (ARD)<br/><code>agenticresourcediscovery.org</code>"] --> Describe["2. Describe It: AI Catalog<br/><code>application/agent-plugins+json</code>"]
    Describe --> Package["3. Package It: Agent Plugins 1.0.0<br/><code>plugin.json</code> + <code>mcp.json</code> + <code>skills/</code>"]
    Package --> Run["4. Run It: FastMCP & Agent Skills<br/><code>tools/mcp/server.py</code> & <code>SKILL.md</code>"]
    Run --> Govern["5. Govern It: Deep State of Mind (DSOM)<br/>Spatial Memory + 10 Guardrails + OKF v0.2"]
  1. Find it β€” Agentic Resource Discovery (ARD): Discovers available plugins and capabilities prior to invocation.
  2. Describe it β€” AI Catalog: Indexes plugin descriptors and resource mappings.
  3. Package it β€” Agent Plugins 1.0.0: Standardised directory layout, manifest (plugin.json), and MCP descriptor (mcp.json).
  4. Run it β€” MCP & Agent Skills: Portable execution contracts across transports (stdio, streamable-http, sse).
  5. Govern it β€” DSOM Metacognition: Tri-Phasic Mind cognitive continuity, 10 Custom Guardrails, and GitOps audit trail.

πŸ“ Package Layout Standard (Rule 30)

Every DSOM-compliant Agent Plugin adheres to the fixed filesystem layout:

my-dsom-plugin/
β”œβ”€β”€ plugin.json              # Required: Closed manifest (schema 1.0.0)
β”œβ”€β”€ mcp.json                 # Optional: Explicit MCP server descriptors
β”œβ”€β”€ skills/                  # Optional: Fixed directory for Agent Skills
β”‚   └── summarize/
β”‚       β”œβ”€β”€ SKILL.md         # Open Knowledge Format (OKF v0.2)
β”‚       β”œβ”€β”€ scripts/         # Idempotent execution scripts
β”‚       └── references/      # Progressive disclosure documentation
β”œβ”€β”€ org.dsom.protocol/       # Optional: Client extension namespace
β”‚   β”œβ”€β”€ guardrails.json      # DSOM custom validators configuration
β”‚   └── hooks/               # Tri-Phasic interception hooks
β”œβ”€β”€ LICENSE                  # GNU GPL v3.0 / SPDX identifier
└── CHANGELOG.md             # Semantic version ledger

πŸ“œ Manifest Contracts

1. plugin.json (Root Manifest)

The root manifest declares metadata and targeted schema version:

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "deep-state-of-mind",
  "version": "10.4.0",
  "description": "Deep State of Mind (DSOM) Sovereign AI Agent Cognitive Architecture, Open Knowledge Format (OKF), and Model Context Protocol (MCP) toolset.",
  "author": {
    "name": "Harisfazillah Jamel (LinuxMalaysia)",
    "email": "linuxmalaysia@gmail.com",
    "url": "https://github.com/linuxmalaysia"
  },
  "homepage": "https://linuxmalaysia.github.io/deep-state-of-mind-for-my-ai/",
  "repository": "https://github.com/linuxmalaysia/deep-state-of-mind-for-my-ai",
  "license": "GPL-3.0-or-later",
  "keywords": [
    "dsom",
    "ai-agents",
    "mcp",
    "agent-skills",
    "agent-plugins"
  ],
  "extensions": {
    "org.dsom.protocol": {
      "tri_phasic_mind": true,
      "okf_version": "0.2"
    }
  }
}

2. mcp.json (MCP Configuration)

Maps Model Context Protocol servers explicitly without guesswork:

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/mcp.schema.json",
  "mcpServers": {
    "dsom-knowledge-server": {
      "type": "stdio",
      "command": "uv",
      "args": [
        "run",
        "--with",
        "fastmcp",
        "--with",
        "pyyaml",
        "python",
        "${PLUGIN_ROOT}/tools/mcp/server.py"
      ],
      "env": {
        "DSOM_PALACE_ROOT": "${PLUGIN_ROOT}",
        "DSOM_STORAGE_DIR": "${PLUGIN_DATA}/storage"
      },
      "cwd": "${PLUGIN_ROOT}"
    }
  }
}

πŸ›‘οΈ Variable Expansion & Path Containment

The specification defines two strict environment variables that conformant clients MUST provide to plugin subprocesses:

  1. ${PLUGIN_ROOT}: Absolute path to the resolved plugin root. Used for referencing bundled binaries, Python scripts (tools/mcp/server.py), and documentation palaces.
  2. ${PLUGIN_DATA}: Client-managed persistent directory dedicated to that installed plugin instance. Used for virtual environments, local SQLite caches, and generated artifacts.

[!CAUTION] Path Containment Invariant (Β§4.1): All relative paths defined in mcp.json or plugin configurations MUST resolve strictly within ${PLUGIN_ROOT} or ${PLUGIN_DATA}. Any attempt to traverse upward (../) outside the root boundary will cause compliant clients to reject the component.


πŸ› οΈ Step-by-Step: Packaging a DSOM Skill as an Agent Plugin

  1. Create the Plugin Directory:
    mkdir -p plugins/my-ops-plugin/skills/sys-audit
    
  2. Write plugin.json:
    {
      "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
      "name": "my-ops-plugin"
    }
    
  3. Add OKF v0.2 SKILL.md: Place your standard DSOM skill inside skills/sys-audit/SKILL.md.
  4. Attach MCP Server (Optional): Add mcp.json declaring any auxiliary stdio or streamable HTTP tools.
  5. Verify with DSOM Guardrails:
    uv run python tools/install_git_guardrails.py
    uv run pytest tests/test_agent_plugins_spec.py
    


Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-08-22
Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0