Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

51 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

AgentSeed β€” One Command to Give Your Agent a Brain

pip install https://github.com/weed33834/agentseed/releases/download/v2.4.0/agentseed-2.4.0-py3-none-any.whl && agentseed forge β†’ A blank agent instantly gains a complete persona, rule system, skills, and tool config.

🌐 Languages: English Β· δΈ­ζ–‡ Β· ζ—₯本θͺž

πŸ“¦ Platforms: GitHub β€” Primary (Releases, pip install, CI/CD) Β· Gitee β€” Backup Mirror Β· Gitcode β€” Backup Mirror

License Personas Platforms Tests Python

AgentSeed is a Persona-Governance Platform for AI Agents. It's what you inject into a blank AI coding assistant (Claude Code, Cursor, Copilot, Trae, Gemini, Windsurf, and more) to give it:

  • 🧬 A Permanent Brain (Governance Engine) β€” safety boundaries, decision formulas, self-evolution triggers
  • 🎭 A Swappable Personality (Persona Packs) β€” coding, writing, research, game design, and custom
  • πŸš€ Zero-Config Platform Sync β€” 13 platforms, one command

How It Works

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚                 AgentSeed                      β”‚
β”‚                                               β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚
β”‚  β”‚  ⚑ GOVERNANCE ENGINE (non-swappable)  β”‚ β”‚
β”‚  β”‚  P0 red lines Β· decision formulas Β·    β”‚ β”‚
β”‚  β”‚  self-evolution triggers               β”‚ β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚
β”‚                    ↓                          β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚
β”‚  β”‚  🎭 PERSONA PACKS (swappable)            β”‚ β”‚
β”‚  β”‚  coding Β· novel Β· paper Β· agent-builder   β”‚ β”‚
β”‚  β”‚  conversation Β· interactive-novel Β· customβ”‚ β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚
β”‚                    ↓                          β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚
β”‚  β”‚  πŸš€ 13 PLATFORMS SYNC                    β”‚ β”‚
β”‚  β”‚  Claude Code Β· Cursor Β· Copilot Β· Trae   β”‚ β”‚
β”‚  β”‚  Gemini Β· Windsurf Β· Cline Β· Continue    β”‚ β”‚
β”‚  β”‚  Amazon Q Β· Qodo Β· Lingma Β· Comate       β”‚ β”‚
β”‚  β”‚  AGENTS.md (universal)                   β”‚ β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Quick Start

# Install
pip install https://github.com/weed33834/agentseed/releases/download/v2.4.0/agentseed-2.4.0-py3-none-any.whl

# Auto-detect β†’ assemble β†’ generate
agentseed forge

# Interactive mode: choose your persona
agentseed forge --interactive

# Specify a persona
agentseed forge --profile coding

# Preview without writing
agentseed forge --dry-run

# Switch persona
agentseed switch --profile novel

# List all available personas
agentseed list

# Sync to a specific platform
agentseed sync --platform cursor

# Start MCP Server (stdio mode)
agentseed serve

# Start MCP Server (HTTP/SSE mode)
agentseed serve --port 8080

AgentSeed will detect your project type (pyproject.toml β†’ coding, chapters/ β†’ novel, etc.), select the best Persona Pack, and generate all necessary rule files.

MCP Server

AgentSeed exposes its governance engine and persona management as MCP tools, enabling any MCP-compatible client to query and enforce AI safety rules programmatically:

Tool Description
governance_check Check if a tool call violates P0 security red lines
persona_list List all available persona packs
persona_activate Switch to a specific persona
gap_detect Analyze context for capability gaps

Configure your MCP client with:

{
  "mcpServers": {
    "agentseed": {
      "command": "agentseed",
      "args": ["serve"]
    }
  }
}

Why AgentSeed?

Problem AgentSeed Solution
AI agents lack consistent behavior rules P0 Governance β€” same safety baseline everywhere
Different tasks need different personalities Persona Packs β€” swap identities without losing safety
Setting up rules for multiple tools is tedious 13-platform sync β€” one command, all covered
Agents get stuck when their preset tools fail Self-Evolution Engine β€” auto-detect gaps, search/fetch/install
Custom personas are hard to create and share Persona Market β€” create once, share with community

vs Competitors

Project What It Does AgentSeed Difference
agent-rules (steipete) Unified .mdc rules for Cursor/Claude Archived; coding-only
agent-rules-books Rules distilled from software books Coding-only; no personas
ACP Agent config + MCP management No governance; no self-evolution
agents.md AGENTS.md format standard Format only; no content
chatgpt_system_prompt System prompt collection Collection only; no toolchain
AgentSeed Full persona + governance + sync platform Complete agent brain

What's Inside

🧬 Governance Engine (Constitution Layer)

The permanent brain. Never overridden by any Persona Pack.

  • core/governance.md β€” P0 red lines (security, truth, boundaries)
  • core/constraints.yaml β€” Machine-enforceable hooks
  • core/agent-modes.md β€” Task / Project / Autonomous modes
  • core/self-evolution.md β€” β˜… Gap detection + self-healing
  • core/dar-spec.md β€” Domain authority scoring (search quality)
  • core/persona-router.md β€” Persona routing & selection

🎭 Persona Packs (Swappable)

Each Pack = SOUL + Rules + Skills + MCP + Prompts

Persona For Key Traits
coding Software Engineers Refactoring, testing, CI/CD
novel Novelists Chapters, characters, worldbuilding
paper Academics Literature review, LaTeX, submission
conversation General Assistants Q&A, research, analysis
interactive-novel Game Writers Branching narrative, state machine
agent-builder Agent Designers Build, evaluate, deploy agents

πŸš€ Platform Sync

Generates platform-native rule files for 13 platforms:

Claude Code, Cursor, Copilot, Trae, Gemini, Windsurf, Cline, Continue, Amazon Q, Qodo, Lingma, Comate, and AGENTS.md (universal).


Architecture

See docs/AGENTSEED_ARCHITECTURE.md for the full architecture design.

Key innovations:

  • Skeleton Mode: Core rules inline, skills on-demand β€” keeps prompts lean
  • Self-Evolution Engine: Agents detect their own capability gaps and self-heal
  • Quality Gates: All auto-fetched content passes safety β†’ quality β†’ compatibility checks

Install

# pip β€” install from GitHub Releases (primary)
pip install https://github.com/weed33834/agentseed/releases/download/v2.4.0/agentseed-2.4.0-py3-none-any.whl

# From source
git clone https://github.com/weed33834/agentseed.git     # GitHub (primary)
git clone https://gitee.com/badhope/agentseed.git        # Gitee (mirror)
git clone https://gitcode.com/badhope/agentseed.git      # Gitcode (mirror)
cd agentseed
pip install -e .

Develop

# Create a new persona
agentseed persona new my-role

# Validate rules
agentseed verify

# Run tests
python -m pytest tests/

# Check rule quality
python scripts/validate_rules.py

License

MIT


AgentSeed: The constitution teaches an agent when to find its own resources; identity decides what it excels at.

About

AgentSeed - One command to give your AI agent a brain. Persona-Governance Platform: governance engine + swappable persona packs + self-evolution + 13-platform sync.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages