Tree-SOP Agent is a Python framework for multi-agent collaboration in a chat-like flow, with 11 preset roles, a four-layer memory system, and a hard-constrained harness. It emphasizes automated SOP pipelines and structured auditing, currently in alpha.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (2 so far).
What it is
Tree-SOP Agent is a group-chat style multi-Agent software development framework. It provides 11 pre-configured roles, a full SOP pipeline, a Harness with hard constraints, and a four-layer memory architecture to automate development workflows from requirements analysis to deployment.
How it works
The project defines an orchestration of agents (e.g., Dispatcher, PM Agent, Coding Agent, TDD Agent, Acceptance + Security, DevOps Agent, Secretary Agent) that collaborate through a structured pipeline. It includes a Control Plane (Harness) with memory routing and a 5-level concept of SOP-inspired workflows, plus an Agent Plane and an Orchestration Plane. The four memory layers are described as:
- Layer 1: CacheEngine with SHA-256 snapshots
- Layer 2: ContextPartitioner with immutable / append-only / volatile partitions
- Layer 3: EmbeddingIndex for semantic retrieval and ConversationCompressor for compression
- Layer 4: CheckpointManager for JSON checkpoints and write_log logging
Getting started
Environment requirements:
- Python 3.10+
- DeepSeek API Key (environment variable DEEPSEEK_API_KEY)
Install:
git clone https://github.com/luyi14-bits/tree-sop-agent.git
cd tree-sop-agent
pip install pydantic pydantic-settings pyyaml
Run:
# CLI mode — load all skills and self-test
python auto_test.py
# Chat group mode — Dispatcher entry
python run.py
# Mount a custom Skill
python run.py --attach my-custom-skill
# Inspect Agent prompt assembly
python -m src.tree_sop_agent.cli.main --skill-dir skills --inspect pm-mentor