Mahabharat Agents Package

One person. One AI company. Infinite leverage.

The Mahabharat Agents Package turns a solo developer into a full software company by deploying specialized AI agents — each named after a Mahabharata character, each embodying FAANG-level role expertise, each following strict cognitive bias prevention protocols.

The Idea

You are Krishna — the CEO, the God, the orchestrator. You don't fight every battle. You direct the war. Your agents are your Pandavas, each a master of their domain:

Architecture

Dharma-Karma-Katha (DKK) Agent Definitions
    + FAANG-Level Role Definitions
    + Composable Skills (Claude Code SKILL.md format)
    + Multi-Topology Orchestration (Vyuha, Sabha, Rajya, Chakravyuha, Akshauhini)
    + Artifact-Based Pub-Sub Communication
    + Tiered Context (Shruti/Smriti/Itihasa)
    + Master/Sepoy Hierarchy
    + Kurukshetra War Room Dashboard

Orchestration Topologies

Core Principles (ALL agents, non-negotiable)

1. Zero Cognitive Bias Protocol — No anchoring, confirmation, availability, survivorship, authority, or sunk cost bias. Question assumptions. Seek disconfirming evidence.

2. 5-Question Research Framework — For every non-trivial decision: Domain Understanding, Business Rules, Industry Standards, Data Requirements, Edge Cases.

Installation

Option 1: pip install (recommended)

pip install mahabharat-agents
mahabharat-install /path/to/your/project

Option 2: Shell script

git clone https://github.com/saikrishnamadavarapu/mahabharat-agents.git
cd mahabharat-agents
bash install.sh /path/to/your/project

Option 3: Manual

Copy agents/, skills/, orchestration/, and context/ directories into your project's .claude/ folder.

Quick Start

1. Install into your project
2. Agents appear in .claude/agents/
3. Skills appear in .claude/skills/
4. Invoke any agent by name in Claude Code: "Ask Arjuna to review this architecture"
5. Run workflows: "Run the feature-development workflow for the new auth system"

Continuous Improvement — Feedback Loop

Every agent gets better over time through structured peer reviews:

EXECUTE → OBSERVE → REVIEW → IMPROVE → repeat

• Self-assessment: Each agent evaluates their own performance
• Peer feedback: Agents review collaborators with specific, evidence-based feedback
• 360 synthesis: Krishna (CEO) consolidates feedback into improvement plans
• Agent updates: Agent definitions updated based on review (new guardrails, skill changes, authority adjustments)
• Promotion/demotion: High-performing sepoys can be promoted; underperforming agents get coaching

Run performance-review workflow weekly or after milestones:

"Run the performance-review workflow for this sprint"

Creating Your Own Company

# Add a new role
cp roles/_template.md roles/my-role.md

# Add a new agent
cp agents/masters/_template.md agents/masters/my-agent.md

# Add sepoys (supporting agents)
cp agents/sepoys/_template.md agents/sepoys/my-sepoy.md

# Add a workflow
cp orchestration/workflows/_template.yaml orchestration/workflows/my-workflow.yaml

License

MIT License. See LICENSE.

Credits

This project is built on ideas from MetaGPT, CrewAI, ChatDev, Swarms, MiroFish, OpenViking, AutoGen, CAMEL, and OpenHands. Full attribution in CREDITS.md.

"Yada yada hi dharmasya glanir bhavati bharata..." Whenever dharma declines, Krishna appears to restore order.

Built by Sai Krishna Madavarapu with Claude Code.