AIGitHub Commercial Radar

A strict agent-loop system that continuously scans GitHub and turns open-source activity into validated commercial opportunity signals.

OpenSource Opportunity Agent: a continuous agent loop that scans GitHub, analyzes open-source projects, validates commercial opportunity, stores qualified opportunities, and creates next actions.

This system is not a user-search tool. The operator runs the harness; the harness orchestrates role-specific LLM agents and evidence scripts.

Why Star This Repo

• Find open-source projects that can become real commercial products.
• Study a strict LLM-agent harness where agents must have role specs, execution specs, output specs, and skills.
• Reuse the validation loop for founder discovery, repo analysis, pain mining, business design, and go-to-market next actions.
• Track how evidence scripts and LLM agents can cooperate without turning the agent into a hardcoded keyword bot.

What It Does

GitHub signals -> LLM scout -> evidence collection -> repo analysis
  -> pain discovery -> business design -> founder-style validation
  -> approval/rejection -> next action plan

The loop is designed for founders, indie hackers, product strategists, and AI-agent builders who want a repeatable way to discover software business opportunities from open-source momentum.

Principles

• Core opportunity judgment is LLM/agent driven.
• Scripts collect facts, enforce contracts, persist state, and execute workflow transitions.
• No hardcoded keyword routing, fake success state, or silent fallback.
• An agent is only valid when it has a role spec, execution spec, output spec, and skill.
• If specs are not compiled, or the LLM provider is missing, the loop fails closed.

Quick Start

Clone and configure:

git clone https://github.com/EvanAI0331/aigit.git
cd aigit
cp .env.example .env

Fill .env with your own provider keys. Never commit .env.

Run one scan:

python3 -m aigithub_radar.cli init-db
python3 -m aigithub_radar.cli compile-specs
python3 -m aigithub_radar.cli run-once --theme-limit 3 --repos-per-theme 20 --deep-limit 2 --validate-limit 1
python3 -m aigithub_radar.cli report-today

Run continuously every 12 hours:

python3 -m aigithub_radar.cli ops-loop --interval-hours 12 --theme-limit 3 --repos-per-theme 20 --deep-limit 2 --validate-limit 1

Run the local frontend/backend:

python3 -m aigithub_radar.server

Open http://127.0.0.1:8028.

Required for real agent execution:

export AIGITHUB_ORCHESTRATOR_API_KEY=...
export AIGITHUB_ORCHESTRATOR_BASE_URL=...
export AIGITHUB_ORCHESTRATOR_MODEL=...
export AIGITHUB_ORCHESTRATOR_DISABLE_THINKING=true

export AIGITHUB_WORKER_API_KEY=...
export AIGITHUB_WORKER_BASE_URL=...
export AIGITHUB_WORKER_MODEL=...

The orchestrator/scout agents use the orchestrator LLM endpoint. The remaining agents use the worker LLM endpoint. Both endpoints must be OpenAI-compatible chat completion APIs.

Optional for GitHub rate limits:

export GITHUB_TOKEN=...

Agent Roles

Agent	Responsibility
orchestrator_agent	Plans the run and enforces the ops sequence.
scout_agent	Uses LLM judgment to generate GitHub search themes and queries.
repo_analyst	Converts repository evidence into adoption, activity, license, and packaging analysis.
pain_finder	Extracts likely user pain and buyer urgency.
business_designer	Designs commercial packaging, pricing, channels, and MVP tests.
validator_agent	Runs the founder-style validation gate and rejects weak opportunities.
strategist_agent	Creates next actions only after approval.

Workflow

theme pool
  -> scout agent
  -> github evidence scripts
  -> repo analyst agent
  -> pain finder agent
  -> business designer agent
  -> validator agent
  -> scoring harness
  -> database
  -> strategist agent

Statuses:

DISCOVERED -> SCREENED -> ANALYZED -> VALIDATED -> APPROVED -> STORED -> NEXT_ACTION_CREATED

Rejected statuses:

REJECTED_LICENSE_RISK
REJECTED_LOW_DEMAND
REJECTED_NO_BUSINESS_MODEL
REJECTED_TOO_HARD_TO_PACKAGE
REJECTED_TOO_GENERIC
REJECTED_ALREADY_SATURATED

Spec Contract

Specs live in specs/agents. Skills live in skills/agents.

python3 -m aigithub_radar.cli compile-specs first validates, verifies, and compiles specx/contracts/opportunity_agent_loop.json through the SpecX plugin CLI, then validates every agent has:

• role spec
• execution spec
• output spec
• skill file

The compiler writes build/specx/opportunity_agent_loop.plan.json and build/compiled_specs/*.compiled.json. The loop refuses to run without these compiled artifacts.

The current SpecX plugin release provides skills and CLI validation/compilation, not live MCP execution tools. The project uses that CLI directly through aigithub_radar/harness/specx_adapter.py.

Distribution

• Star the repo if you care about strict LLM-agent harnesses for business discovery.
• Share the launch copy in docs/PROMOTION.md.
• Use docs/DISTRIBUTION.md to publish on GitHub, Hacker News, X, LinkedIn, Reddit, Product Hunt, and AI-builder communities.
• Open issues for new collectors, validation gates, agent roles, and frontend dashboards.