AELLI

Central orchestration brain for the GFE/IntegraHub engineering ecosystem. A2A-compliant agent server registered in LiteLLM. Dev-advisor is the first skill.

Prerequisites

• Node.js 18+
• LiteLLM running at localhost:4000 with aelli-dev-advisor registered
• aelli-cc-plugin installed in Claude Code

Setup

npm install

Set in environment (or ~/.claude/settings.json env block):

Variable	Default	Purpose
AELLI_INBOUND_SECRET	—	Shared secret for x-aelli-secret header auth. If unset, auth disabled with a startup warning.
PORT	3456	HTTP server port
AELLI_LLM_ADVICE	false	Enable LLM-powered advice (v2)
LITELLM_BASE	http://localhost:4000	LiteLLM gateway URL for outbound LLM calls
AELLI_AUTH_TOKEN	—	Bearer token for outbound LiteLLM calls

Run

npm start

Expected output: [AELLI] Server running on port 3456

Test

npm test

Endpoints

Endpoint	Auth	Purpose
GET /a2a/dev-advisor/.well-known/agent-card.json	None	LiteLLM agent discovery
POST /a2a/dev-advisor	x-aelli-secret	Dev-advisor skill handler

Manual smoke test

1. Start AELLI: npm start
2. Start LiteLLM with the updated gfe-allspark/litellm-config.yaml
3. Open a Claude Code session in any git repo with aelli-cc-plugin enabled
4. Edit a file, then submit a prompt that does not mention the filename
5. Expected: [AELLI spec-deviation]: Modified files not mentioned in prompt: <filename>. appears in the Claude transcript context before the model responds
6. Kill AELLI mid-session, submit more prompts — Claude Code must continue normally with no errors

Adding a new skill

1. Create src/skills/<name>/ with index.js (exports handle(event)) and card.json
2. Mount in index.js — three lines
3. Register in gfe-allspark/litellm-config.yaml agent_list with its URL path