The Faux Foundation

A fake foundation staffed by artificial agents doing real work.

Named after MacGyver's Phoenix Foundation — except faux, because the agents aren't real people, but the knowledge they produce is.

What It Does

Star a repo on GitHub. A few minutes later, MacGyver — the first agent — notices, clones it, reverse-engineers the codebase, and commits a structured product-level spec. Stars become knowledge. The mind grows by paying attention.

Architecture

Dapr-powered monorepo. Each agent gets a mind (genesis-style) and handlers. dapr.yaml defines all apps — one file, three execution tiers.

agents/macgyver/mind/    → Genesis mind (SOUL.md, expertise/, skills)
agents/macgyver/src/     → Job handler, clone+invoke, auto-commit
apps/chat/               → React + Vite chat UI (nginx in production)
apps/tools/              → Tool service (web_fetch, search)
platform/components/     → Shared Dapr components (state, LLM conversation)
scripts/                 → dev.ps1 (launcher), kind-load.ps1 (image builder)

Stack

• Runtime: Node.js + Copilot SDK + bundled CLI (adapted from skeleton)
• Orchestration: Dapr Multi-App Run — sidecars, service invocation, state, pub/sub
• Scheduling: Dapr Jobs API — @every 30s poll, configurable via POLL_INTERVAL
• Trigger: Poll GET /user/starred?sort=created (GitHub App as graduation path)
• Spec format: Product-level reverse specs via reverse skill

Apps

App	Path	Port	Description
MacGyver	agents/macgyver/src/	3000	Star poller + reverse-spec agent
Chat	apps/chat/	8080	React + Vite chat UI (nginx in container)
Tools	apps/tools/	3100	Tool service — web_fetch, Brave search
LLM Proxy	../copilot-llm-svc	5100	Copilot SDK proxy (.NET, external repo)

External Dependencies

Dependency	Local Path	Repo
copilot-llm-svc	../copilot-llm-svc	copilot-llm-svc

Referenced by dapr.yaml (relative path) and azure.yaml (azd service).

Local Development

Three-tier dev loop — one dapr.yaml, three execution modes:

Tier	Command	What It Does	When to Use
Dev	.\scripts\dev.ps1	Processes + auto-sidecars	Daily coding, fast iteration (~3s startup)
Validate	.\scripts\dev.ps1 -Kind	Kind cluster, real containers + K8s	Pre-deploy smoke test, Dockerfile validation
Deploy	azd up	Azure Container Apps	Production

Prerequisites

• Dapr CLI — winget install Dapr.CLI then dapr init
• Kind — winget install Kubernetes.kind then kind create cluster --name faux-foundation and dapr init -k
• Node.js and .NET SDK on host (for process mode)
• Docker Desktop running (for Kind mode)
• azd environment configured with GITHUB_TOKEN, COPILOT_TOKEN, BRAVE_API_KEY

Quick Start

# Process mode — runs apps as native processes with Dapr sidecars
.\scripts\dev.ps1

# Kind mode — runs apps as containers in a local K8s cluster
.\scripts\kind-load.ps1          # build images + load into Kind (first time / after changes)
.\scripts\dev.ps1 -Kind          # deploy to Kind with port-forward to localhost:8080

dev.ps1 automatically loads secrets from azd env get-values and injects them into dapr.yaml at runtime.

Deployment

Azure Container Apps via azd. Three services: macgyver, llm-proxy, chat.

azd env list                      # show environments (default: dev)
azd deploy                        # deploy all services
azd deploy llm-proxy chat         # deploy specific services

Infrastructure is defined in infra/ (Bicep). Use azd up only when infra changes are needed.

Agents

Agent	Persona	Job
MacGyver	Angus MacGyver (1985) — practical, warm, takes things apart	Reverse-specs starred GitHub repos

License

MIT