Agent Factory

Agent Factory is a reference architecture for a quality validation loop:

onboard -> baseline -> compare -> report

It takes a validation request (usually from a pull request), runs reproducible quality checks, compares results to baseline, and emits artifacts for operator review.

Start Here

• I want to run Agent Factory: see docs/users.md
• I want to build/change Agent Factory: see docs/developers.md
• I need the full doc index: see docs/README.md

Audience Split

• Users (operators/integrators): deployment, run submission, metrics, troubleshooting, and day-2 operations
• Developers (contributors): architecture, contracts, roadmap, release flow, and implementation history

Quick Run (Local Golden Path)

npm install
npm run loop-demo

This runs a visible baseline -> regression -> recovery sequence. Artifacts land in artifacts/<run-name>/, including terminal summary result.json and quality report outputs.

Run a single local bot-orchestrated run (no GitHub/webhook required):

npm run bot

Optional: simulate replay failure quickly:

npm run bot -- --proxymock-mode fail

Check gate verdict and fail CI on blocking result:

npm run gate:check -- --run <run-name>

Quick Run (Always-On Compose)

cp .env.server.example .env.server
docker compose --env-file .env.server -f docker-compose.server.yml up -d
docker compose --env-file .env.server -f docker-compose.server.yml ps

Quick Run (Kubernetes)

kubectl apply -k examples/deploy/kubernetes/base
kubectl apply -k examples/deploy/kubernetes/overlays/redis

Submit quality runs through intake API POST /qa/runs using payloads in examples/runs/ (comparison and baseline examples are included).

Core Artifacts Per Run

• run.json
• baseline.json
• build.log
• validation.log
• quality-report.json
• quality-report.md
• gate.json
• evidence.json
• result.json