diff --git a/.gitignore b/.gitignore
index 70cf721..ff83fbb 100644
--- a/.gitignore
+++ b/.gitignore
@@ -227,6 +227,7 @@ Release_Guide.md
 
 # Task tracking (local-only)
 /05_TASKS.md
+DECISIONS.md
 
 # Local documentation artifacts
 openai-docs.md
@@ -234,3 +235,8 @@ tiktoken-docs.md
 
 # Local data directory
 /data/
+
+# Local config files (contains local settings, use env vars for secrets)
+config.yaml
+config.groq.yaml
+config.*.yaml
diff --git a/QUICKSTART.md b/QUICKSTART.md
new file mode 100644
index 0000000..c15c151
--- /dev/null
+++ b/QUICKSTART.md
@@ -0,0 +1,112 @@
+# QUICKSTART.md — Coding Agent Quick Reference
+
+> Get productive in <60 seconds.
+
+---
+
+## Identity
+
+| Field | Value |
+|-------|-------|
+| Repo | `openagent-eval` |
+| Package | `openagent_eval` |
+| CLI | `oaeval` |
+
+---
+
+## First Commands
+
+```bash
+uv sync --group dev          # Install deps
+pytest                        # Run tests
+ruff check .                  # Lint
+```
+
+---
+
+## How to Run
+
+```bash
+oaeval run config.yaml        # Run evaluation
+oaeval audit --corpus ./docs  # Audit corpus
+oaeval diagnose --report r.json  # Blame attribution
+oaeval synth --corpus ./docs  # Generate test data
+oaeval test config.yaml -t faithfulness:gte:0.8  # CI/CD
+```
+
+---
+
+## Where to Put Code
+
+| What | Where |
+|------|-------|
+| New metric | `metrics/{category}/new_metric.py` |
+| New LLM | `providers/llm/new_provider.py` |
+| New retriever | `providers/retrievers/new_retriever.py` |
+| New command | `cli/commands/new_command.py` |
+
+---
+
+## Interfaces
+
+```python
+# Metric
+from openagent_eval.metrics.base import BaseMetric, MetricResult
+class MyMetric(BaseMetric):
+    name = "my_metric"
+    def evaluate(self, **kwargs) -> MetricResult:
+        return MetricResult(score=0.95, reason="Because...", metadata={})
+
+# LLM Provider
+from openagent_eval.providers.base.llm import LLMProvider
+class MyLLM(LLMProvider):
+    async def generate(self, prompt: str, **kwargs) -> str:
+        return "response"
+    async def get_token_count(self, text: str) -> int:
+        return len(text.split())
+
+# Retriever
+from openagent_eval.providers.base.retriever import Retriever
+from openagent_eval.providers.models import Document
+class MyRetriever(Retriever):
+    async def retrieve(self, query: str, k: int = 5) -> list[Document]:
+        return [Document(content="...", metadata={}, score=0.9)]
+```
+
+---
+
+## Testing & Git
+
+```bash
+pytest                              # All tests
+pytest tests/unit/test_metrics/     # Module
+pytest -k "test_faithfulness"       # One test
+```
+
+- **Never** develop on `main`
+- Branch: `feature/{desc}`, `fix/{desc}`, `docs/{desc}`
+- Push → PR → user reviews
+
+---
+
+## Context Files
+
+| File | Purpose |
+|------|---------|
+| `PROJECT.md` | What we're building |
+| `ARCHITECTURE.md` | How it's built |
+| `AGENT.md` | Coding rules |
+| `DECISIONS.md` | Why we chose this |
+| `CONTEXT.md` | Current state |
+| `INSTRUCTIONS.md` | Writing rules |
+
+---
+
+## Rules
+
+1. Never develop on `main`
+2. Never raise generic `Exception`
+3. Never put business logic in CLI
+4. Always use typed exceptions
+5. Always implement ABCs
+6. Always create PR for review
diff --git a/config.groq.yaml b/config.groq.yaml
deleted file mode 100644
index 626a96e..0000000
--- a/config.groq.yaml
+++ /dev/null
@@ -1,35 +0,0 @@
-# OpenAgent Eval Configuration - Groq + Mock Retriever
-# Uses Groq's Llama 3.3 70B for generation, mock retriever for offline testing
-
-dataset:
-  path: data/sample_questions.json
-
-llm:
-  provider: groq
-  model: llama-3.3-70b-versatile
-  temperature: 0.0
-
-retriever:
-  provider: mock
-
-metrics:
-  retrieval:
-    - context_precision
-    - context_recall
-    - mrr
-  generation:
-    - faithfulness
-    - answer_relevancy
-    - similarity
-  performance:
-    - latency
-  cost:
-    - token_count
-
-report:
-  output: terminal
-  output_dir: ./reports
-
-verbose: true
-parallel: false
-max_workers: 1
diff --git a/config.yaml b/config.yaml
deleted file mode 100644
index bd9fcd5..0000000
--- a/config.yaml
+++ /dev/null
@@ -1,27 +0,0 @@
-# OpenAgent Eval Configuration
-# See documentation for options: https://github.com/OpenAgentHQ/openagent-eval
-
-dataset:
-  path: data/sample_questions.json
-  format: json
-  # limit: 100
-
-llm:
-  provider: openai
-  model: gpt-4o-mini
-  temperature: 0.0
-
-retrieval:
-  # provider: chromadb
-  # collection_name: my_collection
-
-evaluation:
-  metrics:
-    - answer_relevancy
-    - faithfulness
-    - context_precision
-    - context_recall
-
-report:
-  output: terminal
-  output_dir: ./reports