Overlook

A self-improving code scanner that surfaces novel, blog-worthy technical angles humans may have missed — not security lint, not style warnings, not things any static analyzer would catch. It scans AI-generated (or any) code repositories and looks for what's genuinely interesting: unusual architecture decisions, clever patterns, non-obvious tradeoffs, techniques worth learning from.

The system runs a negative-feedback loop: scan, measure, propose improvements, human review. Nothing changes automatically.

Quick start

Requires Node >= 20.

npm install
cp .env.example .env
# Add your ANTHROPIC_API_KEY to .env
npm run dev

Open http://localhost:3000 — the dashboard has four tabs:

• Scan — queue a git URL, watch the pipeline run, expand results inline
• Findings — browse/filter all findings across jobs by confidence and keyword
• Tickets — implementation tickets tracking work on the system itself
• System — health, LLM config, build identity, docs, build history

How it works

git URL → Ingest (clone) → Chunk (files → review units) → Analyze (LLM) → Sink (persist)

The host runtime orchestrates jobs, calls the LLM API, and enforces caps. The agent package (src/agent/) owns prompts, model preferences per operating mode, and chunking policy. Model selection flows through versioned config first, with env overrides as an operator knob.

Operating mode	Intent	Default model
Discovery	Breadth, recall, cheaper	claude-haiku-4-5-20251001
Triage	Quick filter/score	claude-haiku-4-5-20251001
Deep-dive	Precision, stronger model	claude-sonnet-4-20250514

Commands

Command	Purpose
npm run dev	Dev server with auto-restart
npm run build	TypeScript → dist/
npm run test	Unit/API tests (Vitest)
npm run check	Build + test
npm run cli -- scan <url> [ref]	Scan a repo via CLI
npm run cli -- doctor	Health check + next steps

Environment

Only ANTHROPIC_API_KEY is required. Everything else has sensible defaults.

Variable	Role
ANTHROPIC_API_KEY	Anthropic API secret (required for analysis)
ANTHROPIC_MODEL	Override model for all modes (optional — agent config is the default)
PORT	HTTP port (default 3000)
NODE_ENV	development or production

See .env.example for the full list.

Architecture docs

Document	What it covers
docs/systems.md	Goal priority, subsystems, control guardrails, update authority
docs/agentic-subsystem.md	Pipeline stages, host vs agent package, contracts, operating modes
docs/diagrams.md	ASCII architecture diagrams
docs/build-log.md	Build log policy (append-only accepted changes)
docs/dev.md	Local development, commands, dashboard

Project structure

src/
  agent/          Agent package config (model defaults, operating modes)
  jobs/           Pipeline: ingest, chunk, analyze, sink, job store
  llm/            Anthropic client and env/model resolution
  routes/         Express routes (git jobs, tickets, info, build history)
  cli/            CLI commands (scan, doctor, check)
public/           Dashboard (single HTML file, no build step)
docs/             Architecture specs
templates/        BUILD_LOG.md entry template
BUILD_LOG.md      Append-only record of accepted changes