agent

A minimal eve agent (Vercel's durable agent framework), structured using the "harness engineering" ideas from OpenAI's Codex team post:

• AGENTS.md is a short map, not an encyclopedia — depth lives in docs/.
• docs/ is the system of record — anything the agent needs to reason about that isn't in code lives here as versioned markdown.
• Design docs, product specs, and exec plans are indexed and graded (docs/QUALITY_SCORE.md), so drift is visible instead of silent.
• Behavior is enforced through structure (typed tool boundaries, a fixed sandbox/tools/skills layering — see ARCHITECTURE.md) rather than through a long list of prose rules.

What's here

agent/
├── package.json
├── AGENTS.md                 # map / table of contents
├── ARCHITECTURE.md           # domain + layering rules
├── docs/
│   ├── design-docs/
│   │   ├── index.md
│   │   └── core-beliefs.md
│   ├── product-specs/
│   │   ├── index.md
│   │   └── trip-planning.md
│   ├── exec-plans/
│   │   ├── active/README.md
│   │   ├── completed/README.md
│   │   └── tech-debt-tracker.md
│   └── QUALITY_SCORE.md
└── agent/
    ├── agent.ts               # model choice
    ├── instructions.md        # identity + policy (no task procedure)
    ├── tools/
    │   ├── get_weather.ts     # mocked weather lookup
    │   └── read_notes.ts      # reads files from the sandbox workspace
    ├── skills/
    │   ├── plan_a_trip.md     # multi-step trip planning procedure
    │   └── summarize_notes.md # summarize saved notes
    └── sandbox/
        └── sandbox.ts         # bootstrap: seeds notes/trip-ideas.md

Running it

npm install
npm run dev

This starts the eve dev server / terminal UI. Try:

• "What's the weather in Lisbon?" → calls get_weather directly.
• "Help me plan a 3 day trip to Lisbon" → loads the plan_a_trip skill, which calls get_weather and read_notes in sequence.
• "What's in my trip notes?" → loads summarize_notes.

Growing this project

Follow the order in AGENTS.md: read ARCHITECTURE.md, check docs/exec-plans/active/ and docs/product-specs/, then add a tool/skill, then update the relevant doc in the same change. Nothing here is meant to stay minimal forever — it's a seed structured so growth doesn't turn into drift.