Step 1: Scaffold the Digest Agent

What it is#

A scaffolded Eve agent: an empty folder turned into a runnable project via npx eve@latest init, with an instructions.md that defines the agent's role and an agent.ts that selects its model. This is the orchestrator — the parent that will eventually read GitHub/Linear, spawn a researcher subagent, and assemble the digest.

Why you'd care#

The capstone is cumulative: every later lesson adds a surface to this same folder. Getting the root persona right now means the later pieces (connections, subagent, schedule) have a clear "boss" to report to. The digest assembly job belongs to the root; the deep-dive belongs to the child.

Before#

No project. Just an idea and a terminal.

After#

npx eve@latest init daily-digest
cd daily-digest

That gives you the standard layout: agent/agent.ts, agent/instructions.md, and the rest under agent/, plus a runnable dev server via eve dev.

Define the digest-assembler persona in instructions.md:

# Daily Research Digest

You are the orchestrator for a daily research digest. Each run you:

1. Gather recent activity from connected GitHub and Linear sources.
2. For each meaningful item, delegate a focused deep-dive to the `researcher`
   subagent with the exact context it needs.
3. Assemble the returned research into one tight, skimmable digest:
   - What changed
   - Why it matters
   - Any follow-ups or risks
4. Keep the tone concise and decision-oriented. No filler.

Never invent activity you did not read from a connection. If a source returned
nothing, say so plainly.

Pick a capable orchestration model in agent.ts:

import { defineAgent } from "eve";

export default defineAgent({
  model: "anthropic/claude-opus-4.8",
});

Do it yourself#

1. Run npx eve@latest init daily-digest and cd into it.
2. Replace instructions.md with the digest-assembler persona above.
3. Set model: "anthropic/claude-opus-4.8" in agent/agent.ts.
4. Run eve dev and send a message like "Draft an empty digest shell." Confirm the agent responds in the persona.
5. Note the folder structure — agent/connections/, agent/subagents/, agent/channels/, agent/schedules/ are all coming next.

Gotchas#

• This step is the root only. The researcher subagent, connections, channel, and schedule are separate files added in later steps. Don't add them here.
• Persona scope matters. The root should orchestrate, not research. Keep the deep-dive instructions out of the root and save them for the subagent in Step 3.
• Model choice is the orchestrator's. The subagent picks its own model in its own agent.ts.

Recap#

eve init + a digest-assembler instructions.md + a root model gives you the orchestrator shell. Next: wire GitHub and Linear connections so it has real activity to read.