Skip to content
RuneRune

Agent-driven workflow

Use case: let an AI agent run your tasks — and write a task whose body is a natural-language instruction — safely.

Demonstrates: agent executor, MCP exposure, destructive gating · Guide: AI agents (MCP)

Prerequisites: an agent CLI (e.g. claude, codex) for the summarize task

Run it

Section titled “Run it”
Terminal window
rune summarize

Expected output

Section titled “Expected output”

A short, agent-written summary of recent changes (the exact text depends on your agent and repository). The list-tasks task is a safe, read-only tool an agent may call.

How it works

Section titled “How it works”
  • summarize (agent) runs its body as an instruction via an installed agent CLI.
  • Exposed over MCP, tasks are read-only by default; reset-db is gated with [confirm], so an agent can’t run it unless the server is started with --allow-destructive reset-db.
  • Secrets come from the environment only — never written in the Runefile, and never shown in any task description exposed to an agent.

Tier-B verification skips this example when no agent CLI is available; the Runefile is still statically validated.