Agent Task Mode

kind=agent executes task body through the agent and converges results via multi-round evaluator flow.

Use cases

• tasks requiring reasoning and iteration
• research / analysis / summarization
• human-readable deliverables

Execution characteristics

• multi-round execution (default max: 3)
• live progress events visible in Console UI
• channel-delivered output written to output.md, with summary in result.md

Best practices

• write explicit goals, boundaries, and acceptance criteria
• require auditable outputs when possible
• do not put raw shell scripts in agent tasks

Example

---
title: daily-market-research
description: daily market research summary
contextId: research
when: 0 8 * * *
status: enabled
kind: agent
---

Generate today's market research summary including:
1. major events
2. key risks
3. next-step recommendations