Agent Manager: a weekly agent that audits all your other agents, with sample prompt
nlarusstone · x · 2026-11-18
Day 30 of nlarusstone's 'an agent a day' series tackles agent management: once you run many agents, you need a way to know whether they're actually working and producing useful output.
The fix: a 'chief of staff' agent that runs weekly and reviews all your other agents' recent runs, checking for recurring errors, missed triggers, unnecessary work, excessive costs, duplicated effort, and whether each agent actually achieves its intended goal rather than just completing runs.
The post includes a full sample prompt requesting a concise report with concrete examples and recommended changes, prioritized by expected impact, with approval required before modifying existing agents. A follow-up on grouping agents into 'Science Foundries' is teased.
More from coding & agent
- Solo game studio runs ~20 AI teammates, goes from 50 to 1,900 PRs/week and 6x revenue — gdechichi · 2026-11-18
- Blueframe: A Browser Canvas for Wireframing UI Structure Before AI Builds It — nima_owji · 2026-11-18
- Open-source Machine Internet turns any site's hidden JSON API into an MCP server — ChessTwo · 2026-11-18
- Neo4j Open-Sources Relate: An Embedded Rust Graph Database with Built-in Agent Memory — JeremyCMorgan · 2026-11-18
- Open-source mail MCP server adds OpenPGP end-to-end encryption, seeks security review — Soft-Lie-434 · 2026-11-18
- Beyond UX: AX and UAX are the new product layers in the agent era — yenkel · 2026-11-18