Agency Open-Loop System: meeting notes to tasks, client follow-ups and a weekly open-loop report

Read this transcript [paste] from [meeting name, date]. List decisions, new commitments, updates to open loops, questions nobody answered, risks and flagged items.
Quote the source line for every commitment. Assign owners only from TEAM.md. Do not write to LOOPS.md yet.

No quote, no row. That one rule keeps invented tasks out of the ledger.

Draft the follow-up for [client] after [meeting] using CLIENTS.md for tone and channel: what we agreed, what we owe with dates, what we need from them.
Output an approval card. Leave a marked gap if scope or price came up. Do not send.

The agent drafts. A person sends.

# Agency Open-Loop System: meeting notes to tasks, client follow-ups and a weekly open-loop report

A PaperChaseWebb pack, version 1.0. Written for Claude, ChatGPT, Claude Code and Grok, and plain Markdown, so the ledger, the rules block, the skills and the prompts work with any assistant that can read your meeting notes and write files or drafts.

An owner-led agency loses money in the gaps between meetings. Someone says "we will send the revised brief Thursday" on a call, nobody writes it down with a name next to it, and the client remembers it two weeks later. This pack is a small operating system for those gaps. Every meeting becomes a list of commitments with an owner and a date. Every commitment lands where your team already works. Every client gets a follow-up draft you approve before it goes anywhere. Every Friday you get one report of what is still open, what is waiting on the client and what has gone quiet.

It is built from public threads where operators described how they give agents team context, extract action items and run recurring reports, plus our own agency workflows. The sources are credited by link in the Sources section. The agent drafts and tracks. A person approves anything a client will see.

## What is inside

| Section | What it is for |
|---|---|
| 1. The loop and the one rule | Four stages and the rule that keeps them safe |
| 2. The three files | LOOPS.md ledger, CLIENTS.md and TEAM.md, with templates |
| 3. Stage one: notes to commitments | Pull only real commitments, quote the line, flag what is unclear |
| 4. Stage two: commitments to accountable tasks | Name an owner, a date and a done test, then post it where the team works |
| 5. Stage three: approval-ready client follow-ups | Recap drafts in the client's tone, held for your yes |
| 6. Stage four: the weekly open-loop report | Opened, closed, overdue, waiting on client and gone quiet, per client |
| 7. Daily sweeps | Small checks that stop loops from aging |
| 8. Capture options | Notetakers, transcripts and what to do when bots are not allowed in |
| 9. Rules block | A paragraph for project instructions, CLAUDE.md or AGENTS.md |
| 10. Three skills | Loop extractor, follow-up drafter and open-loop reporter |
| 11. Prompts | 12 prompts for each stage |
| 12. Two-week rollout | Shadow mode on one client first, and what to count |
| 13. Client data and permissions | Consent, least access and retention |
| 14. Sources | What we used and how we used it |