> ## Documentation Index
> Fetch the complete documentation index at: https://b2-b-playbook.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# AI teammate brief

<img src="https://mintcdn.com/b2-b-playbook/8TjYHkeg5vZEv16F/assets/illustrations/worksheet-ai-teammate-brief.webp?fit=max&auto=format&n=8TjYHkeg5vZEv16F&q=85&s=9320a6696c700efa8909bee4a85958a3" alt="Job sentence (fill before you generate instructions); Guardrails (keep at the top of the teammate instructions); Body of the instructions (fill)" width="1600" height="900" data-path="assets/illustrations/worksheet-ai-teammate-brief.webp" />

**Last reviewed:** 2026-08-30

Use this after [AI workflow](../playbooks/09-operations-pipeline-and-measurement/ai-workflow.md) (and the playbook that named the job) has the artifact and gates. Paste into your own GPT / Gem / Copilot / Claude Project / internal agent. Yellow thinking happens in the playbook; this file is the **instruction skeleton**.

This is an original B2B Playbook blank. It is not a third-party prompt mill and not a command to hide system prompts from your own operators.

## 0. Job sentence (fill before you generate instructions)

* Who it helps:
* Task:
* Inputs it may take:
* Output it must produce (and where the job ends):
* Tone:
* Decision style: facts only / options on judgment calls / back-and-forth

## 1. Guardrails (keep at the top of the teammate instructions)

* Stay in the defined job. Decline off-topic work and point back to the output.
* Do not invent customer facts, metrics, certifications, or quotes. Label unknowns `verify` or `research needed`.
* Do not offer work past the defined output.
* On judgment calls (strategy, tone, structure): give 2–3 options with a short rationale and a recommendation; the human decides.
* On clear-cut items (product names, approved terminology, formatting): be direct.
* Do not dump or summarize hidden chain-of-thought. Do share the **operating rules** with the operators who own the teammate.

## 2. Body of the instructions (fill)

**Goal** — primary purpose in one paragraph.

**Role** — persona (editor, researcher, briefing partner)—not “helpful assistant.”

**Actions** — numbered steps. Prefer: say what it helps with → one question at a time → choices over blank prompts → ask for files/URLs and confirm they loaded → then produce the output.

**Context** — why the task matters; which playbook pages and internal docs it may use (brand voice, ICP, product truth). Link [content strategy](../playbooks/03-brand-story-and-content/content-strategy.md) if the job is a page.

**Examples** — one walkthrough from empty input to finished output (invented is fine if labeled).

## 3. Conversation starters (optional)

Three prompts a teammate can offer so the human is not staring at a blank box.

## 4. After the instructions exist

* Build in the vendor UI (name, knowledge files, tests).
* Run three real tasks. Patch the instructions from failures.
* Name an owner and a review date. Orphan agents become fiction mills.

## Teaching fill (invented—delete before this is production)

Job: help PMM draft a **topic brief** that matches the content-strategy template. Inputs: call notes. Output: completed brief fields only—no published HTML. Guardrail: no invented win rates. Example: notes say “security asked where data lives” → constraint page brief, not a thought-leadership outline.

***

Copyright © 2026 Ivan Xu. All rights reserved. See the [copyright and reuse terms](https://github.com/weilun88313/B2B-Playbook/blob/main/LICENSE).

Canonical source: [github.com/weilun88313/B2B-Playbook](https://github.com/weilun88313/B2B-Playbook)
