Explore AgentPlaybooks01 / A practical field guide

Stop starting over.Start building on it.

Give your AI a project it can pick up again. Keep instructions, reusable skills and decisions in one playbook, then bring the right context into the tools you already use.

Your knowledge stays portable. Your agent does the work.

Browse public playbooks
An illustrative projectPROJECT / NORTHSTAR
ChatGPTCodexClaude
One project. One playbook.Northstar
Private
InstructionsCheck before changing. Explain the tradeoffs.
01
SkillsReview → test → hand off
02
MemoryEmail-link login chosen. Production check next.
03
Load what this task needs
The next conversation

“Continue Northstar. What should we check next?”

Bring context into your existing AI workflow
ChatGPTCodexClaudeCursorHermesand more

01 / From question to useful work

Same AI. A better starting point.

Choose a scenario. See what you ask, which context matters, and what a useful answer could look like.

Choose a workflow
01 / Worked example

A new chat shouldn’t mean a blank slate.

Return to a project after a week, or switch to another assistant, without reconstructing every decision.

01You ask

“Pick up the Northstar portal. What should we do next?”

You have opened a new session. The previous conversation isn’t in its context.

02The playbook provides

01
Project direction

Ship accessible onboarding for the customer portal.

02
Reusable method

Separate confirmed work, open questions and next actions.

03
Saved context

Email-link login was chosen. Staging works; production is still unchecked.

Only the relevant material, not your entire knowledge base.

03Your agent can produce

A handoff you can act on
  1. Current state: staging login is recorded as working. Production remains unverified.
  2. Next checks: email delivery, expired links and keyboard navigation.
  3. Ask for the production environment and permission before making changes.

Illustrative output, not a live execution. Results depend on your agent, sources and available tools.

02 / Worked example

Keep the reasoning, not just the answer.

Use a repeatable research method and saved constraints to build a brief someone can challenge and update.

01You ask

“Compare these two issue trackers for our team. What’s missing before we choose?”

You provide two source documents. The assistant should not fill their gaps with invented facts.

02The playbook provides

01
Project direction

Find a tracker for a small team that needs data export.

02
Reusable method

Compare evidence, cite supplied sources and separate unknowns.

03
Saved context

Export is required. Budget approval and SSO requirements are still open.

Only the relevant material, not your entire knowledge base.

03Your agent can produce

An evidence-led decision brief
  1. A comparison of claims in the supplied documents, with source references.
  2. A gap list: verify export coverage, final price and the required SSO plan.
  3. A conditional recommendation, with the assumptions that could change it.

Illustrative output, not a live execution. Results depend on your agent, sources and available tools.

03 / Worked example

Make your standards part of the review.

Bring project rules and past decisions into a code review instead of repeating the same checklist each time.

01You ask

“Review this API caching change. What must we verify before release?”

Your coding agent can inspect the diff and run checks when its tools and permissions allow it.

02The playbook provides

01
Project direction

Reduce response time without exposing private data.

02
Reusable method

Trace access boundaries, cache keys, invalidation and failure paths.

03
Saved context

Public and private responses must stay isolated. Measure before claiming a speedup.

Only the relevant material, not your entire knowledge base.

03Your agent can produce

A focused release review
  1. Inspect whether cache keys distinguish every relevant access scope.
  2. Check permission changes, stale entries and anonymous versus signed-in requests.
  3. Report verified results, remaining risks and the measurements still needed.

Illustrative output, not a live execution. Results depend on your agent, sources and available tools.

These public, first-party starters are in English and contain generic instructions. Make a private copy before adding your project details. The scenarios above illustrate how to customize them.

02 / Find your starting point

Explore public playbooks.

Search shared playbooks, inspect their skills and tool connections, and make a private copy to adapt to your work.

Loading public playbooks…

03 / Inside the playbook

A working brief. With a memory.

A playbook brings together the pieces an assistant needs to work consistently. Start with instructions; add the rest when your workflow needs them.

Relevant context beats more context.Ask for the skill, reference or memory needed for this task. Keep unrelated material out of the conversation.
01
Persona

Define the role and the perspective you want your assistant to take.

“Act as a pragmatic project reviewer.”
02
Instructions

Keep goals, boundaries and working conventions in one place.

“Verify first. Flag assumptions.”
03
Skills

Package a repeatable method with the reference files it needs.

Release checklist + review guide
04
Memory

Save decisions and useful facts explicitly for later sessions.

Decision: use email-link login
05
Tool connections

Reference the tools your client needs to carry out the work.

Repository, database, issue tracker
06
Scoped credentials

Manage credentials separately from ordinary project instructions, with controlled access.

A key scoped to the required service

04 / Your first useful session

Start small. Build on real work.

You don’t need a giant prompt or a library of tools. One project, a clear goal and one useful method are enough to begin.

  1. 01

    Make it your project

    Create a playbook or copy a starter. Add your objective, boundaries and one repeatable task.

    A focused working brief
  2. 02

    Connect your assistant

    Use the setup guide for your client. Choose MCP, a local skill setup or the API to fit how you work.

    Context where the work happens
  3. 03

    Keep what you learn

    Ask your agent to save useful decisions when memory tools are available. Review them before relying on them next time.

    A better starting point next session

Choose the route that fits your work.

The playbook stays the same. The connection can change.

A few useful distinctions

Before you begin.

Read the setup guide
How is this different from chat history?

A playbook stores chosen instructions, skills and saved facts outside a single conversation. It does not automatically remember every chat. Your assistant needs to read the relevant context and explicitly save anything worth keeping.

Does AgentPlaybooks run the work for me?

Your AI client and its connected tools execute the work. AgentPlaybooks supplies reusable context and configuration. It does not turn a saved instruction into a scheduled job or grant your assistant extra permissions.

Where do my project details belong?

In your own private playbook. Public starters are examples anyone may read. Copy one before adding project information, check who has access, and keep credentials in the dedicated secret-management features.

Should I use MCP, skills or the API?

Use the route your client supports best. MCP offers structured tools inside a conversation; skills provide reusable instructions; the CLI and API suit file-based or automated workflows. Small, targeted reads matter more than the name of the connection.

Make the next conversation count

Give your next idea a head start.

Pick one project you keep explaining. Turn it into a playbook, and build from there.