Agent Serve · Playbooks
Write down how your team does a job once. Agent Serve agents load that playbook in Chat and workflows whenever the work matches.
Agent Serve Playbooks are reusable markdown instructions in Agent Serve, the AI workspace where teams build, deploy, and manage AI agents. Each playbook has a name, a description and up to 50,000 characters of instructions. Agents read the description, load the full playbook when it fits the job, and teams call one directly with a slash in Chat.
How it works
An Agent Serve playbook is a short description plus the instructions behind it. Agents decide when to open it.
Give it a name, a description of when to use it, and the steps in markdown.
Import a markdown file with frontmatter, or a zip of up to 5 MB that holds a SKILL.md.
Pick playbooks in an Agent block, or type a slash and the playbook name in Chat.
The agent sees each description and loads the full instructions only when the job needs them.
In depth
Agent Serve uses one playbook library in Chat and in workflows, so a process is written once and followed everywhere.
In Agent Serve Chat, type a slash and the playbook name, or choose one from the Playbooks button in the message box.
The Skills field on the Agent Serve Agent block attaches playbooks to that agent, or creates a new one without leaving the workflow builder.
Every Agent Serve playbook is shared with its team space. Only its editors and workspace admins can change or delete it.
The library
Agent Serve agents read each playbook’s description first, so a clear “use this when” line decides when it runs.
Use when drafting a credit memo for committee
An Agent Serve playbook is markdown: steps, checklists, tone and the output format the team expects.
## Steps · ## Checks · ## Output
Agent Serve imports a .md file with YAML frontmatter, or a .zip that contains a SKILL.md.
kyc-review.zip → SKILL.md
Agent Serve ships four built-in playbooks: connect an integration, research, create a table and deploy a workflow.
research · create-table
Add editors to an Agent Serve playbook from its Editors card. Everyone else in the team space can use it.
Editors · priya@ · sam@
Agent Serve loads a playbook’s full instructions only when an agent asks for them, keeping each prompt short.
Description first, content when needed
Examples
Three playbooks financial services teams keep in Agent Serve.
An Agent Serve playbook sets the memo’s sections, the ratios to compute and the risks to call out, so every analyst’s draft looks the same.
An Agent block in an Agent Serve workflow loads the KYC playbook for each new file and records what is missing.
An Agent Serve playbook keeps the tone, the required statements and the escalation rules for complaint replies in one place.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| How a job is done | BeforeIn the heads of the two people who have done it longest. | With Agent ServeWritten as an Agent Serve playbook the whole team space can read and use. |
| Prompting an AI | BeforeEach person pastes a long prompt they keep in a note. | With Agent ServeType a slash and the playbook name, or let the agent pick it by description. |
| Changing the method | BeforeTell everyone, then hope each copy of the prompt gets updated. | With Agent ServeAn editor updates the playbook once, and every agent using it follows. |
By the numbers
50,000
Characters of instructions per playbook
1,024
Characters for the description agents read
4
Built-in playbooks
5 MB
Largest zip you can import
Limits are read from the product code.
Under the hood
| Fields | Name (lowercase words joined by hyphens, up to 64 characters), description and markdown content. |
|---|---|
| Where it lives | The Playbooks tab beside Integrations. |
| Scope | Shared with the whole team space. Names are unique per space. |
| Use in Chat | Slash command, the Playbooks button, or picked by the agent from its description. |
| Use in workflows | The Skills field on the Agent block. |
| Import | A .md file with YAML frontmatter, or a .zip up to 5 MB with a SKILL.md. |
| Who can edit | Named editors and workspace admins. Creating one needs write access. |
| API | Create, read, update and delete through the Agent Serve API. |
Governance
Agent Serve treats playbooks as shared team assets, with editors, admin override and a switch to turn them off.
Security and governanceAnyone in an Agent Serve team space can use a playbook. Only its editors and admins can change it.
Agent Serve access policies can block agents from loading playbooks for a permission group.
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Sign in through your identity provider with SAML 2.0 or OIDC, and provision people with SCIM.
Works with
Call a playbook with a slash, or let Chat pick one by its description.
Attach playbooks to an Agent block so each run follows the same method.
Playbooks say how to do the job; knowledge bases hold the documents to cite.
Playbooks sit beside Integrations, and a built-in one helps connect a new tool.
Questions
An Agent Serve playbook is a named set of markdown instructions with a description of when to use it. Agents in Chat and workflows load it when a job matches.
An Agent Serve playbook tells an agent how to do a job. A workflow wires the steps, triggers and actions together in the visual builder. An Agent block in a workflow can use playbooks.
Yes. Agent Serve imports a markdown file with YAML frontmatter, or a zip of up to 5 MB that contains a SKILL.md file.
Not today. Agent Serve keeps the current version of each playbook and when it was last updated.
Last updated
Enterprise access
Agent Serve is available to teams by invitation. Tell us what you want to build and we will set up your workspace with you.