7. Operating at Scale

The previous six chapters were all about a single Agent—how to configure it, tune it, trigger it, make it collaborate. Once that craftsmanship is done, new problems emerge:

• The Prompt you tuned—how do you let others use it too?
• The 5 Agents on your team—how do you keep their configurations consistent?
• An Agent went wrong—how do you send this session to a colleague to look at?
• 100 Workspaces piled up in the sidebar—how do you find the one you want?

This chapter covers everything in Neutree Agent Platform (NAP) related to “reuse, sharing, and organization.”

The Library trio: Prompts, Skills, Templates

Under Library in the sidebar there are three kinds of things, each solving one reuse problem.

Prompts

Store a system Prompt you use repeatedly in the Library. Other Workspaces can reference this Prompt, and once a reference is established:

• When the Prompt in the Library is updated, all referencing parties sync automatically
• A Workspace that wants to deviate from the baseline can “override,” after which it no longer follows updates (it can resume following at any time)

When to extract a Prompt into the Library:

• Multiple Workspaces use the same behavior —the classic case is standardizing a class of Agents across a team
• The same Prompt is being iterated —you don’t want to manually sync to 5 places every time you change it
• You want version management —Library Prompts support multiple versions, making it easy to trace back and compare

If only one Workspace uses it, there’s no need to extract it. Wait until there’s actually a second consumer.

Skills

Skills are also stored in the Library, with a similar mechanism. Guide 3 covered how to enable a skill in a Workspace. Here we cover how to create one:

Library → Skills → Create. You can choose two methods:

• Upload an archive —package SKILL.md together with any tool scripts into a zip and upload it
• Import from a Git repository —specify a repository address and path, and the platform pulls it down. You can re-sync after the repository is updated

A Skill’s content structure follows a convention:

my-skill/
├── SKILL.md       # describes what this skill does and which commands it provides
└── scripts/       # tool scripts
    └── ...

SKILL.md is what the Agent actually reads—it tells the Agent what capabilities this skill provides and how to use them. Writing a good SKILL.md is the core of writing a good skill.

Templates

A Template is a complete snapshot of a Workspace configuration—model, Prompt, Skills, MCP, settings, and resources all packaged together. A Workspace created from a Template directly owns this entire set of configuration.

When to extract a Workspace into a Template:

• You need to batch-create similar Agents —for example, giving everyone on the team a “translation Agent”
• You want to give new members a ready-to-use starting point —they only need to create from the Template, not configure from scratch
• The configuration is being iterated —after the Template is updated, Workspaces bound to it can be upgraded with one click

From the Workspace’s top menu → Save as Template. You can choose whether to bind the current Workspace to this new Template (binding lets it follow updates).

The relationship between Templates and Library Prompts

The two don’t conflict; they’re only complete when combined:

• Templates —manage the whole Agent’s “default persona”
• Library Prompts —manage fine-grained iteration of the Prompt item alone

A common team practice is: the Prompt field in the Template references a Library Prompt. This way the Template provides the overall configuration baseline, the Prompt can iterate independently, and after an update all Workspaces created from the Template receive the new Prompt.

Tags: organizing your Workspace list

Once dozens of Workspaces are piled up in the sidebar, finding one starts to get hard. Tags are NAP’s lightweight grouping tool.

In the sidebar → Manage Tags to create tags. Check the tags to apply from a Workspace’s menu. The tag button at the top of the sidebar quickly filters the list.

Tag design suggestions

• By purpose —production, staging, experiment
• By team —frontend, backend, data
• By status —active, archived, review

Colors are for at-a-glance differentiation. Tag filtering is OR logic—when multiple are selected, anything matching any one of them is shown.

Shared sessions

When debugging an Agent, you often need to send a session to a colleague to look at: you see the Agent took a wrong turn at some step and want a colleague to help diagnose it.

From the Workspace’s top menu → Share Session. This generates a public link that anyone can open to see the full conversation of this Session—messages, tool calls, file operations all visible.

Suitable scenarios:

• Asking for help —send the problematic session to someone on the team who knows it better
• Demos —show a business stakeholder a complete end-to-end flow
• Retrospectives —when an Agent performed especially well or especially badly, archive it for reference

Note that what you share is public—don’t share sessions containing sensitive information.

Workspace visibility and team collaboration

Guide 6 covered how a Workspace’s Visibility affects who can call it. The same field also affects who can see it in their own list:

• Private —only you can see it
• User —you can see it (it doesn’t appear in others’ lists)
• Public —it’s visible in all users’ Library

The Prompts, Skills, and Templates in the Library also have a similar Public/Private distinction. Public suits capabilities the team/platform wants to share; Private suits personal use or the experimental stage.

A common team pattern

1. Experiment in personal space —tune Prompts and try skills in a Private Workspace, experimenting freely
2. Extract to the Library once stable —put the Prompt into Public Prompts and the skill into Public Skills
3. Crystallize into a Template —save the mature Workspace configuration as a Public Template so everyone on the team can create from it with one click
4. Keep iterating —through the Library’s “auto-sync updates” mechanism, configuration improvements roll out automatically

This flow connects “personal exploration” and “team benefit”—what one person spends time tuning, the whole team gets to use.

Next

By now you’ve walked the full path from “creating your first Agent” to “turning it into a team-level capability.”

If you want to dive deep into a specific topic, return to the Concepts chapter for a panoramic explanation of NAP’s core concept groups.