Runstack Docs

Runstack Studio

Build, configure, and publish AI agents in Runstack Studio. Invoke any published agent instantly in chat using the /agentName slash command.

Runstack Studio is where you build and manage agents — pre-configured AI personas with a defined role and an ordered set of tools. Once published to the marketplace, any agent is callable in chat with a single /command.

What is an agent?

An agent in Runstack Studio has:

FieldDescription
NameDisplay name. Also determines the /command — e.g. GitHub Helper becomes /githubHelper
RoleSystem prompt that defines the agent's persona and behavior
TagCategory for filtering in the marketplace (e.g. Developer Tools, Productivity)
Tool sequenceOrdered list of connector + tool pairs the agent prioritizes
VersionSemver string displayed on the agent card (e.g. v1.0)
PublishedWhether the agent appears in the marketplace and is callable via /command

Creating an agent

  1. Go to Studio in the sidebar
  2. Click Create
  3. Fill in the fields:
    • Name — keep it short and descriptive (e.g. PR Reviewer, Notion Writer)
    • Role — write the system prompt that tells the AI how to behave
    • Tag — pick the category that fits best
    • Tools — add tool steps in execution order (connector → tool name)
  4. Click Save — the agent is saved as a draft
  5. Toggle Published to make it visible in the marketplace and callable via /command

The /command system

Every published agent gets a slash command derived from its name using camelCase:

Agent nameSlash command
GitHub Helper/githubHelper
Notion Writer/notionWriter
PR Reviewer/prReviewer
Send Email/sendEmail

Type / in any chat input to see a live suggestion list of all available marketplace agents. Select one and press Enter — the agent's role and tool sequence are automatically applied to that message.

> /prReviewer Review the latest PR in my-org/my-repo

The AI will:

  1. Use the PR Reviewer agent's role as its system prompt
  2. Execute tools in the order defined in the agent's tool sequence
  3. Respond with the review

Designing good agents

Role (system prompt)

The role is the most important field. Write it as a specific instruction set:

You are a GitHub PR reviewer. When given a repository and PR reference:
1. Use search_tools to find github_get_pull_request and related tools
2. Fetch the PR diff and changed files
3. Review for: correctness, security issues, naming conventions, and test coverage
4. Return a structured review with LGTM / Needs Changes verdict
Always be concise and actionable.

Tool sequence

The tool sequence tells the AI which tools to prioritize and in what order. This is not a strict lock — it's guidance. A good sequence for a PR reviewer might be:

StepConnectorTool
1GitHubgithub_get_pull_request
2GitHubgithub_list_pr_files
3GitHubgithub_get_commit
4GitHubgithub_add_pr_review

Versioning

Bump the version string (e.g. v1.0v1.1) when you update an agent's role or tool sequence so users in the marketplace can see it has changed.

Marketplace

The Marketplace tab in Studio shows all published agents — both yours and those published by other users. Browse by tag or search by name/role.

Marketplace agents are available to everyone as /commands in chat. Draft (unpublished) agents are only visible to you under Mine.

Using agents via the MCP server

Published agents are not yet directly callable through the /mcp endpoint — they run through the Runstack chat system. To trigger agent-style behavior from an MCP client, use the meta tools directly:

search_tools → check_authentication → execute_tool

And write the agent's role as the system prompt in your own LLM call. A future version of Runstack will expose Studio agents as named MCP tools.

On this page