AI assistants and skills
Connect an LLM, write an actionable prompt, and give your project reusable authoring instructions.
An assistant combines a model, project tools, and instructions. The model interprets your request; MCP tools perform authorized operations; skills describe how you want the work done. A skill alone cannot connect to Octonode or execute a workflow.
Otto already includes the hosted Octonode MCP connection and the code-author skill. External clients need their own connection or the Octonode Connect client plugin; they do not need Otto's runtime.
Work in Studio
Open Assistant in Studio and select the workspace and project for the session. Ask it to inspect the current project before requesting changes. Existing source, node signatures, and workflow graphs give it concrete context.
Open Otto connection to choose your provider, model, and saved secret variable. Otto's project terminal does not require Claude Code login. Configure external AI clients in their own apps and follow MCP setup.
Start with a useful build prompt
This prompt gives the assistant a specific result, acceptance criteria, and an explicit boundary for external actions. Adapt the input and expected output to your project.
Build an order-total workflow in the selected Octonode project.
First inspect the project files, existing nodes, and workflow graph through MCP.
Read the project's relevant skills and reuse existing code where possible.
Input: { "a": 2, "b": 3 }.
Behavior: add a and b, then square the sum.
Expected terminal output: { "result": 25 }.
Implement the smallest TypeScript change, with JSON-compatible inputs and outputs.
Use the latest file and graph revisions for writes. Validate after editing.
Show the changed files, final node connections, and validation result.
Do not publish, deploy, or call external services without a separate instruction.To permit a test execution, follow up with: "Run this workflow once with the example
input and report its run ID and terminal node output." Its key must include
workflows:run. The workflow API may wrap the node output in run events or
node-keyed results; the prompt's expected object describes the terminal node.
Create a project skill
Open AI resources → Add → New skill and add your instructions. Use Add → Upload local plugin to upload a ZIP containing SKILL.md files and Markdown references, or a supported plugin bundle. Imports attach to the current session when compatible with its selected model. Search, filter, and sort the resource table to review, attach, detach, or remove resources. You can edit resources you own; edits keep the same ID and require the current revision.
A skill has YAML metadata followed by Markdown instructions. For example:
---
name: workflow-review
description: Review a workflow's inputs, connections, and failure behavior before execution.
---
Read the selected workflow and the source of its nodes through Octonode MCP.
Check that every connected output is compatible with the receiving input.
Identify required environment variables by name, without reading secret values.
List external effects, retry behavior, and a small example input.
Report findings with the affected node IDs. Do not modify or execute the workflow.Download the ready-to-use workflow-review/SKILL.md,
then register its instructions or use the development terminal command below.
Click a resource's name and choose Use in this session to include it in the session. Review and update its instructions as the project changes. Keep credentials in Global Variables and reference their names in MCP configuration.
Manage resources in the development terminal
For a managed cloud project, enable Developer mode and open Terminal. The Octonode CLI is already
installed and connected to the selected project's Otto resource registry. The
terminal provides both octonode and octonodes command names.
octonodes otto resources list
octonodes otto mcp add docs --url https://example.com/mcp --variable MCP_KEY
octonodes otto skills add workflow-review --file ./SKILL.md
octonodes otto plugins add review-bundle --file ./resource.json
octonodes otto resources update RESOURCE_ID --file ./registration.json
octonodes otto resources remove RESOURCE_IDPlugin JSON contains skills, connectors, and optional text files, matching
the resource table's reviewed contents. New CLI resources load automatically in
this project's sessions; add --manual to attach them individually. Use
--id UUID when retrying a registration. The terminal's resource connection
expires after 15 minutes; reopening it refreshes the connection.
Remote computer terminals run your host's normal shell. They do not receive this automatic Otto resource connection or show these managed cloud shortcuts.
Skills, plugins, and MCP are different
| Item | What it contains | When it is used |
|---|---|---|
| Project skill | Markdown instructions and references | The assistant follows a repeatable process |
| MCP server | Discoverable tools and resources | The assistant reads or changes authorized project state |
| Octonode workflow plugin | Node definitions, TypeScript handlers, and connection requirements | The workflow runtime executes nodes |
| AI client plugin | Client-specific skills, tools, or other extensions | Your AI client loads its extension package |
Installing an AI client plugin does not automatically install an Octonode workflow plugin. Attach workflow plugins to the target project and configure their service connections. See Nodes and plugins.
The Octonode Connect client plugin packages connection guidance and the hosted MCP dependency. It does not contain the MCP implementation or install workflow nodes. Follow MCP setup for authentication and project scope.
Understand the workspace boundary
Project source edits happen through project tools. A terminal can be a separate scratch environment; do not assume a terminal file edit changed the project's source. Inspect the project again through MCP to confirm the result.
Terminal files can be lost when the environment is recycled. Registered resources remain in Otto’s workspace registry and are snapshotted for each run. Background chat does not execute shell commands or hooks simply because a skill mentions them; the available tools and permissions determine what can run.
Customize Otto
Open AI resources to manage Skills, Connectors, and Plugins. Search, filter and sort the table; click a resource to review it, use it in the current session or delete it. Owners can edit resources and toggle Enabled. Disabling a resource excludes it from future runs without deleting it.
Choose Add → Upload local plugin to drop a ZIP containing skills or a Claude plugin. Browse marketplace loads Claude’s public catalog automatically. Adding a marketplace plugin registers it; click its name and choose Use in this session when you want to attach it. Otto checks source and runtime compatibility when you add a plugin; some plugins need login, transports or features that are not supported yet. Provider secrets stay in Global Variables.
When you connect a project, Otto reads .agents, AGENTS.md, and CLAUDE.md in
the background and shows the files it read. It also discovers .claude/skills.
Guidance is reread from the selected checkout before every run. Background reads
include up to 32 files and 16,000 UTF-8 bytes per project; all selected projects
share a 16,000-byte run context budget. Skipped files can be read through authorized
project tools. Project guidance cannot grant new permissions.