Skip to main content
View source

Skills

View as Markdown

A RocketRide tool node that gives an agent custom skills: written instructions for one kind of task, in the SKILL.md format. Pick it when you want an agent to follow your own procedure, checklist or house style for a task without putting all of it in the agent's prompt.

What it does​

The node has no data lanes and binds to an agent's tools connection. It loads skills when the pipeline starts, from skills entered in the node and from a folder of skills in your RocketRide files. The agent sees each skill's name and description in the load_skill tool description, loads a skill's instructions only when the task matches it, and reads a skill's other files only when the instructions call for them. A skill therefore costs context only when the agent uses it, and one node can hold many skills. Use the Python, Daytona or Luxor Sandbox tool nodes alongside it when a skill's scripts need to run: this node returns files as text and runs nothing.

As a tool​

Tools are namespaced by the node id, so an agent sees tool_skills_1.load_skill.

FunctionDescription
load_skillReturn one skill's instructions and the list of files bundled with it. The tool description lists every loaded skill as - name: description. Required: name (one of the loaded skill names). Returns {name, description, instructions, files}.
read_skill_fileReturn one file bundled with a folder skill as UTF-8 text. Required: name and path, a path from the skill's files list (example: references/style.md). Returns {name, path, content}. Offered only when at least one loaded skill has bundled files.

read_skill_file reads only the files listed for the skill, so a path outside the skill folder, or SKILL.md itself, is refused. A file over 256 KB, or one that is not UTF-8 text (an image, for example), raises an error. When no skill loads, the node offers no tools at all and logs a warning.

Configuration​

Enter skills in the node for short instructions that belong to one pipeline. Use a skills folder for skills you share across pipelines, edit outside the canvas, or that bundle reference files. Both can be used together.

Skills​

Each row is one skill with three fields: Name, Description and Instructions. Paste a whole SKILL.md, frontmatter included, into Instructions and leave Name and Description empty to take them from the frontmatter. Fill in Name or Description to override the frontmatter value.

The name is converted to lowercase letters, numbers and hyphens (Release Notes becomes release-notes), up to 64 characters. The description is what the agent reads when it decides whether to load the skill, so say what the skill does and when to use it: "Write release notes from a list of merged pull requests. Use when asked for a changelog or release summary." Descriptions longer than 1,024 characters are cut, and a name or description longer than 8,192 characters is refused. An empty row is ignored; a row that lacks a name, a description or instructions is skipped with a warning.

Instructions​

Markdown the agent follows after loading the skill. Write steps, rules and examples the way you would brief a colleague. Instructions entered in the node cannot bundle files; use a skills folder when the instructions refer to reference documents or templates.

Skills folder​

A folder in the RocketRide file store, the same store the File System node reads. The folder can hold one subfolder per skill, or be one skill itself:

skills/
release-notes/
SKILL.md
references/style.md
templates/entry.md
triage/
SKILL.md

Each SKILL.md needs a description in its frontmatter; the name defaults to the subfolder name. Every other file in the skill folder, up to 200 files and five folder levels deep, becomes readable through read_skill_file. A subfolder without a SKILL.md, or with one that cannot be parsed, is skipped with a warning naming it. When a skill in the folder has the same name as a skill entered in the node, the node's skill is kept.

Development runs read the folder from your own files. Deployed runs read it from the deployment's files, so copy the folder there before you deploy a pipeline that uses one.

Notes​

SKILL.md format​

A SKILL.md starts with YAML frontmatter between --- lines, then the instructions:

---
name: release-notes
description: Write release notes from a list of merged pull requests. Use when asked for a changelog or release summary.
---

# Release notes

Group the changes by area. Follow references/style.md for wording.

Frontmatter keys other than name and description are ignored. A file without frontmatter is all instructions, which works for skills entered in the node when Name and Description are filled in.

name and description must be text. YAML reads yes, no, on, off, true and false as true or false and a bare number as a number, so put such a value in quotes (name: "yes"). The frontmatter may not use YAML anchors and aliases (&a, *a). A skill that breaks either rule is skipped with a warning that says why, and the other skills still load.

When skills change​

Skills are read when the pipeline starts. Restart the pipeline after you change a skill in the folder.

Upstream docs​

Schema​

FieldTypeDescriptionDefault
tool_skills.folderstringSkills folder
Folder in your RocketRide files that holds skills: one subfolder per skill, each with a SKILL.md. Example: skills. Leave empty to use only the skills entered below.
""
tool_skills.skillDescriptionstringDescription
What the skill does and when to use it. The agent reads this to decide whether to load the skill. Leave empty to use the description in the SKILL.md frontmatter.
""
tool_skills.skillInstructionsstringInstructions
The skill's instructions in markdown, or a whole SKILL.md including its frontmatter.
""
tool_skills.skillNamestringName
Lowercase letters, numbers and hyphens, for example release-notes. Leave empty to use the name in the SKILL.md frontmatter.
""
tool_skills.skillsarraySkills
Skills to give the agent. Paste a whole SKILL.md into Instructions, or fill in the name and description and write the instructions.

Dependencies​

  • pyyaml