Skills

Skills are how you teach your AI agents new capabilities without writing prompts from scratch every time. A skill packages up the know-how for a particular kind of work — how to run deep research, how to produce branded documents, how to summarize a web page — and lets you grant that capability to any agent with a single click.

Skills work hand-in-hand with Building AI Agents: you define an agent's personality and purpose there, then use skills to extend what it can actually do.

What is a skill?

A skill is a reusable bundle of know-how you grant to an AI agent: a set of instructions (how to do something well) plus the tools it needs. Granting a skill is one click — the agent then has that capability.

To keep the agent fast, it only sees a skill's name and description until it actually needs it. When a request comes in that the skill applies to, the agent "loads" the full skill on demand. This is called progressive disclosure — an agent can carry dozens of skills without slowing down, because the detailed instructions only enter its context when they're relevant.

Note: Skills follow the Agent Skills open standard — the same SKILL.md format used by Claude Code, Codex, and the 1,000+ community skills. That means you can import skills from the ecosystem, and skills you build in Portablemind are portable. Learn about the standard at agentskills.io.

Installing a ready-made skill

The fastest way to add a capability is the Skill Catalog in AI Studio. Browse the catalog, pick a skill (for example Deep Research or Office Documents), and install it — it then appears among your workspace's skills, ready to be granted to an agent.

Each installed skill's detail view lists example requests, so you can see exactly what kind of request activates it — just say one to an agent that has the skill.

What's in the catalog

The catalog currently offers these ready-made skills:

Skill	What it does
office-documents	Generate PowerPoint (.pptx) decks and print-quality PDFs from Markdown, or convert Office documents — themed and ready to share.
deep-research	Deep multi-source web research: fan out searches, cross-check sources, and synthesize a cited report (optionally as a PDF).
diagram	Turn a plain description into a polished on-brand diagram — flowchart, architecture, sequence, ER, or state — as PDF, PNG, or SVG.
infographics	Polished on-brand poster-style infographics (architecture, positioning, process, comparison) built from professional templates, as PNG or PDF.
demo-video-recorder	Captioned product walkthrough videos on demand.
file-organization	Audit and reorganize the workspace's Files tree: sweep loose files, consolidate duplicates, archive stale artifacts.
agent-fleet-review	Audit every agent in the workspace — personas, prompts, tools, health, cost — with per-agent recommended actions.
agent-model-audit	Audit agents against real usage and the model catalog; recommend the most cost-efficient model per agent, with projected savings.

The catalog is growing — more skills are on the way.

Importing a skill from the ecosystem

Found a skill on agentskills.io or elsewhere? You can bring it into Portablemind directly. In AI Studio's Skills area, use Import SKILL.md and either paste the SKILL.md text or upload a .md file or a .zip bundle.

Before anything is created, you get a preview showing the skill's name, description, and an honest tier badge telling you what will actually run in Portablemind (see What runs where below). Confirm the import, then grant the skill to an agent.

A SKILL.md is just text — YAML frontmatter plus a Markdown body:

---
name: web-summarizer
description: Summarize a web page
---
# How to use
Fetch the page and summarize the key points.

What runs where — the three tiers

Skills are honest about what actually runs. When you import one, it's tagged with a tier so there are no surprises:

Tier	What it means
native	Instructions only — a playbook. No bundled scripts, so it runs as-is, immediately.
python	Bundles a Python script — imported as a playbook, but the script only runs after a platform admin reviews and approves it.
requires-silolink	Needs a real coding environment (shell, CLI, browser). Imported as a playbook; the code runs via a SiloLink session on your own connected machine — your sandbox, your control.

Every skill card shows its tier badge, with an explanation of what will run. (Separately from these import tiers, skills you build in Portablemind can bundle sandboxed JavaScript scripts that run with no review — see Building your own skill below.)

Assigning skills to agents

Skills do nothing on their own — they take effect when you grant them to an agent. You can do this from the Skills section of the agent's edit dialog, or when creating an agent — the creation wizard's Tools & Roles step includes a skills picker. Each skill's card also shows which agents it's currently granted to. Once granted, the skill's name and description become part of what that agent knows it can do.

To use a granted skill, just talk to the agent in plain language — in AI Assistant or any conversation the agent is part of. The agent recognizes when the skill applies, loads it, and does the work. You don't need to know any commands or parameters.

Tip: With the Deep Research skill granted, say "Do deep research on the open-source agent frameworks landscape and give me a cited report." The agent loads the skill and runs the research.

Building your own skill

You can create skills from scratch in AI Studio. Give the skill a name, a description, and instructions (the Markdown playbook). Then optionally add:

• Scripts (sandboxed JavaScript) — code that runs in a secure isolate with no review needed: it can make safety-checked web requests and use your workspace's data, files, and email helpers, but can't reach the server or other tenants. A granted agent calls it as a tool. Write a JS body that reads its arguments from the params object and returns a value, e.g. return { total: params.values.reduce((a,b)=>a+b,0) };
• Reference files — brand templates or example payloads the skill can use (for example a PowerPoint template for branded decks).
• Built-in tools / Python scripts — wire in approved platform tools.

Note: Need something heavier than sandboxed JavaScript? Connect a SiloLink session and a skill can run real code on your own machine — see Coding Agents.

Coding agents can also work the other way around: an MCP client such as Claude Code can fetch any of your Portablemind skills as a SKILL.md and run it locally. For API-level details, see the API guide.