Skills Reference

Skills are MCP (Model Context Protocol) tool packs that extend the agent with capabilities beyond standard shell commands. When you enable a skill in the Task Composer, its tools become available to the agent during that task.

Stdio-based skills run inside the same sandbox container as the task's run_shell commands, so they share the task filesystem view and the project's sandbox network policy.

Some skills expose deterministic tools directly inside Meowbert (for example DOCX Studio and PPTX Studio). Others are instruction-first workflows that give the agent curated scripts, references, and validation steps for a specific kind of work (for example the imported OpenAI office skills).

Skills are defined in the skills/ directory and loaded by the worker at startup. Each skill has a manifest (skill.json) and documentation (skill.md) shown to the agent when the skill is active.

Choosing the right skill

Use Canvas when you want to build presentation-style or page-style visual documents directly in HTML/CSS and render them into previews or PDFs with Playwright.
Use Image Generation when you want an OpenAI-compatible image generation endpoint to create a new image and save it as a task file.
Use DOCX Studio when you want deterministic Word tools for templates, placeholders, report generation, and SVG-based figures.
Use PPTX Studio when you want deterministic PowerPoint tools for editable slide decks, branded templates, metric cards, and visuals.
Use Google Workspace when a directly attached Google Doc, Sheet, or Slides presentation should be read or edited in its original shared location.
Use OpenAI: doc when the task is broader document work that benefits from shell/Python workflows and render-then-review loops.
Use OpenAI: slides when you want a curated PptxGenJS workflow with render scripts and helper references.
Use OpenAI: spreadsheet for workbook analysis, formula-aware editing, formatting, and spreadsheet review workflows.
Use OpenAI: pdf for PDF generation, extraction, and page-by-page visual validation workflows.

Available Skills

Browser Use

ID: browser-useRequires: Super-admin installation

Automates web browsers using Playwright. Useful for tasks that need to interact with websites — filling forms, navigating pages, extracting content, or taking screenshots.

Example use cases:

"Log into the admin panel and export the latest report."
"Take a screenshot of the homepage on mobile viewport."
"Fill in the registration form and confirm submission."

How to enable: First set skills.browserUse.enabled: true in your server config and rebuild the sandbox runtime image, then select Browser Use in the Skills selector in the Task Composer wrench menu.

This skill is bundled and pinned in the sandbox runtime image when skills.browserUse.enabled is on, and is typically restricted to super-admin accounts. Browser Use now launches through a local wrapper that resolves the bundled Chromium executable path explicitly for more reliable startup inside the sandbox. Contact your platform admin to enable it for your workspace.

Canvas

ID: html-canvas

Creates slide-style and page-style visual documents in HTML/CSS/JS, then renders them locally with Playwright for preview and final export. This is useful when you want editable browser-native layouts instead of generating Office files directly.

Example use cases:

"Make a polished 16:9 HTML deck and export preview PNGs plus a PDF."
"Build a two-page report layout in HTML and check every page before delivery."
"Create a visual handout with custom CSS, then render it to PDF from the same source file."

Available tools include:

Scaffold bundled starter templates for slide decks and report pages.
Render HTML into per-page or per-slide PNG previews.
Generate a preview-index.html contact view for quick scanning.
Export a final PDF from the same HTML source after the preview pass is clean.
Use bundled reportlab when the workflow needs deterministic PDF post-processing after HTML export.

How to enable: Set skills.htmlCanvas.enabled: true in your server config, rebuild the sandbox runtime image, then select Canvas in the Task Composer wrench menu under Core tools.

Canvas depends on the same Playwright Chromium bundle used for browser rendering workflows. When either skills.browserUse.enabled or skills.htmlCanvas.enabled is on, the sandbox runtime image bundles Chromium.

Image Generation

ID: image-generation

Generates a new image through the configured OpenAI-compatible /images/generations endpoint and writes the result to a file in the task directory.

Example use cases:

"Generate a transparent app icon and save it as assets/icon.png."
"Create a product mockup image for this report."
"Make a background image for the Canvas deck and save it in the task files."

Available tool:

Generate one image with model gpt-image-2, output count 1, and a caller-specified filename.

How to enable: Select Image Generation in the Skills selector in the Task Composer wrench menu.

Configure the provider in skills/image-generation/image-generation.config.json. The config follows the same provider shape used by Deep AI Search, including baseURL, apiKey, apiKeyEnvVar, headers, name, retry count, and timeout.

DOCX Studio

ID: docx-studio

Creates and manipulates Word documents (.docx files). Provides deterministic tools for generating polished documents from templates, embedding richer SVG figures, filling placeholders, patching existing files, and fetching reusable templates from the web.

Example use cases:

"Generate a strategy memo and include a process diagram in section 2."
"Inspect the placeholder tokens in invoice_template.docx and tell me what fields need to be filled."
"Find a DOCX template for a project brief and download it into the workspace."

Available tools include:

Create a new DOCX document with structured content and optional SVG figures.
Draw standalone SVG figures for reports and briefs.
Fill a branded template by replacing placeholder tokens.
Inspect an existing DOCX to list placeholder tokens.
Patch specific sections or paragraphs in an existing file.
Search for and download reusable DOCX/DOTX templates.

How to enable: First set skills.office.enabled: true in your server config and rebuild the sandbox runtime image, then select DOCX Studio in the Skills selector in the Task Composer wrench menu.

DOCX Studio depends on LibreOffice for preview/render flows. When skills.office.enabled is off, the sandbox runtime image skips bundling LibreOffice and this skill is hidden.

PPTX Studio

ID: pptx-studio

Creates and manipulates PowerPoint presentations (.pptx files). Provides deterministic tools for generating better-looking slide decks, drawing detailed SVG visuals, filling branded placeholders, and fetching reusable templates from the web.

Example use cases:

"Create a 10-slide pitch deck with roadmap and architecture visuals."
"List the placeholder tokens in branded_deck.pptx so I know what to fill."
"Find a pitch deck template, download it, and then fill it with our company data."

Available tools include:

Create a new PPTX presentation with optional SVG visuals and metric cards.
Draw standalone SVG visuals for slides.
Fill a branded template with content by replacing placeholder tokens.
Inspect an existing PPTX to list placeholder tokens.
Search for and download reusable PPTX/POTX templates.

How to enable: First set skills.office.enabled: true in your server config and rebuild the sandbox runtime image, then select PPTX Studio in the Skills selector in the Task Composer wrench menu.

PPTX Studio depends on LibreOffice for preview/render flows. When skills.office.enabled is off, the sandbox runtime image skips bundling LibreOffice and this skill is hidden.

Google Workspace

ID: google-workspace

Reads and edits Google Docs, Sheets, and Slides that were attached directly from the Google Drive Source.

Best for:

Editing the original shared Google Doc instead of a converted DOCX copy.
Updating ranges and formatting in an attached Google Sheet.
Reading or changing slides in an attached Google Slides presentation.
Reusing an authorized Google file across tasks in the same Project.

How to enable: Attach a native Google Doc, Sheet, or Slides presentation from Google Drive and choose Attach Google files directly. This adds the file to the Project's Google Workspace allowlist, and the skill is enabled automatically for tasks in that Project.

Direct attachment is Experimental and changes the original shared Google file immediately. The skill can access only files deliberately added to the current Project's allowlist; connecting Google Drive or copying a pointer file does not authorize other files.

OpenAI: doc

ID: openai-doc

Imported OpenAI curated workflow for .docx work. Focuses on shell/Python-based document creation, editing, rendering, and visual review with the bundled helper scripts and references copied from openai/skills.

Best for:

Editing or generating Word documents with a shell/Python workflow instead of deterministic MCP-only tools.
Render-first review loops where the agent should convert DOCX files to page images and inspect layout before delivery.
Tasks that need bundled helper scripts, installation guidance, and document-quality checklists.

How to enable: Select OpenAI: doc in the Skills selector in the Task Composer wrench menu.

OpenAI: spreadsheet

ID: openai-spreadsheet

Imported OpenAI curated workflow for spreadsheet creation, analysis, formatting, recalculation guidance, and visual review using openpyxl, pandas, and the bundled references.

Best for:

Building or editing .xlsx, .csv, and .tsv files with formula-aware workflows.
Spreadsheet analysis tasks that mix pandas data work with polished workbook output.
Review flows where the agent should preserve formulas, formatting, and cached values where possible.

How to enable: Select OpenAI: spreadsheet in the Skills selector in the Task Composer wrench menu.

OpenAI: slides

ID: openai-slides

Imported OpenAI curated workflow for PowerPoint authoring with PptxGenJS, render/validation scripts, and bundled layout helpers copied from openai/skills.

Best for:

Agent-authored slide decks where you want editable .pptx output plus the underlying authoring script.
Deck recreation or QA tasks that benefit from bundled rendering, montage, overflow, and font-detection scripts.
Presentation work that needs a guided workflow more than a narrow deterministic tool surface.

How to enable: Select OpenAI: slides in the Skills selector in the Task Composer wrench menu.

OpenAI: pdf

ID: openai-pdf

Imported OpenAI curated workflow for PDF generation, extraction, rendering, and visual review using shell/Python utilities.

Best for:

Reading or producing PDFs where visual fidelity matters as much as extracted text.
Tasks that should render PDF pages to images and verify spacing, clipping, tables, or typography.
Workflows that combine reportlab, pdfplumber, pypdf, and Poppler-style review tools.

How to enable: Select OpenAI: pdf in the Skills selector in the Task Composer wrench menu.

Example (testing only)

ID: example

A simple skill for testing that the skills system is working correctly. Provides an echo tool and a random number generator. Not useful for real tasks.

Configuring the skills directory

Skills are loaded from the directory specified in your config:

json

{
  "skills": {
    "rootDir": "./skills"
  }
}

Each subdirectory in rootDir that contains a valid skill.json is loaded as a skill.

Adding a custom skill

To add a new skill:

Create a subdirectory under skills/ with a unique name (e.g., skills/my-tool/).

Create skill.json with the following structure:

json

{
  "id": "my-tool",
  "name": "My Tool",
  "description": "What this skill does.",
  "mcp": {
    "transport": "stdio",
    "command": "node",
    "args": ["server.mjs"],
    "cwd": "{{SKILL_DIR}}"
  }
}

is replaced at runtime with the absolute path to the skill's directory.
requiresSuperAdmin: true can be added to restrict visibility to super-admin accounts.

Create skill.md with documentation shown to the agent when the skill is active.
Implement the MCP server (e.g., server.mjs) that exposes the tools.
Restart the worker to load the new skill.

The new skill will appear in the Skills selector in the Task Composer.

Skills Reference ​

Choosing the right skill ​

Available Skills ​

Browser Use ​

Canvas ​

Image Generation ​

DOCX Studio ​

PPTX Studio ​

Google Workspace ​

OpenAI: doc ​

OpenAI: spreadsheet ​

OpenAI: slides ​

OpenAI: pdf ​

Example (testing only) ​

Configuring the skills directory ​

Adding a custom skill ​

Skills Reference

Choosing the right skill

Available Skills

Browser Use

Canvas

Image Generation

DOCX Studio

PPTX Studio

Google Workspace

OpenAI: doc

OpenAI: spreadsheet

OpenAI: slides

OpenAI: pdf

Example (testing only)

Configuring the skills directory

Adding a custom skill