logo
Kurashizu Blogwhere ideas flow
  • Home
  • News
  • Blog
  • About
  • Home
  • News
  • Blog
  • About

© 2026 Kurashizu. All rights reserved.

Admin·New Post·Service Status
Powered by Cloudflare·31.10.26
← Back to news

ChatGPT Work Tool and Skill Reference

Original article
codex-tool-reference.simonw.chatgpt.site|110 points|43 comments|by ijidak|Aug 31, 2026

Codex Work Tool and Skill Reference Guide

This document serves as a comprehensive directory of the callable interfaces and reusable workflow definitions currently active in this Work session.

📊 Session Inventory Snapshot

Date of Snapshot: 31 August 2026

MetricValue
Tool Interfaces232
Skill Definitions44
Skill Source Volume615×103615 \times 10^3615×103 characters

Definitions:

  • Tools: These are the actual callable endpoints.
  • Skills: These are instructional packages that provide the logic and guidance on how to utilize those tools.

🛠️ Skill Specifications

Each skill listed below corresponds to a page containing the full SKILL.md source. The reference maintains all TypeScript declarations and exposed descriptions. Note: Availability may fluctuate based on your permissions, connected applications, plugins, and session settings.

🎨 Visuals & Document Generation

  • answers-charts: Deploy when a visual representation (trends, compositions, categorical comparisons, or numeric data) is more effective than a text-based explanation.
  • answers-images: Use to provide visual grounding or enhance an answer using retrieved imagery.
  • answers-learning: Trigger this for interactive educational widgets, such as:
    • Objective multiple-choice quizzes.
    • Flashcards.
    • Pronunciation assistance.
  • documents: Used for the creation, modification, and redlining of .docx, Word, or Google Docs artifacts.
    • Workflow Requirement: Must follow a strict render-and-verify process.
    • Execute render_docx.py to create PNGs (and PDFs).
    • Perform visual QA.
    • Iterate on layout.
    • Deliver final document.
  • imagegen: For creating or editing raster/bitmap visuals (photos, textures, mockups, sprites).
    • Use for: Brand new images or variants of references.
    • Do not use for: Editing SVGs, vector assets, logo systems, or building visuals via HTML/CSS/Canvas.
  • pdf: For reading, creating, and inspecting PDF files (including AcroForms).
    • Tech Stack: Utilizes Poppler rendering alongside reportlab, pdfplumber, and pypdf.
  • Presentations: Specifically for PowerPoint (.ppt, .pptx) or Google Slides requests.
  • Spreadsheets: For any task involving .xlsx, .xls, .csv, or .tsv files, including Google Sheets. Handles formulas, charts, and recalculations.

⚡ Interactive & Textual Tools

  • visualize: Proactively create interactive simulations, maps, or graphs to explain "what happens if" scenarios or to help a user inspect a concept.
  • writing-blocks: Used for short-form drafts (emails, bios, social posts).
    • Note: For long-form content, prefer creating a standalone file over a writing block.

🌐 Web & Site Management

  • control-browser: Manages authenticated or public web interactions.
    • Authentication: Use only the browserAuth capability for credentials.
    • Priority Logic:
      1. Web Search →\rightarrow→ 2. Plugins →\rightarrow→ 3. Browser.
  • sites:sites-building: Used for constructing portfolios, dashboards, or internal hubs.
    • Mandatory: Use this skill if the project contains a .openai/hosting.json file.
  • sites:sites-preview-troubleshooting: Used to recover or diagnose failures when a site preview fails to launch or become healthy.

🧠 Context & Library Integration

  • demos:answers-ask-user-input: Used when a small gap in context exists; provides the user with tappable options or a text field to refine the answer.
  • openai-library:library: The primary interface for managing the ChatGPT Library.
    • Capabilities: Searching/reading Library content, organizing folders, version restoration, and saving new deliverables.
    • Trigger: Use when the user mentions the "Library" or when a reusable artifact needs to be created.
  • personal-context: Activated under two specific conditions:
    1. Continuity: When the user uses elliptical language (e.g., "update it", "same as before", "our plan") referring to prior state.
    2. Decision-making: When a previously established constraint or preference would materially change a current choice.

System Architecture Diagram

// Example of a tool interface declaration
interface ToolCall {
  toolName: string;
  parameters: Record<string, any>;
  sessionContext: string;
}

Reading and tool discovery are only a check; afte...