Copilot Toolbox logo

Copilot Toolbox

Organization
pnp
copilot-toolbox

Creates, initializes, synchronizes, repairs, and publishes the portable English Copilot Toolbox, including its list, review workflow, Toolbox.aspx, and English HTML Toolbook. Blocks incomplete imports, enforces Copilot in SharePoint button actions, and resumes safely after failures. Use when the user says: - "Copilot Toolbox" - "install the toolbox and import all tools" - "synchronize the Agent Toolbox list" - "Synchronize the Copilot Toolbox and fully enrich every current tool." - "create the English Copilot Toolbook"

Overview

Publisherpnp
Repositorysharepoint-skills
Skill namecopilot-toolbox
Stars
141
Forks
60
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by pnp on GitHub. Read the source before you install it.

Installation

Install the Copilot Toolbox AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/pnp/sharepoint-skills/tree/main/Skills/copilot-toolbox/copilot-toolbox
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/pnp/sharepoint-skills.git /tmp/sharepoint-skills
mkdir -p .claude/skills
cp -r /tmp/sharepoint-skills/Skills/copilot-toolbox/copilot-toolbox .claude/skills/copilot-toolbox
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Copilot Toolbox in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Copilot Toolbox on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Copilot Toolbox is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Copilot Toolbox

When to use

Use this portable workflow to install, synchronize, repair, review, or publish the English copilot-toolbox solution on the current SharePoint site unless another site is named.

  • The generic list copilot-toolbox is the installation marker.
  • If absent, provision the complete solution and import the current catalog.
  • If present, never recreate existing artifacts; synchronize, repair, or publish as requested.
  • Never create, replace, edit, or designate Home.aspx as the site home page.
  • Never hard-code site URLs, list/view IDs, timestamps, page URLs, library paths, or tool counts.

Inputs

  • Target list: copilot-toolbox.
  • Authoritative identities and short descriptions: complete current tool catalog plus supplied --agenttools where present.
  • Mandatory enrichment source: complete learn_tool result for every current tool, including direct tools.
  • Identity key: exact Title, compared trimmed and case-insensitive.
  • Lifecycle: new, active, archived, removed.
  • Page: Toolbox.aspx in Site Pages.
  • Publication: copilot-toolbook-en.html in the current site's runtime-resolved standard document library (Documents or its localized equivalent, such as Dokumente).

Managed fields:

FieldRequired rule
TitleExact tool identity; display name Tool Name.
descriptionOriginal short catalog description unchanged.
CategoryMost specific verified functional family.
UseCaseLearned When to invoke, When not to invoke, Usage notes, in that order.
ExampleConcrete valid English request specific to the tool.
PromptTemplateCapability-specific reusable request with meaningful placeholders.
KeyParametersEvery exact learned input name and concise verified meaning.
Statusnew, active, archived, or removed.
TagsConservative English search terms.
Sourcetool-catalog, --agenttools, manual, or generated.
FirstSeenInitial verified detection; preserve existing populated values.
LastSeenCurrent verified detection.
RemovedOnSet only for removed tools.
workstepQuick Steps column for deterministic review.

Do not create Purpose, Prerequisites, Complexity, OutputType, or ReviewNotes.

Steps

1. Detect mode and read state

  1. Discover generic lists and resolve copilot-toolbox by normalized title.
  2. Do not use metadata itemCount as row truth.
  3. If the list exists, read every row with all managed internal fields; re-read at returned count if truncated.
  4. Preserve exact identities, accurate editorial content, FirstSeen, and active/archived lifecycle state unless verified current learning requires correction.

2. First-run provisioning

Run only when the list is absent.

  1. Create the English generic list with all managed fields and rename built-in Title to Tool Name. Seed no fixed records.
  2. Create view New Tools with query <Where><Eq><FieldRef Name="Status"/><Value Type="Choice">new</Value></Eq></Where>, fields LinkTitle, description, Category, Status, FirstSeen, workstep, and row limit 30.
  3. Fetch actual schema and create exactly one active Quick Step named Set tool active, condition [$Status] == 'new', action SetValue, Status = active, Choice overwrite, no prompt and no flow.
  4. Configure only verified workstep with that Quick Step using paired CornflowerBlue classes. Never attach it to Action0.
  5. Create exactly one full-width home-layout Toolbox.aspx through create_page_workspace, without changing the site home page.

Before page creation resolve actual list/view IDs and URLs and pass them in providedContent.

The pageSpec MUST include these exact binding and action rules:

The List web part MUST use the list whose title is exactly copilot-toolbox and the view whose title is exactly New Tools. It MUST NOT use Documents, Shared Documents, Site Pages, or any other library.

Each required Button web part MUST set its Action / Action type to Copilot in SharePoint (the localized UI label may also be Copilot in SharePoint). It MUST NOT use Link, URL, text link, Quick Links, Hero link, or any navigation action. Enter the exact required request in the Prompt field shown below the Copilot in SharePoint action selector.

The page must contain, in order:

  1. Image-free white-to-teal/green-blue full-width surface with strong contrast.
  2. Exact title Copilot in SharePoint Toolbox.
  3. Exact subtitle A Self-Updating SharePoint List of All Copilot in SharePoint Agent Tools.
  4. Short English explanation of create, synchronize, enrich, review, and publish.
  5. Exactly two real SharePoint Button web parts:
    • Update Toolbox: Action / Action type = Copilot in SharePoint, never Link; Prompt field = Synchronize the Copilot Toolbox and fully enrich every current tool.
    • Toolbook HTML: Action / Action type = Copilot in SharePoint, never Link; Prompt field = Create or refresh copilot-toolbook-en.html in this site's standard document library.
  6. List web part bound to actual copilot-toolbox and New Tools IDs.
  7. Built by Michael Greth — yourcopilot.de and Source: github.com/mysharepoint.

The page agent must not put the prompt into a URL field, button URL, description, tooltip, or visible body text as a substitute. The prompt belongs in the Prompt field beneath the Copilot in SharePoint action selection.

Read the page once after creation. Verify what extraction exposes. A button is valid only if all three facts are proven: it is a real Button web part; its action is Copilot in SharePoint and not Link; its Prompt field equals the required prompt. Also verify image-free design, title/subtitle, List binding, and unchanged Home.aspx. If extraction cannot prove action type or prompt storage, report those requirements as unverified and the page provisioning as partial; never claim the buttons work. If a button uses Link, report it as failed provisioning and provide the exact manual correction.

3. Learn every current tool

  1. Enumerate the complete authoritative tool set exactly, including direct-tool identities.
  2. Call learn_tool for every current tool in safe batches and retain every complete result.
  3. Map each tool independently:
    • Keep description unchanged from the short catalog.
    • Build UseCase only from learned guidance and preserve explicit exclusions, prerequisites, sequencing, limits, identifiers, and constraints.
    • Build KeyParameters from every exact learned input plus its verified meaning.
    • Choose the most specific family, such as automation.approvals.* → Approvals.
    • Write a capability-specific Example and PromptTemplate, not boilerplate.
    • Never invent a missing learned section; mark it unavailable and fail the gate below.

4. Mandatory enrichment transaction gate

Treat enrichment and synchronization as one logical transaction. Before any tool-record create or update:

  1. Retain one complete learn_tool result for every authoritative current tool.
  2. Build the complete final row for every current tool in memory.
  3. Finish identity comparison, lifecycle decisions, timestamps, choices, and the full write plan.
  4. Validate every current row against every blocking check.
  5. Write nothing unless all current tools pass.

Blocking checks:

  • UseCase has When to invoke, When not to invoke, and Usage notes in order.
  • UseCase retains tool-specific exclusions, prerequisites, sequencing, limits, and constraints.
  • KeyParameters contains every learned input and a non-empty verified meaning for each; name-only lists fail.
  • Category is the most specific verified family.
  • Example is a concrete valid request, not boilerplate.
  • PromptTemplate has capability-specific placeholders.
  • No managed field contains placeholder, fallback, or generic filler.

Forbidden patterns include Needs review, requires enrichment, exact meanings require, generic verified inputs, concrete task, specific outcome, specific target, copying the short description as the whole UseCase, name-only KeyParameters, and generic Examples or PromptTemplates reusable unchanged for unrelated tools.

This gate applies after timeout, throttling, parser errors, declined or expired confirmation, partial writes, and every fallback/recovery path. There is no emergency, simplified, provisional, placeholder, or seed import mode. If learning or validation is incomplete, stop before tool-record writes and report the failed gate.

5. Compare and lifecycle plan

  • Match exact trimmed lowercase identity; use only confirmed aliases.
  • New current tools → new.
  • Preserve archived.
  • Returning removed tools → new, clear RemovedOn.
  • Mark missing tools removed only when the complete authoritative set proves absence; never delete automatically.
  • Add required Category choices before writing affected rows.

6. Automatic repair and safe resume

At the start of every synchronization, scan all existing current-tool rows against the blocking checks. Treat a violating row as incomplete regardless of Status, including active.

After an ambiguous write result, timeout, expired confirmation, or interrupted execution:

  1. Fresh-read every managed field.
  2. Match by normalized exact Title.
  3. Classify every authoritative identity as verified, missing, incomplete, or duplicate.
  4. Preserve verified rows and human-reviewed lifecycle state.
  5. Create only missing rows.
  6. Update only incomplete rows from retained complete learned definitions.
  7. Never repeat a write whose intended state is already present.
  8. Do not delete duplicates automatically; report exact duplicate identities unless a verified safe correction is available.

When execution can continue safely, resume automatically from the fresh-read state. If the platform requires a new confirmation, stop before degraded writes; after confirmation, fresh-read and repeat the complete enrichment gate.

7. Write and verify

Write only after the transaction gate passes. Set LastSeen, preserve or initialize FirstSeen, set Source, clear RemovedOn for current tools, and fresh-read all rows. Verify exact identity coverage, duplicates, required fields, forbidden markers, lifecycle values, and parameter meanings. Sample every learned batch and compare stored UseCase and KeyParameters with retained learned definitions. After ambiguous timeouts, fresh-read before retrying.

8. Completion assertion

Report created-and-initialized or existing-list-updated as successful only when a fresh read proves exact authoritative identity coverage, zero duplicate current identities, zero missing required values, zero forbidden markers or generic filler, zero name-only KeyParameters, and stored samples from every learning batch matching retained definitions.

Otherwise report partial, name each failed check, list affected identities or a bounded sample plus count, and never describe the import as complete. Continue automatic repair when safely possible.

9. Publish English Toolbook

Create or refresh only copilot-toolbook-en.html in the current site's runtime-resolved standard document library. Discover the site's standard document library at runtime; its title is commonly Documents or a localized equivalent such as Dokumente. Never hard-code the library title or URL, and never save the Toolbook to AgentAssets, the skill folder, or any library other than the resolved standard document library. Save the file at the root of the resolved standard document library. Include new, active, and archived; exclude removed by default. Preserve multiline UseCase and exact KeyParameters, sort categories alphabetically, provide search, status filtering, collapsible categories and lifecycle badges, keep blue/orange design and attribution, use English only, verify file existence in the resolved standard document library, and return its browser link.

Output format

Report briefly:

  • Mode: created-and-initialized, existing-list-updated, toolbook-published, or partial.
  • Artifact status: list, schema, New Tools, Quick Step, workstep, Toolbox.aspx, image-free header, two Button web parts, each button's action type, each Prompt field, and verified List binding.
  • Current, created, updated, removed, reactivated, duplicate, and incomplete counts.
  • Learned sent, successful, incomplete, and failed counts.
  • Blocking-gate result and unresolved learned sections.
  • New, Removed, Reactivated, and Needs repair sets; for more than 20 show totals and about five examples.
  • Links to Toolbox.aspx, the list, and Toolbook when present.

If any tool or page operation fails or returns empty, say so plainly and never invent success.

Constraints

  • Never modify Home.aspx.
  • Never rerun first-run provisioning when the list exists.
  • Never seed fixed tool identities or counts.
  • Never create duplicate lists, views, Quick Steps, pages, or Toolbook files.
  • Never write tool records before complete mandatory learning and validation.
  • Never use generic filler or placeholder imports.
  • Never call name-only parameters complete.
  • Never overwrite accurate editorial fields without verified reason.
  • Never publish the Toolbook outside the runtime-resolved standard document library.
  • Never produce German Toolbox or Toolbook artifacts.
  • Never configure either Toolbox button with action Link; both MUST use Copilot in SharePoint and store the exact request in the Prompt field.
  • A confirmation timeout is not permission to degrade quality; after confirmation, resume from fresh state and rerun the gate.

Frequently asked questions

What does the Copilot Toolbox AI skill do?

Creates, initializes, synchronizes, repairs, and publishes the portable English Copilot Toolbox, including its list, review workflow, Toolbox.aspx, and English HTML Toolbook. Blocks incomplete imports, enforces Copilot in SharePoint button actions, and resumes safely after failures. Use when the user says: - "Copilot Toolbox" - "install the toolbox and import all tools" - "synchronize the Agent Toolbox list" - "Synchronize the Copilot Toolbox and fully enrich every current tool." - "create the English Copilot Toolbook"

Why use Copilot Toolbox on TypingMind?

Because you install it once and use it with any model. Copilot Toolbox is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Copilot Toolbox in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/pnp/sharepoint-skills/tree/main/Skills/copilot-toolbox/copilot-toolbox. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Copilot Toolbox?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Copilot Toolbox?

As many as you like. As long as a model supports skills, you can use Copilot Toolbox with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Copilot Toolbox AI skill free?

Yes. It is published on GitHub by pnp under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇