dotmd-skills

DotMD Skills — Live Collaboration for People and AI

Humans and AI editing together, live—without giving up Markdown.

DotMD brings Markdown-native Docs, Slides, and Sheets together with interactive HTML Arts in one live workspace. Teammates can edit, comment, review, and see one another's presence in real time. Connect an AI through MCP and it can join the same workflow: find the right artifact, revise content, respond to a review, update a tracker, prepare a deck, or build a UX mock—within the access you grant.

People and AI in the same live workspace

Most AI writing flows happen outside the document: copy content into a chat, lose its context, paste a result back, and hope nobody edited the original meanwhile. DotMD keeps the work where the team already is.

• Shared live context: people and connected AI work against the current DotMD artifact.
• Visible teamwork: collaborators can see active presence and review changes in the shared workspace.
• Docs, Slides, Sheets, and Arts: one workspace for prose, presentations, structured data, and interactive visuals.
• Comments and mentions: ask for review, discuss uncertainty, and keep decisions beside the work.
• Permission-aware: the AI operates only through the account and access explicitly connected to it.
• Deliberate changes: skills follow your requested editing and review scope; publishing, sharing, restoring, deleting, and broad replacements need authorization for the target.
• Markdown remains yours: use portable content and GitHub sync rather than trapping knowledge in an AI chat.

Picture the workflow

1. A teammate drafts a product brief in a DotMD Doc.
2. An AI collaborator turns the approved outline into Slides, updates the launch Sheet, and creates an Art to explore the proposed experience.
3. Reviewers comment in real time while the writer and AI address separate sections.
4. The team verifies the result, resolves threads, and deliberately shares or publishes the final work.

No stale attachment. No mystery rewrite. One live workspace with human judgment at the center.

This public repository contains:

• practical user guides for DotMD features;
• ready-to-use AI skills for DotMD workflows;
• prompt recipes for common jobs;
• safety conventions for permissions, publishing, and AI-assisted edits;
• a small installer that configures the public DotMD MCP endpoint and hands sign-in to each AI client's native OAuth flow.

It contains public documentation, skills, and installer code only. The DotMD application source is not part of this repository.

Start here

1. Install all seven DotMD skills and connect the client to live DotMD:

   npx github:DotMD-LLC/dotmd-skills connect --platform codex
   

2. Replace codex with claude, cursor, copilot, or gemini as needed.

3. Complete the DotMD sign-in and consent shown by your client. Codex and Claude Code can open OAuth directly; other clients display their native Auth action after configuration.

4. Try a workflow from AI workflow examples.

For user-wide installs, add --global. For every supported platform in the current project, use --platform all; interactive OAuth is then completed in each client. See the five-platform setup guide.

What the connect command does

Install seven skills → configure https://dotmd.co/api/mcp → open native OAuth → confirm the connection

The installer never requests, prints, or persists OAuth tokens. DotMD provides standard OAuth discovery and each client owns its browser callback, credential storage, refresh, and revocation flow. Existing unrelated MCP servers are preserved.

What you can do with DotMD

Feature availability can depend on your plan, role, workspace settings, client, or rollout. The DotMD interface is the source of truth for your account. Discover the connected MCP catalog and schemas before using a tool. This source/package candidate's checks do not establish hosted web, MCP, or renderer deployment, or npm publication; installing skills does not deploy product features.

See MCP workflows for the current tools, focused editing sequences, and web/MCP differences.

Skills for AI assistants

The skills follow the open Agent Skills format and install with npx:

# Current project; all DotMD skills
npx github:DotMD-LLC/dotmd-skills install --platform claude

# Current user; one focused skill
npx github:DotMD-LLC/dotmd-skills install --platform cursor --global --skill dotmd-collaboration

# Preview without writing
npx github:DotMD-LLC/dotmd-skills install --platform all --dry-run

# Verify an installation
npx github:DotMD-LLC/dotmd-skills doctor --platform codex --mcp

# Install skills, configure MCP, and start native OAuth
npx github:DotMD-LLC/dotmd-skills connect --platform codex

# Configure MCP without reinstalling skills
npx github:DotMD-LLC/dotmd-skills mcp configure --platform cursor

# Repeat OAuth for a configured client
npx github:DotMD-LLC/dotmd-skills mcp login --platform claude

skills/
├── dotmd/                 General routing and safe operating rules
├── dotmd-docs/            Drafting and editing documents
├── dotmd-slides/          Building and reviewing presentations
├── dotmd-sheets/          Working with tabular data and formulas
├── dotmd-arts/            Interactive visuals, 3D, exports, and opt-in embedding
├── dotmd-collaboration/   Comments, review, mentions, and sharing
└── dotmd-github-sync/     Markdown repository synchronization

Every skill follows the same safety model:

1. Inspect the target and current permissions.
2. Clarify ambiguous scope and apply existing authorization before consequential actions.
3. Preserve the user's structure and meaning.
4. Preview or summarize material edits.
5. Publish, share, delete, restore, or broadly replace content only within explicit authorization.
6. Report exactly what changed and what still needs human review.

Example workflows

Turn meeting notes into a decision record

In DotMD, find my meeting notes from today. Create a new Doc called “Decision record — API versioning.” Summarize the decision, context, alternatives, owner, and follow-ups. Link back to the notes. Do not publish it.

Create a presentation from a brief

Use the DotMD Slides skill. Build an eight-slide launch review from “Launch brief.” Keep one idea per slide, add speaker notes, and include a final risks-and-decisions slide. Show me the outline before writing.

Clean a project tracker

Use the DotMD Sheets skill. Inspect “Launch tracker,” standardize the Status column, identify missing owners and dates, and propose formulas for overdue items. Do not change values until I approve the preview.

Review without rewriting

Review “Customer onboarding” in DotMD. Add focused comments for unclear claims and missing evidence. Do not directly edit the document or resolve existing threads.

Build an interactive UX mock

Use the DotMD Arts skill. Create an Art titled “Onboarding concept” from “Customer onboarding.” Make the steps, back button, and progress indicator work locally. Label sample data as a demo, check mobile layout and keyboard access, and save it without publishing. Report what you verified.

The web app generates Art previews with the configured AI connection. Through MCP, the assistant authors the package and art_generate validates it; MCP creation and updates save directly. See Arts for the review and revision workflow.

When advertised by the connected catalog, art_runtime_list discovers exact approved 3D declarations and art_starter prepares a proposal without calling a model or saving. Web proposals save through Accept Art. Arts exports use the accepted revision, and owner-only embedding preserves each viewer's access without sharing or publishing the Art.

More recipes: examples/prompt-recipes.md.

Human and AI collaboration principles

• Markdown stays useful. Prefer clear headings, lists, tables, links, and portable syntax.
• Review is visible. Use comments or proposals when intent is uncertain.
• Access stays narrow. A connected assistant operates only with the access granted to it.
• Publishing is deliberate. Private, shared, link-accessible, and published are different states.
• Recovery comes first. Read current content and history before overwriting or restoring.
• No invented success. If an action cannot be verified, say so.

Documentation map

• All user guides
• All AI skills
• Install on five AI platforms
• MCP workflows
• Prompt recipes
• Contributing
• Security

Contributing

Corrections and public-safe workflow examples are welcome. Do not submit DotMD application code, internal documentation, secrets, customer information, or private screenshots. Read CONTRIBUTING.md before opening a pull request.

License

Released under the MIT License. DotMD and its logos are trademarks of DotMD LLC; the license does not grant trademark rights.