Doc Co-Authoring: Structured Collaborative Document Creation

The doc co-authoring skill from Anthropic provides AI assistants with structured workflows for guiding users through collaborative document creation. Rather than unstructured back-and-forth editing, this skill establishes a three-stage process—context gathering, refinement through iteration, and reader testing—that ensures documents work effectively when others read them.

What This Skill Does

This skill guides users through professional document creation via active workflow management across three distinct stages. Context gathering closes knowledge gaps between users and AI, enabling smart guidance later. Refinement builds documents section-by-section through brainstorming options, curating selections, and iterative editing. Reader testing validates documents with fresh AI instances having no context bleed, catching blind spots before human readers encounter them.

The workflow applies to various document types: technical specifications, decision documents, proposals, PRDs, design docs, or RFCs. Each follows the same structure but adapts to type-specific templates and audience requirements. Memos for executives emphasize brevity; technical specs provide detail; proposals focus on persuasion—the skill adjusts appropriately.

What makes this approach effective is the emphasis on testing. Documents written collaboratively between users and AI often make sense to participants but confuse fresh readers. By testing with a fresh AI instance (no shared context), the skill identifies assumptions, ambiguities, or missing explanations before documents circulate to stakeholders.

Getting Started

The skill triggers when users mention writing documentation, creating proposals, drafting specs, or similar structured content tasks. Upon triggering, it offers the three-stage workflow with explanations of each stage, then asks if users want structured guidance or prefer freeform writing.

When users accept the workflow, the skill begins with meta-context questions: document type, primary audience, desired impact when readers engage with it, template or format requirements, and any constraints. This establishes the framework before content creation begins.

Integration with messaging platforms, document storage, and collaboration tools (via MCP servers or connectors) enables direct context pulling. If Slack, Teams, Google Drive, or SharePoint integrations are available, the skill can read team channels, shared documents, or related threads directly rather than requiring manual copying.

Key Features

Three-Stage Structured Process: Context gathering ensures sufficient background for smart guidance. Refinement builds documents iteratively rather than all at once. Reader testing validates with fresh AI that hasn't seen the conversation, revealing blind spots.

Section-by-Section Construction: Rather than drafting entire documents then editing, the skill builds one section at a time. For each section: ask clarifying questions, brainstorm 5-20 options for what to include, curate selections based on user feedback, check for gaps, draft the section, then refine through surgical edits. This incremental approach maintains focus and quality.

Integration-Aware Context Gathering: When team discussions happen in Slack channels or key context lives in SharePoint documents, the skill leverages integrations to read directly. This reduces manual copying and ensures comprehensive context capture.

Brainstorming with Curation: Instead of writing directly from sparse inputs, the skill generates multiple options (5-20 potential points per section), then asks users which to keep, remove, or combine. This surfaces forgotten context and provides choices rather than imposing structure.

Reader Testing Validation: Fresh AI instances test whether documents work for readers without background context. They answer likely discovery questions, identify ambiguities, check for false assumptions, and flag contradictions. Issues revealed trigger targeted refinement before stakeholder review.

Usage Examples

When creating a technical decision document, the skill first gathers context about the problem being solved, technical constraints, alternative solutions considered, organizational politics, and stakeholder concerns. During refinement, it brainstorms considerations for each section—trade-offs, implementation risks, migration paths—and lets users curate what matters. Reader testing confirms whether someone unfamiliar with the project can understand the decision rationale without asking clarifying questions.

For proposals seeking executive approval, context gathering emphasizes desired outcomes, budget constraints, success metrics, and stakeholder objections. Refinement focuses on executive summary first (where unknowns typically concentrate), then supporting sections. Reader testing validates that busy executives can grasp value proposition and required action without deep reading.

When drafting technical specifications, the skill asks about system architecture, API contracts, data models, error handling approaches, and performance requirements. Brainstorming generates implementation options; curation selects appropriate approaches. Reader testing confirms engineers unfamiliar with the project can implement correctly based solely on the spec.

Best Practices

Dump all context during Stage 1 without worrying about organization. The skill explicitly encourages stream-of-consciousness information dumps, links to channels, or pointers to related documents. Providing comprehensive context upfront enables better brainstorming and curation later.

Answer clarifying questions concisely using shorthand. When the skill asks numbered questions, respond with numbered answers—no need for complete sentences if the point is clear. Efficiency matters during iteration.

Be specific when refining sections. Instead of "make this better," indicate exactly what to change: "Remove bullet 3—already covered in section 2" or "Make paragraph 4 more concise." Specific feedback helps the skill learn style preferences for subsequent sections.

Don't edit documents directly during early stages. Let the skill make changes based on feedback so it learns preferences. If you edit directly and request reads, the skill notes changes mentally but learns less about why those changes mattered.

Take reader testing seriously. When fresh AI struggles with questions or identifies ambiguities, those are real issues future readers will encounter. Fix them now rather than discovering problems after stakeholders have already read the document.

When to Use This Skill

Use this skill when creating substantial documents where stakeholder understanding matters. Decision documents, technical specifications, project proposals, architecture designs—anything requiring clear communication beyond the immediate team benefits from the structured workflow.

The skill is particularly valuable when collaborating with AI on complex topics. The three-stage process prevents common pitfalls where documents make sense to authors (who share context) but confuse readers (who don't).

It's ideal when templates or organizational standards exist. The skill asks about format requirements early, then ensures documents conform to expected structure while maintaining quality content.

When NOT to Use This Skill

Don't use this skill for quick notes, drafts, or informal communications. The three-stage workflow takes time—appropriate for important documents, excessive for casual writing.

Avoid using it when you need to write solo without iteration. If you're drafting alone and won't involve AI in revisions, freeform writing may be more efficient than guided workflow.

It's not appropriate for real-time collaboration in meetings. The workflow assumes async document development, not live editing during calls.

Don't expect the skill to make strategic decisions. It guides process and tests comprehension, but doesn't determine whether proposals should be approved, designs are optimal, or specifications are technically sound.

Related Skills

This skill complements doc for DOCX document creation, docx for document editing and formatting, and notion-research-documentation for research documentation workflows.

Source

This skill is maintained by Anthropic. View on GitHub