Anthropic★ 4.3

Doc Co-authoring

Anthropic's official Skill for collaborative document authoring with humans in the loop. Tracked suggestions, structure-aware edits.

View repo Homepage

A+

Safety & Trust

Delv Safety Grade: A+

Score 94/100 · assessed 2026-04-18

Maintainer100

Permissions95

Supply chain90

Transparency90

Incidents100

Doc Co-authoring is Anthropic's official Skill for collaborative document editing with tracked changes and structure-aware modifications. As a first-party Anthropic product, it benefits from the highest maintainer trust and direct alignment with Claude's safety model. The Skill operates on document content passed to it rather than requiring filesystem or network access, making its permissions footprint minimal. It's distributed through Anthropic's official Skills repository with clear documentation and version control. The tracked-changes approach means edits are proposed rather than automatically applied, keeping humans in control. No security incidents are known. The main consideration is that as a newer Skill category, the supply chain maturity is still developing compared to established package ecosystems, though Anthropic's stewardship mitigates this substantially.

Green flags

Official Anthropic product, first-party trust
Human-in-loop design with tracked changes, not auto-apply
Structure-aware editing reduces unintended document corruption
No filesystem, network, or shell access required
Open source with clear documentation in official repo

Red flags

Skills ecosystem is newer, less battle-tested than MCP servers
No independent package registry yet, relies on GitHub distribution
Limited third-party security audits of Skills framework

Permissions requested

Read filesWrite files

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Doc Co-authoring is Anthropic's official Skill for structured, iterative document editing. It teaches Claude to propose inline edits with tracked changes rather than rewriting entire sections, and to maintain consistency across multi-section documents. The Skill understands document hierarchy, so it can expand outlines into prose, tighten arguments, or merge sections without losing context. It keeps a log of suggestions, which means you can reject edits without Claude forgetting what you've discussed. Particularly useful for long-form reports, white papers, or technical documentation where you're iterating over many rounds. Works best in plain text or Markdown; rich-text workflows require manual translation of suggestions. Not a style guide, so you'll need to provide examples if you want Claude to match a specific voice.

Review

I've used Claude for document work before, but without this Skill it tends to rewrite entire sections when you ask for a tweak, or lose track of which suggestions you've accepted. Doc Co-authoring changes that. It treats your document as a living structure with tracked changes, so when I ask Claude to tighten a paragraph or expand an argument, it proposes edits inline rather than dumping a wall of replacement text. The Skill understands outlines, so if I'm drafting a technical report with nested sections, it can expand bullet points into prose while keeping the hierarchy intact. It also maintains a log of suggestions, which means I can reject an edit without Claude forgetting the context or re-proposing the same thing five minutes later. In practice, this makes multi-round editing feel like working with a human co-author who remembers what you've already discussed. The real win is for long documents where you're iterating across many sections. I used it to draft a 20-page strategy doc, and Claude kept voice consistent across sections I wrote weeks apart. It also handles structural changes well, like splitting a section or merging two arguments, without mangling the surrounding text. The rough edges: it's not magic at understanding your personal style unless you give it examples upfront, and it can still over-explain in early drafts if you don't rein it in. Also, the tracked-change format works best in plain text or Markdown; if you're working in a rich-text editor, you'll need to translate the suggestions manually. But for anyone drafting reports, white papers, or documentation where multiple rounds of edits are inevitable, this Skill makes Claude far more useful than the base model's tendency to rewrite everything from scratch.

Verdict

Load this if you're drafting long documents with multiple revision rounds and need Claude to track what's been changed. Overkill for quick blog posts or single-draft work where you'd just paste the output and move on.

Good at

Tracked changes instead of full rewrites, so you can accept or reject edits cleanly
Maintains document structure and hierarchy across nested sections
Logs previous suggestions, avoiding repeated proposals after rejection
Keeps voice consistent across sections written at different times
Handles structural edits like splitting or merging sections without mangling text

Watch out

Requires upfront style examples to match your personal voice reliably
Tracked-change format works best in plain text or Markdown, not rich-text editors
Can still over-explain in early drafts unless you guide tone explicitly
Overkill for single-draft or short-form writing tasks
Not a replacement for human judgement on argument structure or logic

Use cases

Long-form report drafting alongside a human
Structured edits across many sections
Maintaining voice across multiple authors
Outline-driven document expansion