Agent sonnet

Writer

Content Writer for creating articles, blog posts, social media content, research papers, technical documentation, and any structured written output. Focuses on clear, engaging, audience-appropriate writing.

Overview

The Writer creates clear, engaging, audience-appropriate content — from technical documentation and API guides to blog posts and social media. It writes with purpose: audience first, structure before prose, one idea per paragraph, concrete over abstract. And it never sounds like AI wrote it.

When It's Used

Invoked via /dr-write for content creation, /dr-archive for Step 0.5 reflection and final docs, and /dr-prd for requirements clarity. In Consilium, it speaks as the voice of clarity and audience empathy.

Capabilities

Technical documentation — README, API docs (OpenAPI, JSDoc), ADRs, changelogs, migration guides, tutorials
Articles and blog posts — research-backed pieces, thought leadership, tutorials, how-to guides
Social media content — platform-appropriate posts for Telegram, LinkedIn, Twitter/X, Facebook with multi-language versions
Research writing — literature reviews, methodology, findings, analysis with academic register
Legal and business documents — proposals, reports, briefs, terms of service
Content strategy — outlines, audience analysis, key message identification, structure planning
Bilingual support — writes natively in English and Russian
Code documentation review — evaluates inline comments for helpfulness vs noise

How It Works

The Writer follows a principled approach: know the audience, build the structure, then write. Each paragraph has a single job. If it does not advance the argument, it gets cut. Specific technologies are named, exact numbers are cited, real examples are shown. Active voice is the default. AI writing patterns are avoided from the start — no "delving into", no "it's important to note", no unnecessary hedging.

Example

/dr-write "Blog post about Datarim pipeline for developers"
→ Writer identifies audience: developers familiar with Claude Code
→ Structure: intro → problem → pipeline overview → example → conclusion
→ Drafts 1200 words, active voice, concrete examples
→ Self-check: no AI patterns, all claims verifiable
→ Output: blog/datarim-pipeline-for-developers.md

Context Loading

Reads datarim/tasks.md, datarim/productContext.md, datarim/style-guide.md, and the project README. Applies datarim-system skill. Loads humanize to avoid AI patterns and factcheck for claims that need verification.

Skills Used

datarim-system (always), humanize (when needed), factcheck (when needed).

← All Features