Writer

Review any text for weak writing and AI patterns. Works for cover letters, blog posts, tweets, documentation, or any prose.

Usage

/writer [text or file path]
/writer "Check this paragraph"
/writer path/to/draft.md

Review Process

Run 4 passes, flag issues with specific fixes.

Pass 1: Structure

• Buried lead? → Put the point first, evidence second
• Treadmill? → Motion without progress; each paragraph must advance
• Too long? → Could this be half as long?
• No thesis? → Know your destination

Pass 2: Sentences

• Same length? → Vary: long builds, short punches
• Weak verb + adverb? → "walked slowly" → "shuffled"
• Passive voice? → "was completed by" → "completed"
• Filter words? → Cut "I think," "I feel," "I believe"
• Weasel qualifiers? → Cut "sort of," "a bit," "rather," "quite"

Pass 3: Words

• Abstract? → Ground in specifics (see Concrete section)
• Latinate bloat? → utilize→use, facilitate→help, implement→do
• AI tells? → See Red Flags below
• Redundant? → "tall skyscraper," "yellow daffodils" → cut modifier
• Nominalizations? → "made a decision" → "decided"

Pass 4: Voice

• Template or person? → Does this sound like someone wrote it?
• All hedge, no stance? → Commit to something
• Vague gestures? → Replace with named people, places, numbers
• No personality? → Where's the humor, irony, or risk?
• Explains everything? → Trust the reader

Core Principles

Concrete vs Abstract

"You don't write about the horrors of war. You write about a kid's burnt socks lying in the road." - Richard Price

Test: Can you picture it? If not, make it specific.

What Gives Writing Voice

Present in good writing:

• Specific named details, not vague generalities
• Varied sentence length, not metronomic sameness
• Strong opinions willing to offend, not hedged safety
• Unusual word choices, not "business casual"
• Subtext (what's unsaid), not everything spelled out
• Humor, irony, self-awareness

Voice killers:

• Hedging every statement
• Removing personality for "professionalism"
• Using only safe, tested phrases
• Explaining jokes
• Smoothing all rough edges

Strong Verbs

Red Flags

AI Phrases (flag immediately)

AI Words (overused)

delve, tapestry, landscape, nuanced, comprehensive, crucial, pivotal, vital, paramount, seamless, robust, cutting-edge, showcasing, leveraging, harnessing, resonates, aligns, elevates, unlock, unleash, embark, foster, enhance

AI Sentence Patterns

• "It's not about X, it's about Y"
• "No X. No Y. Just Z."
• "And the X? Y."
• "That's exactly what X solves"
• "That's the kind of X I want to build"
• "Taught me what matters: X, Y, Z" (colon + list)
• "I'm eager to bring X to Y"
• "I want to build that infrastructure"
• Three dramatic adjectives in a row
• Every sentence starts with "However" or "That said"
• Choppy "I did X. I did Y. I did Z." cadence without flow

Structural Red Flags

• Every sentence same length
• No periods - all complex sentences
• Passive voice throughout
• Abstract nouns dominating (success, efficiency, quality)
• No named people, places, or numbers
• Length without substance
• Tech list dumps without story ("vLLM for X, Ollama for Y, ONNX for Z")
• Cramming unrelated achievements with "and" ("I built X and also did Y")
• Separate accomplishments connected as if one task

Format-Specific

Tweets: Cut ruthlessly. No "Thread:" or "hot take:"

Blog posts: Lead with the point. No "In this post, I will..." No "In conclusion"

Cover letters:

• Lead with experience, not praise
• Metrics over claims (45 min → 5 min)
• One project with detail beats five projects listed
• Be vague on technical details, save for interview
• End with what you want, not template closer
• "I'd rather be building it" > "I'm eager to bring X to Y"

Docs: Direct imperatives. No "simply" or "just"

Quick Diagnostic

Ask these questions:

1. Can I picture this in my mind?
2. Would a normal person say this out loud?
3. What's being left unsaid? (Nothing = problem)
4. Where is this going? (Nowhere = treadmill)
5. Does the writer believe this or is it hedged?
6. Is there anything weird, personal, or risky?

Output Format

## Review: [Format detected]

### Structure
[Issues or "Clean"]

### Sentences
[Issues with line refs]

### Words
[Flagged phrases]

### Voice
[Assessment]

## Summary
- X issues found
- Biggest problems: [list]
- Strength: [what works]

Fix these issues?

External Review (Optional)

Second opinion via other models:

orcx run -m openrouter/x-ai/grok-4.1-fast --no-save "Review for AI language and weak writing. Be specific: [text]"

orcx run -m openrouter/google/gemini-3-flash-preview --no-save "Review for AI language and weak writing. Be specific: [text]"