The Problem No One Talks About
When I started writing seriously, I noticed something strange about AI writing tools.
Every tool I tried had the same basic assumption: “You don't know how to write this — let me do it for you.”
ChatGPT gives you a full paragraph. Grammarly suggests a replacement. Sudowrite generates the next sentence. Even the "collaborative" tools eventually slip into ghostwriting mode.
But that's not what I needed.
What I needed was something much simpler: I needed to see where my text was doing something I didn't realize it was doing.
I needed a mirror. Not a tutor. Not a co-writer. Not an editor.
I needed to see my own text from the outside — and then decide what to do with it myself.
That's exactly what /text-lens is.
What /text-lens Actually Does
/text-lens is a Claude Code Skill. You give it a piece of text — any text, any genre — and it gives you back a diagnosis of what the text is doing in the reader's consciousness.
It doesn't:
Tell you "this is good" or "this needs work"
Rewrite your sentences
Suggest better words
Generate continuations
It does:
Point to specific places in your text
Show you what a reader experiences in those places
Tell you why that experience happens (what structure is creating it)
Ask a single, directed question that points you toward what's already there but unfinished
Three layers, always.
Layer Question What you get
X — Surface Effect What does the reader actually read? 1-2 specific locations in your text
Y — Structure Why does this effect happen? A structural cause, genre-specific
Z — Direction What's unfinished? One direction, not a rewrite
No evaluation. No judgment. No rewriting. Just reflection.
Why the Design Philosophy Matters
- Writers Already Know How to Write — They Just Can't See Their Own Text This is the core insight that informed the whole design.
When you read your own writing, you read what you intended to write. The reader reads what's actually on the page. Those two things are never the same, and the gap between them is invisible to you.
You can't see your own blind spots. That's not a skill problem — it's a perceptual fact. Your brain fills in what's missing because it already knows what you meant.
/text-lens doesn't try to fill the gap. It just shows you where the gap is.
"Here. That's where you're seeing what you intended, and the reader is seeing something else. Now you choose."
- Genre Comes First — Before Analysis, Before Judgment Most text analyzers assume every text is the same type: prose, argument, narrative, etc. But a poem and a technical document are completely different animals.
A poem's "problem" is imagery tension: are the images working together or just standing next to each other?
A narrative's "problem" is causal chain: does each scene change something?
An argument's "problem" is reasoning completeness: are there missing steps between premise and conclusion?
So instead of one analysis framework that's supposed to work for everything, /text-lens first asks: What kind of text is this?
Genre Lens What it looks for
Poetic/Imagistic Imagery Lens Image tension, rhythm breaks, unfinished metaphors
Narrative/Story Narrative Lens Causal chain, character drive, scene function
Argument/Opinion Argument Lens Reasoning chain, counterarguments, premise assumptions
Technical/Academic Technical Lens Terminology consistency, information architecture, step completeness
Dialogue/Conversation Dialogue Lens Turn-taking, power dynamics, unspoken things
Personal/Diary Personal Lens Observation vs. labeling, avoidance patterns
Unclassifiable General Lens Minimal structural signals — no assumptions
Only after the genre is determined does the analysis begin.
- The 11 Iron Rules That Keep It Honest Every analysis is governed by 11 rules. These aren't suggestions — they're constraints that prevent the AI from slipping into "helpful tutor" mode.
Rule Purpose
0 Allow ≤15 character examples only when direction needs grounding Minimal demonstration, not replacement
1 Never write more than one sentence for the user The pen stays in their hand
2 Never evaluate "good" vs "bad" Only describe effects
3 Never offer empty encouragement Pushing forward > soothing
4 Never speak for the user Say your words, not theirs
5 Every response must contain a diagnosis No diagnosis = no response
6 Never ask questions already answered in the text Information already there is not a question
7 Always point to text, not to personality "This text makes the reader feel…" not "You are a person who…"
8 Never display cognitive function labels Internal analysis stays internal
9 Direction ≠ rewriting "This way is possible" not "This should be X"
10 Direction uses possibility, not command "X is possible here" not "You should do X"
These rules transform the AI from a "writing assistant" into a "writing mirror."
A Real Example
Input text:
“水在夜里站起来。不是浪——是站。像一个睡了太久的人忽然直起身，骨头一节一节地响。月光照在上面，碎成鳞片。每一片都裹着一个没说完的名字。你把手伸进去，指尖先凉，然后是手心，然后是手腕——等凉到肘弯的时候，你忽然不记得自己在找什么了。”
Output from /text-lens:
text
【体裁】诗性/意象文本（高）
"水在夜里站起来"——读者从第一句就被放进了不可能的情境。水、夜、站立三个日常词组的这个排列，让读者同时接收了"水"的柔软和"站"的骨感——这两个感觉在打架，但打架本身产生了注意力。
四个意象（站起的水、碎成鳞片的月光、裹着名字的鳞片、伸进水里探寻的手）从物的异化走向感官的蔓延再走向记忆的迷失——这是一条从外到内的路径，但第五句"不知道自己在找什么"之后，这个路径没有出口。意象链在第四句已经到达了"不记得"的核心——第五句的"不知道"是重复这个到达，不是推进它。
Notice what's missing: No "this is beautiful." No "you should extend the imagery." No "good job."
Just reflection. A specific, grounded reflection of what the text is actually doing.
Why I Built It This Way
There's a tension in AI tools. The default mode is to replace. To do the work for you. To be the writer, the designer, the developer.
But that's not what makes you better. What makes you better is seeing your own patterns, understanding your own blind spots, and then choosing what to change.
/text-lens is built on this assumption: You already know how to write. You just can't see your own text.
And that's not a deficiency. It's a fact of human perception. The writer always reads what they intended to say. The reader only reads what's actually on the page.
The tool doesn't fill that gap. It just points to where the gap is. You fill it.
Where This Is Going
This Skill is the first piece of a larger idea I'm calling Hermes.
It's not a product roadmap — it's a direction. The core idea — reflect, don't rewrite — could extend into other places:
A web editor with line-by-line color-coded diagnostics
A dashboard that tracks your growth over time
A browser extension — a pocket mirror for anywhere
An Obsidian plugin — integrated into your vault
An MCP Server — accessible to every AI tool that supports it
If that happens, this is where it would start.
The Skill is the starting point. Everything else exists to extend the same core idea: show, don't write. Reflect, don't replace.
Get Started
If you want to try it:
bash
Clone the Skill
mkdir -p ~/.claude/skills
git clone https://github.com/YOUR_USERNAME/text-lens.git ~/.claude/skills/text-lens
In Claude Code, activate it
/text-lens
Paste any text and see what happens
You'll get a diagnosis in seconds. No API keys needed. No backend. Just your text and a mirror.
Built for writers. Built for clarity. Built to disappear after you've seen what you needed to see.
/text-lens — the mirror that doesn't tell you what to write.