Prompt Details
Model
(claude-4-6-sonnet)
Token size
1,239
Example input
[WRITING_SAMPLES]: Sample 1, internal Slack: ok so I looked at the bug report and it's not what I thought. The crash isn't in the parser, it's in the layout pass - the parser handles the malformed input fine but then we try to lay it out and there's a missing width attr that we should be falling back on but aren't. quick fix is adding the fallback in the layout pass, longer fix is making the parser emit a default attr when it sees this pattern. ship the quick fix today, file the longer fix as a follow-up? --- Sample 2, short blog post: A thing I keep noticing in code review: nobody writes the deletion PR they need to write. Every codebase has a few hundred lines of dead code that everyone knows is dead, but the PR to delete it never gets prioritized. I think this is because writing code feels productive and deleting code feels like cleanup. But cleanup compounds. --- Sample 3, email reply: Yeah I can take this one - give me until Thursday. The thing I'd want to check first is whether the metric is actually measuring what we think; the dashboard last quarter had a similar shape and turned out the join was wrong. If the metric checks out I can write up the recommendation by EOW.
Example output
# Voice Profile: writer (casual-technical)
## Quick summary
Writes long sprawling sentences punctuated with em-dashes, then drops a short fragment for emphasis ("But cleanup compounds."). Casual register even in technical contexts - opens with lowercase "ok" and "Yeah." Confident on technical analysis, hedged on inference ("I think this is because").
## Sentence structure
- Average ~22 words, high variance (4 to 40+ words)
- Fragments: yes, for emphasis at paragraph ends
- Long em-dash sentences: common
- Common openers: ok, Yeah, A thing, The crash, I think
## Vocabulary
- Recurring words/phrases: "the thing," "I think," "actually" - mild hedges before technical points
- Notably absent: leverage, robust, transform, the AI-tells
- Reading level: professional but conversational - precise technical terms (layout pass, fallback, join) inside casual structures
- Contractions: yes (it's, isn't, I'd, we're)
- Profanity: not in these samples
## Rhythm and punctuation
- Em-dashes: high density, used for clarification not decoration
- Commas: standard, for serial clauses
- Parentheticals: rare
- Semicolons: occasional
- Exclamation marks: none
## Structural habits
- Paragraph length: 3-5 sentences; one paragraph per sample here
- Openings: "ok so I looked at..." / "A thing I keep noticing..."
- Closings: a proposed next step, often as a question ("ship the quick fix today, file the longer fix as a follow-up?")
- Lists: not in these samples
## Voice qualities
- Register: casual professional
- Self-positioning: confident on facts, hedged on inference
- Personal anecdotes: implicit (describes specific situations), not extended
- Metaphors: light ("pulling its weight," "compounds")
- Humor: dry, mostly absent
- Distinctive ticks: lowercase opener on casual messages, em-dash for mid-sentence clarification, fragment ending for emphasis
## What this writer avoids
Exclamation marks, marketing vocab, formal email openers ("I hope this finds you well"), capitalized first word in casual messages, lists in short pieces
## How to use this profile
- Open with the specific point, not a context-setting sentence
- Use em-dashes for clarification, sparingly elsewhere
- Default to contractions
- Hedge inferences with "I think" or "I'd want to check"; don't hedge technical facts
- End with a question or a next action when proposing a path
## Confidence level
Sample size: 3 samples, ~280 words. Confidence: medium-low - samples are short and from similar internal/technical registers. To improve: add a 500+ word long-form piece and a public-facing or persuasive sample to confirm the patterns hold across registers.
By purchasing this prompt, you agree to our terms of service
CLAUDE-4-6-SONNET
Extracts a writer's voice profile from 3-5 samples of their actual writing - it analyzes rather than edits. It names what the samples actually show: sentence structure, vocabulary, rhythm and punctuation, structural habits, voice qualities, and what the writer avoids, with quotes, plus how-to-use notes you paste into any writing prompt to re-voice its output. It observes, never invents, and rates its own confidence. Returns a structured, reusable profile. Personal and commercial use permitted.
...more
Added 2 days ago
