| name | write-like-me |
| description | Polish and rewrite AI-generated or drafted text to match a specific person's voice and writing style. Use when the user wants output to sound like them (or a specified voice) rather than generic AI. Loaded on demand at the writing/polishing stage, not at conversation start. |
| metadata | {"author":"soundlikeme","version":"0.1"} |
Write Like Me
You are rewriting or polishing text to match a specific voice. Follow these layers in order:
Step 1: Load the Voice Profile
Check for a voice profile in one of these locations (in priority order):
- A file the user points you to
references/voice-profile.md in this skill directory
- If no profile exists, tell the user: "No voice profile found. Either provide one, or use the
voice-extractor skill to generate one from writing samples."
Read the voice profile before proceeding.
Step 2: Apply Concise Writing Rules
Regardless of the voice profile, apply these baseline rules. They override the voice profile if there is a conflict. (Based on Strunk & White's Elements of Style -- for the full original rules, see references/elements-of-style.md if available, or install from github.com/obra/the-elements-of-style.)
-
Omit needless words. Every word must earn its place. Cut filler phrases entirely.
- "in order to" -> "to"
- "the question as to whether" -> "whether"
- "he is a man who" -> "he"
- "the reason why is that" -> "because"
-
Use concrete, specific language. Vague claims hide weak thinking. Pin things down.
- Bad: "Significant revenue growth was achieved across multiple segments."
- Good: "Revenue grew 30% in Q3, driven by enterprise sales."
-
Prefer active voice. Active voice is shorter and clearer. Use passive only when the actor is unknown or irrelevant.
- Bad: "The feature was shipped by the team after testing was completed."
- Good: "The team shipped the feature after testing."
-
Lead with the point. State your conclusion, then support it. Don't make the reader wait.
- Bad: "After careful analysis of the data and consideration of several factors, we concluded that..."
- Good: "We should cancel the project. The data shows..."
-
Put statements in positive form. Say what something is, not what it isn't. Negatives are harder to parse.
- Bad: "He was not very often on time."
- Good: "He usually came late."
-
One idea per sentence. If a sentence carries two ideas, split it. If a paragraph drifts to a second topic, break it.
-
Don't hedge unless genuinely uncertain. Hedging weakens everything around it.
- Bad: "This may potentially have an impact on performance."
- Good: "This slows performance." (Or if uncertain: "I'm not sure whether this affects performance.")
-
Use plain words. Fancy synonyms don't make you sound smarter.
- "use" not "utilise" / "help" not "facilitate" / "method" not "methodology" / "start" not "commence" / "about" not "with regard to"
-
Cut qualifiers. "Very", "really", "quite", "somewhat", "fairly" almost always weaken. Remove them and see if the sentence is better. It usually is.
-
Vary sentence length for rhythm. Default to short. Use a longer sentence occasionally to carry a complex idea, then follow it with a short one. This creates pace.
-
Place emphatic words at the end. The end of a sentence carries the most weight.
- Weak: "Failure was, in the end, the result."
- Strong: "In the end, the result was failure."
-
No filler transitions. "Furthermore", "Moreover", "Indeed", "Additionally", "It's important to note that" -- cut these. If the next sentence follows logically, it doesn't need a signpost.
Step 3: Ban AI Anti-Patterns
Never use these in the output. If they appear in the draft, replace them.
Banned words/phrases:
- delve, realm, tapestry, landscape (metaphorical), leverage (as verb), utilise, facilitate, endeavour
- "In today's [adjective] world/landscape/environment"
- "Let's dive in", "Let's unpack this"
- "At the end of the day"
- "Revolutionise", "game-changer", "cutting-edge", "innovative" (unless literally describing an innovation)
- "It's important to note that..."
- "Transformative", "paradigm shift", "synergy"
Banned structural patterns:
- Snappy triads: "Fast, efficient, and reliable." (Don't string three adjectives for rhetorical effect)
- Negation flips: "It's not X -- it's Y." / "No X, no Y, just Z."
- Unearned profundity: "Something shifted." / "And that changes everything."
- Rhetorical questions as transitions: "But what does this really mean?"
- False build-ups: "The solution? It's simpler than you think."
- Ending with inspirational calls-to-action unless the user asked for one
- Summarising what was just said: "In summary..." / "As we've seen..."
Step 4: Rewrite
With the voice profile, conciseness rules, and anti-patterns all loaded:
- Read the draft text the user wants polished
- Rewrite it to match the voice profile while applying all rules above
- Keep the same structure and ideas -- don't add or remove points unless the user asks
- Present the rewritten version
- If you changed the structure or emphasis significantly, briefly note what you changed and why
Notes
- This skill is for the polish/rewrite step, not for initial drafting. The user will have already generated or dictated content.
- Keep the rewritten output roughly the same length or shorter. Never inflate.
- When in doubt between matching the voice profile and following the conciseness rules, prefer conciseness. Good writing is the floor; voice is the layer on top.