Style Match

Mirror the writing style and copywriting of any social account. Given a target account (X/Twitter or Instagram) — or a few example posts the user pastes — this skill pulls real posts, extracts a quantitative + qualitative "style card," and uses it to write new copy that reads like it came from that account, on the user's OWN topic, without copying its actual words.

When to use this skill

Load this skill when the user wants to match an existing account's voice, e.g.:

• "Write a post in the style of @account" / "make it sound like @account"
• "Match the voice/tone of this account"
• "Rewrite this so it reads like <brand/creator>"
• "Mirror the copywriting of @account"
• Draft captions / threads / tweets that fit a specific account's established voice

If the user just wants generally good copy with no reference account to imitate, you don't need this skill.

The workflow

1. Identify the target. Get the @handle (X) or username (Instagram) whose style to mirror. If the user pasted example posts directly, skip fetching and feed those texts into step 3 instead.

2. Pull real posts — never guess the style. Fetch the account's own recent posts so the analysis is grounded in real text, not assumption:

   • X: x_user_posts (handle) for the timeline; x_post for a single linked post.
   • Instagram: instagram_user_posts (username); instagram_post for one post. Grab 15–30 posts when you can — more samples = a sharper style card. Use the post text (X tweet text) or the caption (Instagram).

3. Compute the quantitative style card. Collect the post texts into a JSON array of strings and run the bundled script with run_skill:

   • skill_id: "style-match"
   • command: "python3"
   • args: ["scripts/style_stats.py", "posts.json"]
   • input_files: [{ "path": "posts.json", "content": "<JSON array of post texts>" }]

   The script prints metrics: post length, sentence rhythm, emoji / hashtag / mention rates, question / exclamation / ALL-CAPS usage, formatting (line breaks, lists), the most-used emojis, common opening hooks, and a vocabulary signature. These are the measurable bones of the voice.

4. Read the qualitative dimensions. Use read_skill_file to load:

   • references/style-dimensions.md — the full taxonomy of what makes a voice (diction, syntax, rhythm, POV, hooks, CTAs, formatting, attitude). Read the real posts through these lenses to catch what numbers can't.
   • references/mirroring-playbook.md — how to replicate a style faithfully WITHOUT plagiarizing, the style-card template, and a do/don't list.

5. Assemble the style card. Merge the script's numbers with your qualitative read into one compact style card (template in the playbook). This card is your spec for the draft.

6. Write to the card. Draft the user's copy so it hits the card's targets — same rhythm, same opener pattern, same emoji/hashtag density, same attitude — on the user's own subject. Match structure and voice, not content.

7. Self-check before delivering. Re-read your draft against the card: does the average sentence length match? The emoji density? The hook pattern? Fix any drift. Optionally re-run style_stats.py on your own draft and compare the numbers to the target's.

Other voice jobs

The workflow above mirrors a voice into new copy. Two related jobs use the same style card:

• Document the voice — turn the card into a voice guide someone else can follow: tone, vocabulary to use and avoid, sentence rhythm, openings and closings, formatting habits, and do/don't pairs. Every rule cites two or more real examples from the samples.
• Check a draft against a voice — the user's own guide or a card you built. Point to each line that drifts, say which rule it breaks, and give the fix; keep the draft's facts untouched. Raise conflicts between rules only when they matter.

Evidence rules

• A pattern seen once is a hint, not a rule. Separate what the account does often from its exceptions and from your suggestions.
• Measure before claiming: lengths, emoji and hashtag rates come from the script or code, with the counting method stated.
• A post that performed well does not, by itself, make its style a rule of the voice.
• Never infer biography, values or audience with more confidence than the samples allow.
• Keep dates and formats of the samples in view; a voice drifts over time.

Saving the guide — as a document or as a skill — happens only when the user asks for it.

Video scripts (reels, TikToks, brand integrations)

When the deliverable is a script rather than a post — reel, TikTok, story sequence, talking-head, or brand integration — build the style card as above, then read references/script-writing.md with read_skill_file. It adds the script-specific voice sources (the creator's comment replies, video transcripts, past integrations), a reusable voice-profile template, format structures, and the robotic-script edit pass. Deliver the finished script via create_artifact.

Guardrails

• Mirror, don't copy. Replicate the patterns (rhythm, structure, tone, hooks), never the account's specific phrasings, jokes, or signature lines. Output must be original copy on the user's topic.
• Ground in real posts. Don't invent what an account "probably sounds like." If you couldn't fetch posts and the user gave no samples, say so and ask for a handle or a few example posts rather than guessing.
• Respect intent. This is for legitimate voice-consistency and inspiration — not impersonation. Don't produce copy meant to deceive readers into thinking a real person authored it (fake endorsements, impersonation, etc.).
• One account at a time unless the user explicitly asks to blend voices.