2.13.0-beta.20). Features may change before the next stable release.Switch to stable →Skills
Write conversion copy with ak:copywriting
Turn an audience, offer, proof, and format brief into reviewable copy, variants, and evidence without inventing claims or hiding external processing.
Use ak:copywriting to turn a concrete audience, offer, channel, and brand-voice
brief into persuasive copy. The Skill supplies formulas, headline and CTA
patterns, channel structures, a copy-brief template, and a helper for extracting
writing-style signals from project assets.
Choose ak:copywriting for persuasive copy
Use ak:copywriting when
- You need headlines, subject lines, email sequences, landing-page sections, product descriptions, social posts, CTAs, or A/B variants.
- You have an audience, offer, proof points, desired action, and constraints to turn into a reviewable draft.
- You want to analyze an approved writing sample and apply its observable style dimensions without copying its substance.
Choose another workflow when
- You need factual product or API documentation. Use
ak:docsand treat persuasion as a secondary edit. - You need a full interface implemented around the copy. Use
ak:frontend-designorak:frontend-developmentafter the message is clear. - You need market evidence, legal review, brand approval, or current platform-policy research. Obtain that evidence separately; this Skill's templates do not establish it.
- The source contains personal, confidential, licensed, or unpublished material that is not approved for model processing.
Prepare the brief and source material
Before you start:
- Complete Onboarding, and confirm Engineer Kit is installed for the current runtime and scope.
- Provide the audience, awareness level, pain points, goal, offer, primary CTA, format, length, tone, must-include items, must-avoid items, and success metric.
- Supply exact, supportable proof: product behavior, prices, dates, guarantees, customer quotes, study results, and legal qualifiers. Mark anything that is a placeholder rather than a verified claim.
- Identify the approved brand or writing samples and whether the runtime may read them or send them to an external processor.
- State the output location if you want files written. Otherwise, request copy in the response and keep the project unchanged.
| Runtime | Invocation | Availability boundary |
|---|---|---|
| Claude Code | /ak:copywriting ... | Native delivery can use files and commands available to the session; external style extraction still needs its own dependency, network, credential, and data approval. |
| Cursor | /ak:copywriting ... | Slash invocation is user-verified; file access, command execution, and approval behavior depend on the Cursor session. |
| Codex | $ak:copywriting ... | Native Skill discovery is supported; identical access to project assets, external processors, or runtime prompts is not implied. |
The declared argument hint is [copy-type] [context], but the Skill does not
define a parser, fixed copy-type list, or default output file. State the desired
deliverable and boundaries in natural language.
Run the Skill with a complete brief
/ak:copywriting "Draft a landing-page hero and three headline variants for finance leads who already know the problem. Use only the supplied product facts and approved customer quote, keep the tone direct and professional, use one CTA to book a demo, and flag every missing proof point instead of inventing it. Return copy in chat; do not write files."/ak:copywriting "Draft a landing-page hero and three headline variants for finance leads who already know the problem. Use only the supplied product facts and approved customer quote, keep the tone direct and professional, use one CTA to book a demo, and flag every missing proof point instead of inventing it. Return copy in chat; do not write files."$ak:copywriting "Draft a landing-page hero and three headline variants for finance leads who already know the problem. Use only the supplied product facts and approved customer quote, keep the tone direct and professional, use one CTA to book a demo, and flag every missing proof point instead of inventing it. Return copy in chat; do not write files."The bundled brief organizes project information, audience and awareness, deliverable details, key messages, proof, constraints, formula, and success metrics. You can provide the same information directly or ask the Skill to make missing decisions visible before drafting.
Select patterns without treating them as proof
| Need | Bundled guidance | Review question |
|---|---|---|
| Persuasive flow | AIDA, PAS, BAB, 4Ps, 4Us, FAB, ACCA, 1-2-3-4, and Star-Chain-Hook | Does the structure fit the reader's awareness and the available proof? |
| Headline or subject line | How-to, number, question, revelation, negative, command, comparison, guarantee, curiosity, benefit, and urgency patterns | Is the promise specific, accurate, and supportable? |
| Landing page | Hero, social proof, benefits, process, FAQ, offer, and repeated primary CTA | Does every section advance one outcome without fake scarcity or unsupported social proof? |
| Short, standard, story, sequence, preview-text, CTA, and P.S. patterns | Are consent, frequency, sender identity, and claims appropriate for the campaign? | |
| Social copy | Templates for X, LinkedIn, Instagram, and TikTok or Reels | Are platform limits and recommendations still current? |
| Voice | Tone, pace, vocabulary, emotion, humor, perspective, sentence patterns, and do/avoid rules | Does the result preserve brand intent without imitating protected content too closely? |
The reference material contains historical conversion tips, platform limits, and example performance claims. Treat them as prompts to test, not current benchmarks or guarantees. Verify platform rules and campaign data at the time of use.
Extract a writing style deliberately
The packaged helper scans assets/writing-styles/ for approved text,
documents, images, or video. Its user-facing modes are --list,
--style <name>, --all, --json, and --verbose.
python scripts/extract-writing-styles.py --list
python scripts/extract-writing-styles.py --style founder-voice --jsonPlain .md and .txt files are read locally. PDF, DOCX, XLSX, and PPTX inputs
invoke Multix document conversion. Image and video inputs invoke Multix Gemini
analysis. The helper can therefore download and execute
@mrgoonie/multix, use the network, send source material to a model
provider, and require GEMINI_API_KEY for those external paths.
For document conversion, the helper writes a hidden temporary Markdown file
beside the source style asset and deletes it after a successful readable output.
Failed or interrupted runs need an explicit directory check. The release's
--all result does not match its text formatter's expected shape, so do not
rely on that mode without fixing or independently verifying it; process named
styles individually instead.
Observe the stages
- The Skill confirms the copy contract. It identifies the reader, awareness level, offer, channel, objective, CTA, proof, voice, constraints, and output boundary.
- It selects a structure. A formula and channel pattern organize the draft; they do not supply missing facts.
- It studies approved style evidence. It uses stated dimensions or analyzes authorized samples, separating observed traits from the source's content.
- It drafts and varies. The main copy and requested variants keep the same factual contract while testing a bounded message difference.
- It checks the work. Claims, specificity, reader fit, CTA consistency, tone, prohibited language, and channel constraints are reviewed.
- It returns evidence. The result names the chosen pattern, source facts, assumptions, unresolved proof gaps, variants, and any files or services used.
Keep claims, rights, and external effects controlled
A persuasive pattern is not evidence
Do not turn a template's sample numbers, guarantees, urgency, testimonials, or superiority language into real claims. Each factual statement needs approved support, and regulated or contractual copy still needs the appropriate human review.
- Get approval before sending source documents or media to an external provider, especially customer data, internal strategy, unpublished work, or licensed creative material.
- Style transfer should reproduce abstract traits such as pace and tone, not distinctive passages, personal likeness, or deceptive authorship.
- Do not create fake testimonials, logos, statistics, deadlines, stock levels, guarantees, security claims, or endorsements.
- The Skill may draft copy or write an explicitly requested artifact. It does not publish a page, send an email, post to a social network, start an A/B test, purchase media, or approve legal claims unless a separately authorized workflow performs that action.
Verify outputs and evidence
A useful result should include:
- The final copy in the requested format and length, plus clearly labeled variants when requested.
- Audience, awareness, offer, CTA, formula or structure, and style dimensions used to make the draft.
- A claim ledger that distinguishes supplied facts, cited proof, assumptions, placeholders, and statements still needing legal, brand, or subject-matter approval.
- Exact files read or written, temporary files checked, helper commands run, external provider use, and failures or skipped stages.
- A compact review against must-include, must-avoid, voice, accessibility, channel, and success-metric requirements.
The bundled writing-style guide presents six detailed preset definitions even though the Skill's reference table labels it as a 30-style quick reference. Do not report 30 ready-to-use styles. The stable and beta releases contain the same Skill files, and no directly scoped automated tests were found for the copy workflow or extraction helper.
Troubleshoot and interpret limits
| Symptom | Safe next step |
|---|---|
| The brief produces generic copy | Add audience awareness, concrete pain, differentiated offer, verified proof, channel, tone, one CTA, and clear exclusions. |
| A requested claim has no support | Mark it as a placeholder or remove it; ask for approved evidence rather than polishing it into certainty. |
| No writing styles are listed | Confirm the project root and assets/writing-styles/ location. The helper searches upward for .claude and otherwise falls back to its own directory. |
| Document or media extraction fails | Confirm format, provider credential, network approval, dependency execution, timeout, and hidden temporary files; use a local text sample when external processing is not approved. |
--all crashes or formats incorrectly | Use --list, then run --style <name> for each approved asset and combine reviewed results manually. |
| Social limits or performance tips look stale | Verify current platform documentation and your own campaign data before publishing. |
| The runtime does not recognize the Skill | Confirm target and scope, restart the runtime session, then follow Runtime cannot find a Skill or Agent. |
Use ak:docs for factual documentation or
ak:frontend-design when the copy belongs inside an
implemented interface.
Create and edit office artifacts with ak:document-skills
Route PDF, spreadsheet, Word, and PowerPoint work through the bundled format guidance with explicit intermediate files and visual or formula verification.
Author and render diagrams with ak:mermaidjs-v11
Turn a process, system, data model, timeline, or journey into Mermaid v11 source with explicit parsing, rendering, accessibility, and security evidence.