Generates audience-facing prose in a configured voice register and runs a dual review gate.
npx skills add https://github.com/TimSimpsonJr/copydesk --skill write-timsimpsonjr
Please help me install this Agent Skill. Skill: write Source: https://github.com/TimSimpsonJr/copydesk/tree/main/skills/write Command: npx skills add https://github.com/TimSimpsonJr/copydesk --skill write-timsimpsonjr
~/.claude/data/copydesk/registers/ and matches it to the writing context, asking when ambiguous. - Dual review gate: Dispatches prose-review and craft-review agents in parallel; hard fails (em dashes, fatal patterns, ChatGPT-isms) are fixed silently, everything else returns as an advisory table you accept or reject row by row. - Learning loop integration: Snapshots drafts and review dispositions so the copydesk:learn skill can sharpen rules from your hand edits. - Use Case: Paste research notes from an investigation and ask for a blog post; the Skill drafts in your personal register, fixes banned patterns, and returns the draft with an advisory table of suggested improvements. ## Quick Start Ask Claude to write a blog post from your research notes using your configured voice register.High-intent search queries and answers about installing and using this skill.
Run /copydesk:init once to build a voice register from 10-20 samples of your writing, then ask Claude to write the post. The write skill loads the matching register, drafts in that voice, and runs a dual review gate before showing you the result.
The skill dispatches two parallel agents after drafting: a prose-review agent that checks voice drift against your register and a craft-review agent that checks structural depth. Hard fails like em dashes and banned phrases are fixed silently; other findings appear as an advisory table.
Yes. You configure multiple registers as markdown files in ~/.claude/data/copydesk/registers/, each with trigger conditions in its frontmatter. The skill matches the writing context to a register and asks you to choose when the match is ambiguous.
The skill checks for at least one configured register before generating anything. If none exists, it stops and tells you to run /copydesk:init, which sets up the data directory and walks you through voice extraction from your writing samples.
Hard fails include em dashes in any form, the "this isn't X, this is Y" fatal pattern and its variants, AI vocabulary like "delve" and "it's worth noting", and ChatGPT-isms like "let's be honest" and "sit with that". These are rewritten silently before you see the draft.
yuqueAnalyze Yuque documents to extract vocabulary, sentence patterns, and tone into a reusable style profile.
cy-waliDraft marketing content for blogs, emails, social posts, and landing pages.
u9401066Validate content accuracy against source materials and generate verification reports.
elasticValidate documentation content against the Elastic style guide with Vale.