gooseworks-ai/goose-skills / skills/marketing/voice-card
voice-card
VerifiedBuilds or updates a voice card from a company blog ("extract our blog voice") or how an account or person actually writes, measured from about 8 or more of their real posts (length, casing, emoji, openings, pet phrases, opinions with their source). When there are too few posts, it runs a short interview instead. It ends with a test that checks a draft against the card. The brand's rules always beat the card. Use alone when you already have the account's posts or the person on hand; otherwise run the playbook for the job (e.g. run-organic-social, write-outreach, email-lifecycle).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.9 KB |
| references/blog-voice.md | markdown | 4.7 KB |
| skill.meta.json | config | 655 B |
| tests/examples/2026-10-22/blog-voice.md | markdown | 7.1 KB |
| tests/fixtures/company-blog.md | markdown | 6.8 KB |
| tests/test-by-hand.md | markdown | 2.4 KB |
references/blog-voice.md
View on GitHub# Human version
A blog voice guide records how the company actually writes, with examples and measurements a writer can reuse. It keeps different authors and channels visible and checks the result against the brand's confirmed rules.
---
# Agent version
## Gather the sample
1. Read the current `brand/voices/blog-<brand-slug>.md`, kit voice and brand rules using voice-card's precedence. Use supplied URLs or text first. Otherwise discover the company's own blog from its site/feed/sitemap with the host's public fetch/browser; read the article bodies, not search snippets or navigation. Target 6–10 varied articles: recent opinion/how-to/product content, with a few older pieces if voice change is relevant. Honour a supplied set/count instead of imposing that default. Log each title, URL or supplied-file location, author, publication date (or unknown), retrieval date, usable body-word count and why selected. No paid crawl or customer vendor key.
2. Separate company-authored, founder-authored, guest/sponsored and unknown-author pieces. Do not pool unlike writers into a fictitious single voice. Keep a company baseline and note distinct modes; a named person's voice follows voice-card's own-person rules. Fewer than five substantial articles or mostly snippets → a provisional guide with exactly which dimensions lack evidence; still analyse all supplied text. Ask for additional examples only in the closing decisions, not as a substitute for reading what was given.
## Measure and write
3. Measure by article and summarise median/range: sentence and paragraph length, article length, casing, punctuation/emoji, first/second/third person, jargon and explanations. Add long-form mechanics: heading style/cadence, list and emphasis patterns, opening/transition/conclusion moves, evidence/examples, CTA wording/placement and format. Define the counted unit (body words; headings separate), keep sample counts and source references. Show observed exceptions. With no performance data, mark performance unknown; publication is not proof it performed well.
4. Extract 3–5 specific writing rules with an actual on-voice quotation and a plausible off-voice version; distinctive vocabulary, terminology, tone/reader relationship, things the writer avoids, and channel/author differences. Bans inferred only from absence are style suggestions, not confirmed brand prohibitions. Screen quotations for claims: they are style evidence only, never proof or approval of a promise. Use safe quotations for examples a writer might reuse. Confirmed brand rules win; show clashes rather than silently replacing them.
5. Produce the voice-card shape, adapted to articles: source table and sample limits; rules with measurements; five short sourced examples (or all available, explaining a smaller sample); article opening/body/closing pattern; three generic → in-voice rewrites about a **supplied product fact or neutral topic**; terminology and never-says with status; performance known/unknown. Apply voice-card's match test to a fresh sample paragraph, including a blind line-up with source paragraphs and the claims check. Revise when the sample violates a measured rule. The guide must let a second writer reject an off-voice line, not merely list personality adjectives.
6. Show the guide and ask for corrections/confirmation. Save a sourced **provisional** card immediately; only the owner/author's actual yes confirms it, subject to voice-card's provisional-brand precedence. Their corrections are recorded in their words. Preserve any older confirmed rule until they approve its replacement. Do not change the kit voice, claims or positioning merely because a blog sample sounds different.
## Save, deliver and check
Use voice-card's existing `brand/voices/<handle>.md` save contract with handle `blog-<brand-slug>` and tags `[voice-card, voice-blog-<brand-slug>]`; supersede only this handle's earlier imported card. Read before write; source evidence is retained in the card. For a host with files, also deliver `gooseworks-reports/<brand-slug>/<YYYY-MM-DD>-blog-voice.md` (a same-day second run gets a suffix); hosted chat gets the full guide/artifact. Give the opening link and whether the brand save succeeded. Content/social writers use the applicable channel rules, not blog article length as a social-post limit.
Done when every supplied page is read or listed unread, measurements trace to article bodies, observed rules have quotations, the three rewrites and fresh paragraph pass the card/claims checks, author and coverage limits are visible, and the saved guide has honest confirmation status. A fetch that cannot read body text produces an incomplete/provisional result with the missing read named; never a manufactured corpus or a request to describe the voice in adjectives.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |