ssheleg/super-ux / plugins/super-ux/skills/brand-voice
brand-voice
VerifiedUse when defining or checking how a product speaks: tone of voice, terminology, canonical facts, register. Triggers: "tone of voice" / "тон оф войс", "brand voice" / "голос бренда", "брендбук", "наша терминология". Not for writing the text (copywriting). Refusal phrase "no brand" / «без бренда».
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 7.8 KB |
| references/ai-tells.md | markdown | 20.6 KB |
| references/brand-contract.md | markdown | 25.4 KB |
| references/channel-playbooks.md | markdown | 20.0 KB |
| references/hooks.md | markdown | 26.4 KB |
| references/localization.md | markdown | 5.2 KB |
| references/marketing-copy.md | markdown | 8.1 KB |
| references/research-outliers.md | markdown | 5.4 KB |
| references/seo-aeo-safety.md | markdown | 9.9 KB |
| references/store-copy.md | markdown | 7.4 KB |
| references/surface-registers.md | markdown | 9.1 KB |
| references/system-map.md | markdown | 12.4 KB |
| references/video-script.md | markdown | 14.6 KB |
| references/voice-packs.md | markdown | 15.9 KB |
| templates/brand/README.md | markdown | 2.3 KB |
| templates/brand/channels.md | markdown | 4.2 KB |
| templates/brand/copy-brief.md | markdown | 187 B |
| templates/brand/facts.md | markdown | 1.3 KB |
| templates/brand/locale.md | markdown | 2.2 KB |
| templates/brand/strings.md | markdown | 1.7 KB |
| templates/brand/terminology.md | markdown | 1.2 KB |
| templates/brand/voice.md | markdown | 2.2 KB |
references/brand-contract.md
View on GitHub# Brand Contract (v1): Voice, Terminology, Facts, Channels, Strings, Locales
This is THE contract for `docs/brand/` in a target project. The `brand-voice`
and `copywriting` skills, the `copy` scope of `ux-audit`, `brand_lint.py` and
the Cursor rules all follow it. Do not deviate from file names, field names,
statuses or surface names, because tooling keys off them. The reasoning behind the
model lives in [surface-registers.md](surface-registers.md); the pack library
it selects from is [voice-packs.md](voice-packs.md).
## Contents
- [Files in the target project](#files-in-the-target-project)
- [`copy-brief.md`](#copy-briefmd)
- [The marker](#the-marker)
- [Where the voice comes from](#where-the-voice-comes-from)
- [`voice.md`](#voicemd)
- [`terminology.md`](#terminologymd)
- [`facts.md`](#factsmd)
- [`channels.md`](#channelsmd)
- [`strings.md`](#stringsmd)
- [`locales/<code>.md`](#localescodemd)
- [`README.md` and the `Sources:` block](#readmemd-and-the-sources-block)
- [The three rules that hold the layer together](#the-three-rules-that-hold-the-layer-together)
- [The check codes](#the-check-codes)
- [Versioning](#versioning)
`docs/ux/` answers **what the product does**. `docs/brand/` answers **how it
speaks**. They are separate roots on purpose: the brand pack also governs
surfaces that are not UX at all: a blog post, a store listing, an ad.
## Files in the target project
```
docs/brand/
├── voice.md # identity: pack, axes, narrative, locale transfer
├── terminology.md # our words, banned words, entity and tier names
├── facts.md # canonical numbers and proof, the only source
├── channels.md # one record per surface: register, limits, bans
├── strings.md # interface string registry -> file:line -> scenario
├── copy-brief.md # current marketing task inputs, added when needed
├── locales/
│ └── <code>.md # per-locale delta (en.md, ru.md, de.md, …)
├── README.md # this map, project-side, plus the Sources: block
└── lint.py # the brand linter, seeded by the installer
```
## `copy-brief.md`
The current marketing writing task's input record, owned by `copywriting`.
It is optional for existing packs until the next marketing writing task; other
UX and interface-only tasks do not need to load or create it. Seed from the
copywriting skill's template. Keep past briefs in Git history; a new task
rechecks the current Surface and Action before reusing it.
```markdown
Contract: brand-contract v1
Surface: unknown
Action: unknown
Reader: unknown
Customer words: unknown
Proof: unknown
Success metric: unknown
Constraints: unknown
Voice reference: unknown
```
Each field occurs once and has a value. `unknown` is a supported unresolved
input, not evidence, approval or a successful result. `Surface` identifies the
channel/page; `Action` names its one ask; `Reader` traces a persona/job/stage;
`Customer words` links source language and terminology; `Proof` links existing
facts; `Success metric` names an observable outcome and only records a target
when supplied; `Constraints` records supplied limits or explicit `none`.
`Voice reference` is `voice.md#voice-references` (its filled Admired/Refused pair)
or `unknown`; keep the voice's source rather than copying its text. Copywriting
reports gaps and never changes voice/facts to make the brief look complete.
`B074` advises migration when marketing sources have no brief. `B075` refuses a
present brief with missing/empty/duplicate fields or a broken voice reference.
It verifies structure and the local voice pointer, not whether a metric or
assignment is true. Other provenance remains an agent review responsibility.
## The marker
The **first line** of every file under `docs/brand/`, including
`locales/<code>.md`, is exactly:
```
Contract: brand-contract v1
```
`brand_lint.py` reports a missing marker as `B001` and disagreeing markers
across files as `B002`. `ux_doctor.py` reads the same line to tell a base
written to an old contract from one written to the current one. A base three
versions behind is internally consistent, which is precisely why a linter
cannot see it.
## Where the voice comes from
The pack is **derived from** `docs/ux/foundation.md`: personas, jobs, and
what the user stands to lose when the product fails. The dependency runs one
way. A persona does not change because a tone was appealing.
`voice.md` records the trace in `Derived-from:`. When no foundation exists,
`brand-voice` still works, stamps `Derived-from: inferred`, and says plainly
that the WHY layer should be built.
---
## `voice.md`
```markdown
Contract: brand-contract v1
Voice pack: operator-brief
Locales: en (primary), ru, de
Locale parity threshold: 80%
Derived-from: P-01, P-03, JTBD-02
Status: validated
Humanization: on
Humanization pass: own
Last calibrated: 2026-08-05
## Axes
| Axis | The product IS | The product IS NOT |
|---|---|---|
| Confidence | states its position | "we think this could help" |
| Register | terse, operational | corporate marketing speak |
| Distance | peer to peer | vendor to account |
| Humor | dry, rare, earned | quirky, emoji-led |
| Density | every sentence carries weight | filler, throat-clearing |
## Narrative
Hero: the operator who owns the outcome
Enemy: guesswork that costs money before anyone notices
Product role: instrument
Promise: you stop guessing within one run
## Invariant in every language
- never hedges
- never claims a number that is not in facts.md
## Reconsidered per locale
- address form
- humor level
- idioms and wordplay
```
**Header fields.** `Voice pack` is a pack id from
[voice-packs.md](voice-packs.md) or `custom`. `Locales` lists every locale,
marking exactly one `(primary)`. `Locale parity threshold` is the percentage
below which `B071` warns. `Derived-from` lists `P-` / `JTBD-` ids from
`foundation.md`, or the single word `inferred`. `Last calibrated` is an ISO date.
**A header value ends where an aligned comment begins.** Two spaces or more
before a `#` starts a comment and the value stops there, which is how the
seeded templates annotate a field. One space does not, because a reason may
legitimately read `per ticket #431` and truncating it would leave no reason
at all.
**`Humanization` and `Humanization pass` answer two different questions**, and
conflating them is why the second one existed for releases while nothing read
it. `Humanization` is **whether the pass runs at all**; `Humanization pass`
names **which implementation** runs, and absent it is `own`, the only one that
reads this pack's registers and canonical facts.
`Humanization` defaults to `on`, and it is a default rather than a preference:
every mode that produces text runs the sweep, because a draft nobody swept
carries the markers `ai-tells.md` grades and a reader registers them before
they can say why. An absent field is `B064` at warning level, since the default
is the safe state. `off` is legitimate and it is a decision that outlives the
person who made it, so it requires a `Humanization declined:` line carrying the
reason and the date; without one it is `B064` at error level. The value is
what every status this pack prints reports, so a project can always see which
state it is in without reading a file.
**The `Status` enum, in one home:**
- `voice.md` **Status** — `draft | validated`
- `voice.md` **Humanization** — `on | off`
`validate_status_enums_match_contract` reads that line and compares it against
`VOICE_STATUSES` in `brand_lint.py`, so neither side can move alone. It is here
because the two had already disagreed and it worked by accident: this pack's own
`voice.md` said `Status: approved` for two releases, and every read in the linter
asked `== "draft"` or `!= "draft"` — so `approved` behaved like `validated` and
would have read as **not** validated the first time any check tested for the
value. An out-of-enum status is `B034`.
**Status lifecycle.** Same as a scenario: `draft` until the operator has seen
and approved it, then `validated`. `approved` is not a third state; it is what
the operator does, and `validated` is what the file then says. Any edit drops it
back to `draft`.
**The five axes are fixed**: `Confidence`, `Register`, `Distance`, `Humor`,
`Density`. A pack fills them; a project may sharpen the wording but may not
add or remove an axis, because `channels.md` expresses register as deltas
against exactly these five.
---
## `terminology.md`
```markdown
Contract: brand-contract v1
## Product terms: always
| Our term | Never write | Applies to |
|---|---|---|
| Run | Execution, Job | what the product performs |
## Entity and tier names: exact spelling
| Name | Wrong forms seen |
|---|---|
| Pro | PRO, Pro plan, pro |
## Banned
| Word or phrase | Why | Use instead |
|---|---|---|
| leverage | filler verb | use |
| seamless | claims what it cannot show | describe the step that disappeared |
## Glossary
| Term | Meaning |
|---|---|
```
The **Banned** table seeds from three places at init: weak verbs and
buzzwords, hedging chains, and the marker vocabulary in
[ai-tells.md](ai-tells.md). Calibration adds what is specific to the product.
`brand_lint.py` reads the first column of **Product terms** and **Banned**
for `B010` and `B011`, and the **Entity and tier names** table for `B012`.
---
## `facts.md`
```markdown
Contract: brand-contract v1
| Fact | Value | Source | Checked | Review by | Public |
|---|---|---|---|---|---|
| tools in the catalog | 448 | api/catalog.json | 2026-08-05 | 2026-11-05 | yes |
| median run cost | $3.10 | internal billing export | 2026-07-30 | 2026-10-30 | no |
```
**This is the only source of any figure in public copy.** A number in a
public surface with no row here is `B030` and blocks. `Public: no` marks
internal figures that must never be quoted. A row with no `Source`, or past
its `Review by`, is `B031` and warns.
A missing fact is reported, never invented to close a gap.
### A claim binds to a row, not to digits
**A numeric coincidence is not proof.** One row `supported integrations = 500`
must not license "We serve 500 million paying customers" — the digits match and
the subject, unit and population all differ. So a public claim resolves against
a fact ROW, never against the bag of known figures:
| Axis | In the claim | Against the row |
|---|---|---|
| claim-id | a stable id where the copy registers its claims | — |
| fact-id | the `Fact` name the claim cites | the row must exist |
| subject | what is being counted | must match the row's subject |
| unit | integrations, customers, ms, % | must match |
| population | whose: all users, paying, one region | must match |
| date | as-of when the claim speaks | inside the row's `Checked`→`Review by` window |
| transformation | rounding or aggregation applied | only those the row declares |
A transformation the row does not declare (`round down`, `sum with <fact>`)
makes a NEW fact that needs its own row. A claim that resolves no row is
**unresolved and prints as UNVERIFIED** — it never passes because its digits
happened to appear in the sourced set.
**What the linter checks, and what it cannot.** `B030`'s promise is
token-level: a KNOWN numeric token traced to a row, unit and precision beside
it. Whether the sentence's SUBJECT matches the row is the semantic audit's
question — a reader's, not a regex's. The extractor's blind spots are stated,
not papered over: bare numbers under 100 and year-like tokens need context —
"2026 integrations" is a figure, not a year, so a token is read as a year only
when nothing is counting it.
---
## `channels.md`
One record per surface, in this shape:
```markdown
### landing hero
Register: confidence +1, density +1, humor -1
Format: one headline under 60 characters, one subhead, one CTA
Limits: title 60, meta description 160
Forbidden: physics: none | brand: superlatives without a facts.md row
CTA: one primary, verb plus outcome
Proof: one number, sourced
Locales: de headline budget 60 * 1.30
```
**Register** is expressed as deltas against the five axes in `voice.md`.
**Forbidden** always carries both halves, `physics:` and `brand:`, even when
one is `none`; see the second rule below.
**What the linter reads in a record.** `Limits:` is `field N` pairs, `body`
meaning the document body (`B040`, `B073`). In the `physics:` half of
`Forbidden:` it reads four tokens and nothing else: `link in body` (`B042`),
`max N hashtags` (`B043`), `fold N`, the characters the feed shows before
"… more" (`B045`), and `one ask` (`B046`). A physics clause in other words is
documentation for a reader, not a rule the linter applies, so the platform
fact belongs in the playbook and the token belongs here.
**Short video.** A script document carries its hook in front matter:
`hook:` (and `hook-2:`, `hook-3:` … for variants) for the first spoken line,
`on-screen:` (`on-screen-2:` …) for the overlay text. `B044` filters each hook
and the overlay length; `B030` reads both fields as public copy. The method
and its limits are in the `copywriting` skill's `hooks.md`.
### Surfaces
Product: `primary action` · `empty state` · `error` · `loading` ·
`success/toast` · `onboarding` · `paywall and upgrade` ·
`destructive confirm` · `billing and receipts` · `settings and legal` ·
`transactional email and push` · `docs and help`.
Marketing: `landing hero` · `landing body` · `pricing` · `blog` ·
`changelog` · `X` · `Instagram` · `TikTok` · `YouTube Shorts` ·
`short video` · `Reddit` · `LinkedIn` · `HN and Product Hunt` ·
`App Store` · `Google Play` · `ads` · `lifecycle email`.
`Instagram`, `TikTok` and `YouTube Shorts` are the caption or post on that
platform; `short video` is the spoken script and on-screen text, shared across
the three when one video ships to all of them.
A project may omit a surface it does not have. It may not rename one: the
linter and the audit both address surfaces by these names.
---
## `strings.md`
```markdown
Contract: brand-contract v1
| Key | Text (primary) | Location | Scenario | Status | Kind |
|---|---|---|---|---|---|
| action.project.publish | Publish | src/ui/ProjectBar.tsx:47 | SCN-014 | agreed | copy |
| help.usage.table | ` --cursor <dir> Cursor rules` | bin/cli.js:12 | SCN-002 | agreed | layout |
```
Statuses: `agreed` · `proposed` · `drifted` · `orphan`.
**`Kind` decides which rules may judge the row**, and it is the sixth column so
that a five-column registry written before it existed keeps every index it had.
Absent means `copy`, because the safe default is to check.
- `copy` is language. Every language rule applies: banned words, product terms,
entity spelling, casing, the dash, a full stop after a label.
- `layout` is a string whose shape carries the meaning: a column-aligned option
table, an ASCII frame, a banner. Those rules would be judging typesetting, so
they do not run. `B022` still does, which is the point of registering it
rather than exempting it from registration: the string is a decision somebody
made, and calling it furniture must not be a way to hide a real one.
A row is `layout` because its **shape** is load-bearing, never because a rule
was inconvenient. Two strings inside one aligned block can differ: the block's
title is `copy`, and the aligned rows under it are `layout`.
- `strings.md` **Kind** — `copy | layout`
**This is a decision registry, not a message catalog.** It does not replace
i18n keys and holds no translations. It records which strings have been
reconciled with the pack, which scenario each serves, and where it lives.
It is what makes `B020`, one action under two names, checkable at all; without
it that defect is only findable by reading the whole interface.
`Key` is dot-separated and stable. Two rows sharing a `Key` with different
`Text` is `B020`. A `Location` that no longer resolves is `B023`.
The registry is populated by an inventory sweep at init, not by hand.
---
## `locales/<code>.md`
```markdown
Contract: brand-contract v1
Locale: de
Primary: no
Address form: Sie
Length coefficient: 1.30
Humor: -1 from base
Never translated: product name, entity names, tier names
Keywords: own research, not translated from primary
Dead idioms: "ship it" -> "raus damit"
Legal differences: Impressum, VAT-inclusive pricing
```
`Length coefficient` multiplies every field limit in `channels.md` when the
linter checks that locale (`B040`, `B073`).
Two rules:
1. **The primary locale is the source of meaning, not of form.** A
word-for-word CTA translation is a finding (`B072`) even when the grammar
is perfect.
2. **A locale need not be complete, but must declare that it lags.** Parity
below `Locale parity threshold` warns (`B071`) with the percentage, rather
than letting an unfinished locale look finished.
---
## `README.md` and the `Sources:` block
The project-side map, and the one thing the linter cannot infer: where this
project keeps its text.
```markdown
Contract: brand-contract v1
Sources:
ui: src/**/*.{ts,tsx,js,jsx,vue,svelte}
marketing: content/**/*.{md,mdx}
store: store/{ios,android}/*.md
robots: public/robots.txt
locales: src/locales/*.json
```
Nothing outside these paths is scanned. A missing block is `B006`; an empty
declared source glob is `B009`. Both block. Comma braces are expanded, including
the examples above; overlapping patterns read a file once per source key.
`robots` is exempt from the generic empty-glob check because `B050` owns its
crawler policy. Omitted keys are unscanned, not passed; the findings array does
not provide a whole-project coverage verdict. Report which source keys and
files the changed surface actually used.
HTML sources contribute static visible text, not markup or quoted attributes.
B021 checks explicitly registered strings against that full visible text. B022
suggests registry rows only for explicit HTML links, controls, labels and
alert/status roles; a marketing paragraph does not become an interface decision
because one action on its page is registered. Adjacent navigation/CTA anchors
remain separate candidates; ordinary inline prose links do not split sentences.
CSS-styled link cards may still need human review of the extracted control name.
`B063` reads h1–h6 with nested inline tags, entities and `<br>` fragments, as well
as Markdown ATX headings and front-matter titles. Preserve those roles in any
copy projection. Hidden/template/script text is excluded; `aria-hidden` alone
is visual text. Browser review covers CSS and runtime content this parser cannot
resolve. Normal paragraph punctuation is outside the title rule.
`ui` and `marketing` also classify a finding, because several checks apply to only
one of the two.
---
## The three rules that hold the layer together
1. **Register moves the axes; it never crosses the invariants.** A Reddit
post may run long, drop the CTA and turn self-deprecating. If the brand
does not hedge, it does not hedge there either. The invariants are listed
in `voice.md`; everything else is negotiable per surface.
2. **Platform physics and brand choice are separate fields.** "A link in the
post body suppresses reach" and "we do not post links" are different
claims. Merged into one line, nobody can tell six months later which was
an algorithm and which was a decision, so `Forbidden:` always carries
both halves.
3. **Humor is forbidden on `error`, `destructive confirm`,
`billing and receipts` and `paywall and upgrade`**, in every pack,
including `playful-consumer`. The user is losing data, access or money at
that moment; a joke reads as mockery. Enforced as `B061`, not left to
taste.
---
## The check codes
`brand_lint.py` emits these and nothing else. The contract owns the meanings;
the linter owns the detection. A code the linter can emit that is absent here
is a validator failure, because its meaning would otherwise live only in the
source of the thing doing the checking.
Severity is stated per code: **E** blocks with exit 2; **W** reports without
blocking by default. `--strict` makes any warning exit 1; `--fail-on B063` makes
only selected warning codes exit 1 (repeat the flag or use comma-separated
codes). Unknown warning codes fail with usage/exit 2. Errors always block.
`--json` remains an array of findings; selecting blockers does not hide advice.
| Code | | Fires when |
|---|---|---|
| B001 | E | a file under `docs/brand/` has no contract marker |
| B002 | E | markers disagree across the pack |
| B003 | W | `voice.md` is `draft` while `strings.md` holds agreed rows |
| B004 | E | `Derived-from` cites an id absent from `foundation.md` |
| B005 | W | `foundation.md` changed after `Last calibrated` |
| B006 | E | `README.md` has no `Sources:` block, so nothing to scan |
| B007 | W | `## Voice references` names no admired or no refused brand, once the voice leaves `draft` |
| B009 | E | a declared non-robots source glob matches no files |
| B010 | E | a banned word appears in a registered string |
| B011 | E | a generic word used where a product term exists |
| B012 | E | an entity or tier name spelled inconsistently |
| B020 | E | one action carries two different names |
| B021 | E | a registered string diverged from the code |
| B022 | W | a code string or explicit HTML control/message has no registry row, reported at the line it first appears on; ordinary marketing prose is outside the HTML candidate sweep |
| B023 | E | a registry row points at a location that does not exist |
| B024 | E | declared casing violated |
| B025 | W | a button label names no outcome |
| B026 | W | a label, button, menu item or title ends in a full stop |
| B030 | E | a KNOWN numeric token in public copy has no row in `facts.md` — token-level only: subject/population match is the semantic audit's job, and a claim resolving no row prints UNVERIFIED, never a silent pass |
| B031 | W | a fact has no source, or is past its `Review by` |
| B032 | E | a superlative with no fact beside it |
| B033 | E | two rows in `facts.md` under one `Fact` name — a figure cited by that name is ambiguous, and the duplicate also widened the sourced set |
| B034 | E | `voice.md` carries a `Status` the contract does not declare — every read here asks whether it is `draft`, so an unrecognised value behaves like `validated` today and reads as not-validated the moment a check tests for the value |
| B040 | E | a field exceeds its surface limit |
| B041 | E | an iOS keyword-field rule broken |
| B042 | E | a link in a body where the surface's physics forbid it |
| B043 | W | more hashtags than the surface tolerates |
| B044 | W | a short-video hook fails the filter: it scores WEAK, carries a dealbreaker (a "stop scrolling" opener, a video preamble, a greeting, a hashtag, an emoji), or its on-screen text runs past ten words. A filter, not a predictor: AUC 0.83 at catching a bad hook, 0.56 at picking a winner; a hook in a non-Latin script gets only the hashtag and emoji dealbreakers |
| B045 | W | a caption's first line runs past the surface's `fold`, opens with a hashtag or a mention, or spends the window on a hashtag |
| B046 | W | a caption on a `one ask` surface makes no ask, or more than one |
| B050 | E | AI search declared a target while a retrieval agent (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User) is blocked, by its own group or by `*`; a training-crawler block is never reported |
| B051 | W | a token REPEATS unnaturally on one rendered page (advisory: ≥5x and >4%, min 40 words, registered terms exempt) — Google's spam policy is manipulative repetition, not a fixed percentage; no claim about citation likelihood |
| B052 | E | a filler opener |
| B053 | W | no named author where the surface needs one |
| B054 | W | the title promises more than the body delivers |
| B060 | W | machine-drafting markers above the advisory threshold; not an authorship verdict |
| B061 | E | humor where the user is losing something |
| B062 | E | AT-06, a rhetorical dash standing in for a full stop, comma or colon |
| B063 | W | AT-07, a document title, Markdown ATX heading or HTML heading fragment ends in a full stop |
| B065 | E | a registry row carries a `Kind` the contract does not declare |
| B066 | W | AT-16, an invisible or look-alike-space character in a declared source file, outside the numbered exemptions; it reports and never rewrites, and it never examines a dash or a quotation mark |
| B067 | E/W | AT-06's hard layer, the em dash (U+2014) in newly authored text (owner decision 2026-10-18): an em dash joining a range errors (write an en dash); an em dash opening a line where the locale has no dialogue dash errors; a lone dash in a locale with a grammatical dash warns, because the copula is allowed only where no rewrite reads naturally. Never a second finding for a sentence `B062` reports |
| B064 | W/E | the humanization pass: absent field warns that the default `on` applies unrecorded; an out-of-enum value errors; `off` with no `Humanization declined:` reason errors |
| B070 | E | a declared locale has no locale file |
| B071 | W | locale parity below the declared threshold |
| B072 | W | a locale row left identical to the primary |
| B073 | E | a field overflows under the locale's coefficient |
| B074 | W | marketing sources have no current copy brief; advisory migration |
| B075 | E | present copy brief has missing, empty or duplicate inputs, or an unresolved voice reference |
---
## Versioning
The version in the marker moves when a field is renamed or removed, or when a
new required field appears. Adding an **optional** field does not move it,
the same rule the UX contract follows. `ux_doctor.py` reports what each
version introduced, so a project can see what it is missing without reading
the diff.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |