Clud-bug brand voice

The clud-bug product surface — cludbug.dev, the App's marketing pages, README, public docs — speaks like a 1950s naturalist's field guide. Observational, irreverent, reader-trusting. The brand DOES NOT explain its own jokes. Visitors who get the field-guide vocabulary get it; visitors who don't get a slightly mysterious naturalist vibe, which is the actual goal.

Your job on this skill is to catch copy that drifts away from this voice — toward generic SaaS speak, toward over-explaining the brand affordances, toward bossy second-person commands, toward features we haven't shipped.

Vocabulary keepers (use freely; never explain)

• plate, specimen, frontispiece, marginalia, field notes, observations
• Spec. as an abbreviation prefix on specimen-card tags (Spec. brand-voice, Spec. api-contract)
• Fake-Latin binomials in italics — Cluddus bugfindii, the signature joke. Form: capitalized genus + lowercase species + observational verb. Examples that work:
  • "Cluddus bugfindii, observed crawling on every PR."
• § numbering for sections, № for plates, MMXXVI Roman numerals on the masthead, Vol./No. notation
• Naturalist verb choices over corporate verbs: observed, catalogued, cites, pinned, foraged — not "delivers", "drives", "powers"

Section affordances (the field-guide UI scaffolding)

• Section heads: § I — Field Procedure, § II — Specimens for your habitat, § III — A Habit of Three Naturalists, § IV — Field Notes, § V — Observations from the field, § VI — Self-hosted alternative
• Plate cards: Plate I — Frontispiece, № I, the 🐛 bug-pin glyph
• Masthead: A Field Guide to Code Specimens · Vol. I · No. 1 · MMXXVI
• These are brand affordances, not labels-pretending-to-be-functional. Don't replace them with Step 1 / Step 2 / Step 3.

NEVER do (with reasons)

1. Explain the joke

Tonight we killed <em>Plate</em>: a labeled illustration in a field guide. <em>Frontispiece</em>: the cover plate. The gloss block was a dictionary entry under the plate. The joke landed weaker because the field-guide vocabulary had a translation layer pasted next to it.

• Bad: Marginalia (the annotations in the page margin) note that the App carries its own credentials.
• Good: <aside className="marginalia">No workflow file. No ANTHROPIC_API_KEY rotation. The App carries its own credentials.</aside>

Don't write parenthetical translations of plate, specimen, frontispiece, binomial, the Latin binomials themselves, or the Roman numerals. Don't add i.e., meaning, translates to, or footnote-style glossaries. Trust the reader.

2. Generic SaaS speak

Reflexive blacklist:

• seamless, powerful, transform your workflow, supercharge, blazing fast, 10x, robust, scalable, world-class, enterprise-grade, mission-critical, next-generation, cutting-edge, best-in-class, drive value, unlock potential, leverage, synergize, take it to the next level

Each of these is a copy-paste from every other AI dev tools homepage. They have no information content. Replace with a specific observation:

• Bad: "Powerful AI PR review that scales with your team."
• Good: "AI PR review with project-aware skills. Install the GitHub App, write a skill, get reviews graded against your conventions on every pull request." (from site/app/page.tsx hero)

3. Exclamation marks outside success-state microcopy

Toast like Review posted! is fine. Hero like Install today! is not. The voice is observational, not salesman-excited. One exclamation per page max, and only when something actually completed.

4. Second-person bossy

• Bad: "You should install the GitHub App."
• Bad: "You must configure your skills."
• Good: "Install the GitHub App on the org or repos you want reviewed. Approve the permissions for pull requests, contents, and checks." — describes the action, doesn't command.

Observe, don't command. The reader knows they're reading; they don't need to be told to act.

5. ALL-CAPS CTAs

• Bad: INSTALL THE GITHUB APP
• Good: Install the GitHub App with sentence-case button styling.

6. Generic step-explainers without field-guide framing

Step 1 / Step 2 / Step 3 is corporate. The field-guide form is § I — Field Procedure: Three steps. Two minutes per review. (with naturalist verbs: Install / Authorize / Reviews land).

Locked product facts

Multi-pass. Beetle = Pass 1 broad scan. Wasp = Pass 2 cross-check (catches what Beetle missed, suppresses false positives). Mantis = Pass 3 arbiter on disputes only. Order locked. Names locked (NOT Inspector/Auditor/Arbiter — those were pre-2026-06). Multi-pass is App-tier only — never claim it on the npm/OSS workflow path (PR #165 reviewer caught this). Honest hedge form: "Same skill engine, same review quality. Multi-pass is App-tier only — the hosted bot orchestrates the three roles server-side."

App-first. Primary CTA every page: "Install the GitHub App" → https://github.com/apps/clud-bug/installations/new. Secondary, below the fold: "Self-hosted? Use the open-source workflow →". Never lead with the workflow Action.

Pricing. Solo $9/mo · Team $29/seat/mo · Pro overage 1.20× Anthropic. Always link to app.cludbug.dev/pricing — don't restate prices on other surfaces (the lib/pricing-display.ts drift gate exists because we already had this bug). Never name a fixed $/review — overage anchors to actual Anthropic cost × multiplier.

Domains. cludbug.dev = marketing site. app.cludbug.dev = product (dashboard + /install + /pricing + /compare). When visible link text is bare domain app.cludbug.dev, href must be bare domain — not a subpath (PR #165 footer-fix lesson).

Brand exemplars — quote these as anchors when reviewing

These are the load-bearing lines from site/app/page.tsx. When marketing copy lands in review, cite these as the voice anchors. If the new copy could swap in for one of these without changing the brand vibe, it's on-brand. If it'd feel jarring next to them, it isn't.

"AI PR review with project-aware skills." (hero subtitle)

"Install the GitHub App, write a skill, get reviews graded against your conventions on every pull request." (hero body)

"— Cluddus bugfindii, observed crawling on every PR." (binomial italic, the brand signature)

"No workflow file. No ANTHROPIC_API_KEY rotation. The App carries its own credentials and posts inline comments under its own GitHub identity." (marginalia voice)

"Drop a Markdown file into .claude/skills/ and Clud Bug cites it by name on every review. Your team's standards become the reviewer." (specimens marginalia)

"Three steps. Two minutes per review." (§ I subtitle — short, observational, no salesman energy)

"Same skill engine, same review quality. Multi-pass (Beetle / Wasp / Mantis) is App-tier only — the hosted bot orchestrates the three roles server-side." (§ VI honest hedge — the model for any feature-availability copy)

What this skill does NOT cover

• Internal log strings, error messages, function names, code identifiers — brand-voice-review (the generic catalog skill) handles these if installed.
• API contract / schema review — api-contract-enforcement covers that.
• Accessibility (contrast, ARIA, etc.) — separate skills handle those.

If a review finds a violation that doesn't fit one of the buckets above, surface it under the most specific skill that matches. Avoid double-flagging.