菜单
Voice & Presence — fires during cruise content writing. Guards the sound of the prose so it reads as written by someone who has actually sailed the route, not by a brochure or a model. Shapes voice markers, rhythm, vocabulary, and pastoral honesty. For post-draft diagnostics, see voice-audit.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
The dedication under which all work in this household is done — every task offered to God, excellence as worship. Automatic posture skill, present and byte-identical in every repository and sibling workspace. Preserve the invocation; getting it right is the worship, not getting it fast.
Hard ban on any reference to the external domain getbets.us on cruisinginthewake.com. Enforced via PreToolUse hook (.claude/hooks/no-getbets-guard.js) and git pre-commit hook (.githooks/pre-commit). Fires automatically on every Write/Edit/MultiEdit and every commit; no manual invocation.
Post-draft diagnostic for InTheWake content. Scans for cruise-marketing tells, AI authorship-cluster signals, and absence-of-authenticity gaps. Assesses authenticity risk by clustering signals across layers (Layer 1 strong, Layer 2 supporting, Layer 3 counter-signals). Required: every voyage-pack ships with a voice-audit attestation in its .factcheck.json sidecar. Fires before committing content edits or before publishing a new page. For during-writing standards, see like-a-human. For corpus measurement, see voice-dna.
External red-team critique pass. Sends a commit range or proposed change to a second model (defaulting to a different LLM) with the explicit charter to find every flaw "as though you're trying to take the engineer's job." Grok-native usage notes.
Foundational integrity guardrail. Automatic on every file modification and any non-trivial task. Demands verified claims with specific evidence, explicit scope, anomaly dispositions, and rejection of narrow or vague assertions. The single most important discipline on this project. Grok-native adaptation.
| name | like-a-human |
| description | Voice & Presence — fires during cruise content writing. Guards the sound of the prose so it reads as written by someone who has actually sailed the route, not by a brochure or a model. Shapes voice markers, rhythm, vocabulary, and pastoral honesty. For post-draft diagnostics, see voice-audit. |
| version | 3.2.0 |
The reader can tell a sailor from a brochure inside one paragraph. They can also tell a model from a sailor inside one paragraph. This skill shapes prose so the answer is always sailor — someone who walked off the pier, paid the price, watched the queue form, ate the food, and reports back.
The standard is not detection-software-pass. The standard is a 62-year-old first-time cruiser saying "this person has been where I'm going."
If prose lands all four, voice is intact regardless of polish.
Prose disappears when unnoticed. Avoid:
Human writing has seams — places the prose changes direction, where a tension stays unresolved, where the writer admits something they didn't expect to admit. Protect those seams. Don't sand them.
Cruise-marketing tells (zero tolerance):
world-class, stunning, luxurious, elevate, elevating, unforgettable, pristine, breathtaking, majestic, idyllic, paradise, exclusive, indulge, indulgent, opulent, lavish, sumptuous, hidden gem, must-do, must-see, bucket-list, jaw-dropping, dream destination
Generic AI tells:
delve, tapestry, leverage, framework, holistic, unpack, resonate, garner, showcase, underscore, encompass, nestled, enduring, boasts, facilitate, exemplify, robust, seamless, seamlessly, vibrant, nuanced, comprehensive, in essence, dive into, dive in, navigate (as metaphor), unlock, transform
Filler transitions:
Moreover, Furthermore, Additionally, In essence, In conclusion, Ultimately, At its core, Whether you're, Look no further than
Promotional drift:
offers, features, provides (as a verb of marketing), boasts, showcases, presents — use plain copulas (is, has, includes) instead.
AI replaces plain "is/are/has" with inflated constructions. The watchtower IS the watchtower. The buffet IS the buffet. The waterpark IS open from 10 a.m.
Replacement table:
| Inflated | Plain |
|---|---|
| serves as | is |
| stands as | is |
| represents | is |
| constitutes | is |
| marks | is |
| features | has / includes |
| offers | has / serves |
| provides | has / serves |
| boasts | has |
| presents | has |
| showcases | shows / has |
Use is. Use are. Use has. Plain copulas are the backbone of clear cruise reporting.
Kill decorative adverbs: quietly, deeply, fundamentally, remarkably, arguably, certainly, really, just, literally, genuinely, honestly, simply, actually, truly, definitely, particularly, especially.
If an adverb is doing work the verb cannot do alone, keep it. If removing it changes nothing, cut it. "He quietly suggested" → "He suggested." "Genuinely surprising" → "surprising." "Particularly good" → "good" (or name what was good about it). The verb does the work. The adverb watches.
Exceptions: adverbs in actual quoted reviews, stated specifications ("sails weekly"), and schedule descriptions are untouchable. The kill list applies to editorial adverbs, not factual ones.
AI defaults to triads: three adjectives, three examples, three bullet points, three applications. Real cruise reporting follows the shape of the fact, not a template. Sometimes two is right. Sometimes four. Sometimes one strong observation is better than three weak ones.
Let the fact determine the grouping. If a port has five observation decks, use five. If a ship has two main pools, use two. If a buffet station has only one redeeming dish, name it once and stop. Do not force triads.
Triplet-of-absences sub-rule (v3.2.0). A triplet built entirely from negations ("no urgency, no upsell, no countdown") is a corpus-native brand move and defensible — once per top-level document. Twice in the same document slides into AI-rhythm shortcut. Cap: one triplet-of-absences per document. If you reach for a second, either pair it with a concrete positive ("...no countdown — this pack is the calm read on one specific week" — the second half anchors the brand claim) or restructure.
AI rotates through synonyms to avoid repeating a word: "the port... the destination... the locale... the stop." This is a tell because it prioritizes surface variety over clarity. If you mean the same thing, use the same word or "it."
"The bridge was completed in 1910. The bridge connects the harbor and the old city" is better than "The bridge was completed in 1910. The structure connects the harbor and the old city." Repeat the noun. Don't cycle.
AI generates the most statistically likely content. Humans include oddly specific, pattern-breaking details that no algorithm would produce: the creak of the gangway, the smell of teak oil after rain, the exact weight of a casino chip, the moment you noticed the ship had moved away from the dock without you feeling it, the sound the elevator makes between decks 8 and 9. These details are what make a logbook feel lived rather than generated.
When writing pastoral application, sensory observation, or first-person attestation, reach for the specific detail that only a person who has been there would know.
AI tacks present participle phrases onto the end of sentences to add fake depth: "The pool deck filled up by 9 a.m., reflecting the popularity of family cruises." Cut everything after the fact. If the editorial matters, make it its own sentence with its own evidence.
The rule: end sentences at the fact. State what happened. Stop. If you need to say why it matters, start a new sentence. The -ing clause is almost always a throat-clear disguised as analysis.
Watch for: "highlighting the appeal," "reflecting broader trends," "underscoring the value," "ensuring continued relevance," "showcasing the line's commitment to," "emphasizing the need for," "solidifying its position as."
Distinguish from gear-shift markers. "Here's the thing:" followed by a genuine reporting move is a gear-shift — keep it. "But here's the truth:" followed by a restatement or platitude is synthetic earnestness — cut it. The marker is fine. The platitude is the tell.
Also watch for: "Perhaps the most striking thing is..." "It is a reminder that..." "But here's what really matters..." If what follows is genuinely new content, the phrase earns its place. If what follows is something already said, the phrase is AI throat-clearing dressed up as insight.
"From X to Y" requires X and Y to be endpoints of a real scale. "From the pool deck to the engine room" works. "From relaxing to adventurous" does not — those are not on a meaningful scale. AI uses this construction to gesture at breadth without being specific. Name the actual items instead.
Self-help and consultant vocabulary creeps into travel writing through specific verbs: optimize (your itinerary), unpack (your options), process (the experience), calibrate (expectations), reframe, leverage, curate. These are brochure-and-TED-talk words, not the words of someone telling you what a week at sea was actually like.
Replace with plain verbs the reporter would use: "optimize your sea days" → "plan your sea days," "unpack the dining options" → "walk through the dining options," "curate your excursions" → "pick your excursions." Carve-out: literal uses are fine ("the app optimizes the route"). The tell is the abstract self-help use where a plain verb works.
When a page gestures at "who this cruise is for," AI reaches for a four-part traveler stack: "Some guests want to relax by the pool. Some want adventure ashore. Some are traveling with kids. Some are celebrating an anniversary." The four-part shape is the tell — exhaustive market-segment coverage, not observation. (Distinct from a real grouping under Rule of Three: there the count follows the fact; here four parallel "some guests…" clauses are a template.)
Fix: name the specific reader this page actually serves, or collapse to one or two concrete clauses. "the first-time cruiser who's nervous about getting seasick" beats four parallel "some guests want…" clauses.
"Having sailed this route myself…" "I've done this excursion." "Trust me, I've been there." When a first-person claim is followed by generic description with no specifics — no date, no weather, no tender time, no dish you actually ate — the claim is unsubstantiated and the reader feels it. (Pairs with Low-Probability Details: specificity is the proof of having been there.)
Two honest fixes: drop the claim and report it as research ("the route runs…"), or earn it with one concrete detail only a passenger would know ("the 7:15 tender from Cabo was already full by 7:00"). The half-measure — claim plus vague gesture — is worst: it asks for testimony credit the prose can't back. This site's whole credibility rests on sounding like someone who actually sailed it; an empty "I've been there" spends that credibility for nothing.
"Picture this." "Let me tell you." "Here's the thing about Cozumel." Sparingly, at a real gear-shift, fine. Used to manufacture intimacy before an ordinary sentence, cut. Diagnostic: does the sentence after the cue need it, or land harder alone?
A sentence that narrates what the next sentence is about to do, instead of just doing it. Always a throat-clear. Almost always deletable.
Forbidden openings (flag all of these at the start of a sentence or paragraph):
The positive rule:
Gear-shift markers are welcome. Announcements are not. The difference is length and content:
Gear-shifts turn. Announcements narrate. Keep the turns. Cut the narration.
The fix is almost always deletion. If you wrote "Let me walk you through what the tender process looks like. The tender leaves the ship every twenty minutes..." cut the first sentence and start with the second. The second sentence is the walk-through. It does not need to be announced.
AI imitates the cruise voice by reaching for clever phrases — metaphors, surprise words, fingerprint-level imagery — when a plain word would serve better. The voice profile permits one or two surprise words per page. The failure mode is producing three to five per paragraph.
Diagnostic: count images per paragraph.
Test for distinct work: if two images in the same paragraph name the same thing (both describe the buffet, both describe the harbor, both describe the experience of arrival), keep the more concrete one and cut the more abstract one. The concrete image is what a person who has been on the ship writes. The abstract image is what a machine writes to sound like one.
Test for cleverness vs. truth: say the plain version aloud. Then say the surprising version aloud. If the plain version lands, keep it. If the surprising version earns its keep by doing work the plain version cannot, keep the surprise. When in doubt, plain wins.
One metaphor used more than twice across an entire piece becomes invisible — and then becomes a tell. Use a metaphor once at its pressure point and move on. If "home away from home" carries the introduction, do not bring it back in the conclusion. If the ship is described as a "floating city" early on, that's its single use; do not return five paragraphs later to call it a "city at sea."
Let the image do its work and step aside.
The same observation restated across four paragraphs without adding evidence or texture. AI does this to fill space. A human reporter says it once, adds the specific that earns it, and moves on. If a section feels like it's circling, it is. Cut the circles.
Example: if you've already said "the buffet at lunch was mediocre," do not also say "the lunch buffet underwhelmed," "the noon meal disappointed," or "the midday spread was forgettable" in the same section. Pick the sharpest version and cut the rest.
AI repeats the same sentence structure across consecutive sentences: Subject-Verb-Object, Subject-Verb-Object, Subject-Verb-Object. Vary the machinery, not just the words. Follow an S-V-O with a fragment, a question, a sentence that starts with a dependent clause, or an inversion.
Rule: if three consecutive sentences have the same grammatical shape, break one.
Caught by orchestra audit of the Anthem voyage pack: the same header pattern applied identically across all parallel sections. "Honest read on [port]" appearing on every port day; "For first-timers:" as a sub-section heading on every port day. The phrase itself can be corpus-native, but its mechanical repetition across all sibling sections is the AI-shape tell. A human writer varies the framing by what each section actually contains.
Rule: When writing parallel sections (port-day sections, day-by-day entries, ship-feature breakdowns), let each section's header be shaped by the content it covers, not by a shared template. "Sitka — the small Russian-Tlingit port" + "Skagway — the gold-rush town" + "Juneau — the unreachable capital" + "Victoria — the foreign-port stop" reads written. "Honest read on Sitka" + "Honest read on Skagway" + "Honest read on Juneau" + "Honest read on Victoria" reads templated. Three or more identical-shape headers at the same level is the during-writing equivalent of running an LLM template across siblings.
Caught by orchestra audit and by reader feedback (Erin Upshur Jones, Anthem pack, 2026-06-04): when writing about a specific ship, do not let venue names, brand partners, or class-generic features from sibling lines or sibling classes drift into the prose. Each ship/line has its own naming:
If you find yourself reaching for a venue name and you're not certain it belongs on this specific ship, look it up in the deck plan before writing. Cross-ship name-drift is the AI-shape signal of training-data contamination — the kind of error a reader who has been on the ship catches immediately.
Royal Caribbean's port marketing uses constructions like "Sitka truly offers the best of Alaska," "Skagway symbolizes the spirit of Alaska," "a perfect snapshot of Alaska's wild beauty." Your pack can avoid those exact phrases and still mirror the cadence — "This is the day Alaska reveals itself" is corpus-marketing rhythm even without overlapping vocabulary.
Test while writing: if a sentence feels like it could appear on the cruise line's own marketing page (or in a third-party port guide that gets its language from the cruise line), rewrite. Replace with a specific, friction-bearing observation. Marketing voice telegraphs research-synthesis even when the words are different.
Everything above guards against general AI-slop and the Claude/GPT tells. A draft from a local model (Qwen, Gemma) carries its own accent. While writing or editing such a draft, also strip:
Full grep-able detection lives in voice-audit; these are the during-writing reflexes to catch first.
The cruise voice has a specific rhythm: short declarative sentences that stack, then a longer sentence or a single phrase that lands.
Example:
The tender was rough. Twenty minutes, swells. We landed wet but laughing, and the dock crew already had towels stacked at the rail.
Example:
She cost more. She was worth it.
Every dollar.
This is not a rhythm trick. It's the way a person reports compression then release. When this pattern appears naturally, leave it alone. When it's being manufactured to sound like the author, cut it. The difference is felt, not measured.
The cruise voice uses:
If a paragraph has more than two em-dashes, compress. Find the clause that's doing the interrupting and decide: does it earn its dash, or is it hiding in one?
AI never starts a sentence with "And." AI never uses a sentence fragment on purpose. AI never writes a paragraph that's just one word.
Human writers do all of these.
These are not tricks. They are the natural movements of a mind reporting in real time, not executing a template.
For accessibility, solo, and grief content the standard is heightened:
All machine tells share one source: writing for safety instead of report. The diagnostic question is: Am I including this because the page needs it, or because I'm not sure the next sentence will land?
When in doubt, trust the specific. Cut the crutch. The reader feels the difference.
The PDFs are the cron job. The architecture is the mechanism. The voice is what the reader actually feels.
voice-audit. This work shipped on main as v3.2.0 (2026-06-02) but was merged into the branch alongside the parallel v3.2.0 work (2026-06-04) and renumbered to v3.1.5 for chronological clarity.like-a-human (in cruise voice): copula avoidance with replacement table, decorative adverb kill list, rule of three, low-probability details, dead metaphor, one-point dilution, syntactic template repetition, punctuation fingerprint, the controlled flaw, the building pattern with concrete examples, cadence escalation (two kinds of acceleration + pause-and-pivot). Reorganized Plain Language Discipline as the spine of the document.