Writing Helper (Questions · Genre Coaching · Tightening)

A quiet editing partner from blank page to final polish. Pick a genre skeleton before you write, questions while you're still developing ideas, editor-style genre coaching while revising, or tightening when the prose reads mechanical. It never gives scores. ✍️

← Home: ObsiKO 101 · Up: 16 AI Agent

Where is it? 🧐

• Open 'Writing Helper' (pencil + question mark icon) in the right-sidebar nav — it works on the note you're viewing.
• Press '?' next to the title for a one-glance guide to all four tabs plus comments, history and the AI access guard.
• The tabs at the top switch between four modes, in writing order: 'Structure' / 'Questions' / 'Coach' / 'Tighten'.
• The speech-bubble button in the panel header (next to refresh) toggles %%comments%% in the note between expanded text ↔ folded chips. And every AI question/coach/tighten card has a "To comment" button — it drops that suggestion right below the quoted sentence (or at the cursor) as a %%comment%%: a margin note that never leaves the note on export, undone with one ⌘Z. 💬

① Structure tab — genre skeletons before you write 🏗️

When the blank page feels heavy, it scaffolds your draft instantly, no AI.
- Pick a genre → structure. Web novels alone offer serial episode · chapter-1 opening · regression ch.1 · romance first meeting · battle episode — 89 forms across 13 genres (16 for Character, 13 for Worldbuilding, 11 for Beings).
- Time-tested formal theories are built in, unabridged — the Hero's Journey (Vogler's 12 stages), Save the Cat! 15 beats, 7-Point Structure (Dan Wells), Story Circle (Dan Harmon), the full Romancing the Beat romance arc, four beats (Ki-Seung-Jeon-Gyeol) and the five-stage Freytag arc (shared by web novel & literary), Story Grid & Truby's seven steps for literary fiction, the braided essay, SCQA (Minto's Pyramid), the Amazon 6-pager and Toyota A3 for reports, NABC and the Golden Circle for proposals, StoryBrand SB7 and PASTOR for marketing, the CARS 3 moves and five-move abstract for papers, STAR and PREP for cover letters, and AIDA for landing copy. Plan memos also carry classic catalog questions like "which of the 20 Master Plots is yours?" Hover (or tap) the ⓘ next to the structure description for a detailed explanation and a concrete example of each structure — a Lord of the Rings mapping for the Hero's Journey, a real STAR answer for cover letters, and so on for every one.
- There's also a Character genre 🎭 — sixteen forms: basic profile sheet, character arc (wound→lie→truth), inner life (dimension = contradiction), voice design, values & beliefs, emotional reaction sheet, daily-life details, interview (the character answers in their own voice), villain design, web-novel protagonist, four tabletop RPG sheets (generic, fantasy class-and-level, point-buy generic, gothic horror), relationship design, and a side-character quick sheet. Each step lays down concrete fill-in fields (Name · Age · Job…) with the cursor placed in the first one, so you just fill them in. Once the sheet is written, hit "Roleplay the written character with AI" — it uses the currently open note as the character sheet: the agent panel opens and the conversation starts in character right away (say "end roleplay" to drop the role). To play a randomly rolled character, insert it into the note first with 'Insert at cursor'.
- Tabletop RPG sheet 🎲 — the sheet in the rulebook is meticulous about the numbers and thin about how to actually play them, so this one keeps both on one page. Nine steps: basics · ability scores · health & resources · what they can and can’t do · gear & belongings · origin & why they left · how to play them (creed, bond, flaw, first line) · place in the party · session log (optional). The stat boxes come laid out as Strength · Agility · Endurance · Intellect · Perception · Presence, but those are examples to rename for whatever system you play. You can roll the numbers in Random Box ▸ Character and drop the radar chart in with them.
- If your system is settled, three family sheets — Fantasy RPG sheet (class & level): the six ability scores · AC, hit points, hit dice, initiative, proficiency bonus · passive perception · personality, ideal, bond, flaw · page-2 appearance and backstory (D&D 5e family). Point-buy generic RPG sheet: ST, DX, IQ, HT and the seven values derived from them · advantages, disadvantages, quirks, skills · encumbrance (GURPS family). Gothic horror RPG sheet (vampire): ambition and desire · the nine attributes · disciplines · humanity, stains, hunger, blood potency · predator type · convictions and touchstones · the Masquerade (Vampire: The Masquerade family). All three follow the real sheets’ layout, but none of them is a rulebook — no dice procedures, tables, spell or clan lists; each sheet’s ⓘ states up front that it is unofficial and whose trademarks are whose.
- The random character generator moved to the Random Box panel (first tab, "Character") 🎲 — it used to sit at the top of this structure list, but it's something you reach for while writing or while looking at a canvas, so it now lives in the right sidebar. Pick a setting, gender, age bracket and name style and it rolls a whole character instantly (no AI), now including a birthday with a matching star sign, birthstone and birth flower. Filling in a character sheet here and stuck on a name? Open Random Box, roll one, and pull it in with [Insert at cursor].
- Worldbuilding genre 🗺️ — thirteen forms just for settings: a location sheet, a one-page world bible (only the visible tenth of the iceberg), magic & power systems (Sanderson's Laws — limits beat powers), class & vocation, items & artifacts, gear (weapons & armor), a history timeline, society & taboos, a faction sheet (stated goal vs. real goal), the texture of daily life (one breakfast beats ten lines of exposition), geography & travel (distance is tension), scouting a scene's stage (Chekhov's gun — plant the props early), and naming rules. All are fill-in sheets, just like the character sheets.
- Beings genre 👻 — eleven forms for what is not human. Eight are sheets: entity sheet (basic), nonhuman kind, monster (threat design — fear comes from rules, not power), summoned being (pact & price — what you lose each call), mythic beast (lore — when it appears is its identity), spirit (ghost, elemental, god — the fixation becomes the rule), yokai (rules of folklore — do not look back, do not answer), and a containment file (anomaly report). That last one is written in office-paperwork voice (the layout the SCP Foundation wiki settled into); it carries no setting from that project and is an unofficial form, which its ⓘ says up front. Three old-tale structures come with it — ghost story (rule, violation, an ending that never closes), folktale (Propp's morphology — lack, trials, taboo, and "that is why, today"), and fairy tale (a story the child solves alone). Those three also live under Literary, so either place works.
- Three setting-asset sheets 🗡️ — right after the magic system sit class & vocation / items & artifacts / gear (weapons & armor). Class & vocation is not only for fighters and rogues — wuxia schools and modern or SF specialist roles share the seat, and it asks not for a list of powers but what was given up to gain this. Items start from the premise that the cost and the person who wants it make the story, not the power. Gear puts weapons and armor on one sheet — set the branch in the first box (“weapon or armor”) and the rest of the questions (handling, what it cannot do, upkeep & price, provenance) fit both. What it cannot do makes the scene, not how strong or hard it is — a spear cannot swing in a narrow hall, and no one sheds plate alone. None of the three is a rulebook: they carry the boxes to fill and the questions to ask, nothing more. Need an organization? Use the faction sheet.
- A fourth selector, Name style, picks the cultural flavor of the name — modern offers Korean · Japanese · Chinese · Anglo-American (for foreign side characters), fantasy offers Nordic, Elven and Latin (Lusius, Myla, Laetitia… — a great fit for royal-court and western-flavored fantasy), and sci-fi generates codenames (Echo-7, NX-402). Wuxia has only one style, so the selector is hidden there. The style affects the name only — roles and speech habits stay as they were. The Default option behaves exactly like before, and each setting remembers its own choice.
- Every setting rolls a separate epithet / title / nickname / callsign — wuxia draws from some 1,600 combinable jianghu epithets (Iron Sword, Snow Wolf, Azure Cloud Wanderer…), fantasy from roughly 40,000 titles (Red Hawk, Warden of the Storm, Last Herald of the Night, Twin Warden of the Moon…), modern rolls one of 62 idiomatic nicknames (the Neighborhood Insider, Walking Encyclopedia, Café Fixture…), and sci-fi rolls a crew callsign from 336 (Scrap Appraiser, the Ghost of the Engine Room, Last Compass… — it coexists naturally with code-name style names, designation and callsign being different things). It appears as its own line under the name — labeled to match the setting — rerollable on its own, and never attaches to children.
- Every character also rolls a keepsake, an MBTI type and a zodiac sign — the keepsake is a story-bearing combined phrase rather than a bare object ('a pocket watch, left behind by their mother'; 288 combinations per setting), MBTI comes in all 32 refinements (A/T included, e.g. INTJ-A · cool-headed architect), and the zodiac changes clothes with the setting: the plain twelve signs for modern, hanja palace names (白羊宮) for wuxia, etymological Latin forms (Scorpius) for fantasy, and natal sectors ('Tau Ceti Sector') for sci-fi. Children get all three too.
- Role pools are richer per setting too — fantasy has 78 roles (from wandering knight and village midwife all the way up to emperor, pope and master of the mage tower), wuxia 68 (from lay disciple to leader of the martial alliance and lord of the Demonic Cult), and modern and sci-fi include top-tier figures like conglomerate chairmen and fleet admirals — everything from walk-ons to leads.
- A Sore spot line gives you the character's weakness or trauma as something visible in a scene ("when thunder rolls, they go quiet mid-sentence") — 1,656 for adults, 224 for children. It isn't all flinching: anger triggers (being talked about behind your back, credit stolen, a rule that bends for everyone but you) sit alongside outward reactions, including ones that surface days later over something unrelated. No diagnoses or condition names: a label isn't a story. The cause is deliberately left blank — that belongs in the Character arc form (wound → lie → truth), where you decide it. Children draw from their own age-appropriate pool.
- Fantasy characters also get a surname, built the way each style would do it: terrain house names by default (1,921 of them — Ashford, Frostmarch, Ravenstone), nature compounds for elves (1,283 — Silverleaf, Shadowstep, Moonheart), true patronymics for Nordic (Eriksson ↔ Eriksdóttir, 48 stems), and Roman gens names for Latin (Valerius ↔ Valeria, 48 gentes). The parts were researched from real naming practice — the second element is an actual British place-name generic (-ton, -ley, -by, -thorpe…), so genuine surnames like Ashby, Ashley and Ashton fall out naturally.
- 'Insert suggested skeleton' drops in ① plan notes (this piece in one sentence? who reads it? the feeling at the end? — click a chip to jot answers) and ② step-by-step guides at your cursor. Or tap individual step cards. Undo with one ⌘Z.
- There are two places to put the skeleton — Insert suggested skeleton (at the cursor in the current note) and Make a new note (a fresh .md holding just the skeleton). The first is highlighted for an empty note, the second once you’ve already written something. The new note asks for a name before creating the file, so cancelling leaves nothing behind; afterwards it opens with the cursor in the first field. It works even on screens with no visible document, like the dashboard.
- Guides are all %%comment%% chips — folded in the editor and never included in exports, so there's nothing to clean up.
- Format matches the genre: reports, papers, and blogs get ## headings (switchable to guides-only), fiction and essays get unobtrusive comments only, and poems get numberless "cues".
- A structure is one possible flow, not the answer — use only the steps you need, in any order.

Structure check — instant, no AI ⚡

Below that, the Structure tab also flags structural issues:
- Empty sections, broken [[wikilinks]], orphaned note (no backlinks), no outgoing links

Click an item to jump to that spot. (Always works, no AI key needed.)

② Reflect with questions 🤔

• AI makes questions only when you press the button — never silently while you write.
• No scores, no rewriting — questions only: hidden assumptions, missing counterexamples, vague terms, weak evidence.
• Choose depth (light/normal/deep) and count (1–5); tap the quoted passage under a question to jump to that sentence. 📍

③ Get edit suggestions — through a genre lens ✍️

The revising mode. Pick a genre lens and the AI reads your draft like an editor who has worked in that genre for years.

10 genre lenses 🎭

Web novel · Literary fiction · Essay · Blog/info post · Cover letter · Report · Proposal · Marketing copy · Academic · Poetry
- If the note has writing-kind: in its properties (or uses the web-novel view), that lens gets a (suggested) mark — but it's only a suggestion; switch freely in the dropdown.
- 'Remember this genre for this note' writes it to the note's properties (writing-kind) so it's auto-selected next time (nothing is written until you press it).
- Each lens looks differently — the cover-letter lens never invents experience or numbers, and the poetry lens never applies prose rules (sentence length etc.).

Measurements — instant, no AI 📏

Before running, the app measures your draft and shows numbers and locations only (judgment stays yours):
- long sentences/paragraphs, dialogue ratio, repeated sentence endings, intensifier adverbs, double passives, character count, and more — varying by genre (poetry measures nothing).
- Tap a location chip to jump; more than 3 unfold with 'N more'. Works without an AI key.

Run the coaching 🖋️

• (Optional) Fill in "What reader effect do you want to keep?" — the AI treats it as your intent to verify. Leave it empty if you like.
• Before running you see "Sends about N characters · 1 AI call" (long notes send the first 8,000 characters — it tells you when trimmed).
• Results come as cards:
  1. Editorial thesis — the single biggest reader-experience issue + its cause + what must be kept (compass card, on top)
  2. Whole piece — structural suggestions (direction only)
  3. This part — specific passages as before → after pairs with [Copy]; the after is one example, not "the answer"
  4. Keep this — one thing that already works (green heart card)
• It never pads the count — only well-grounded items, never scores or grades.

④ Tighten — where it reads mechanical 🧹

For text you pasted from an AI, or your own writing that feels stiff. No genre needed.

First it counts things without any AI (tap a chip to jump there 📍): how even your sentence lengths are, repeated sentence endings, sentence-initial connectives, empty hedges, template openers and closers, list-heavy sections, and sentences that all start the same way.

• Repetition map 🔁 — which words and phrases you used how many times and where. Korean particles are folded together (지식 / 지식을 / 지식은 count as one), and repeated two-word phrases are caught too. A small band under each entry shows where in the document they cluster (left is the start, right is the end); tap a spot to jump to that line.
• Thresholds keep it from nagging — words need 4+ times across 3+ lines, phrases 3+ times across 2+ lines. Deliberate repetition within a single line is left alone, short pieces under six sentences aren't measured at all, and function words plus code blocks and frontmatter are excluded.

No scores. There is no "87% AI" reading here — a machine can count patterns, not who wrote it. You get numbers and locations; the judgment stays yours. This is an editing tool for readability, not a way to dodge AI detectors.

Two buttons that fix things without AI 🧰

At the top of the Tighten tab sit [Tidy formatting] and [Fix symbols]. No AI call — they just fix it.
- Tidy formatting — trailing spaces, runs of blank lines, the missing space in ##Heading, the key:value colon in frontmatter, all in one pass (details → 03 Markdown Basics).
- Fix symbols — turns typed marks into typeset ones: three dots (...) become an ellipsis (…), straight quotes (" ') become curly quotes (“ ” ‘ ’), --/--- become dashes (– —), and ->/<- become arrows (→ ←).
- Both show you what and how many places they'll change first, and one Cmd/Ctrl+Z puts it all back. Code, maths, frontmatter, table separator rows, horizontal rules, and quotes inside link URLs and HTML attributes are never touched.

Your style sample (optional) ✍️

With no sample, it only trims filler. Pick one note you wrote earlier and it uses that voice, rhythm and vocabulary as the reference — rewriting toward "generic human prose" would just swap one machine voice for another. It reads only the note you picked, never your whole vault.

Get suggestions · 1 AI call 🖋️

Results use the same cards as genre editing (biggest reason → whole piece → this part → keep this). If there is little to fix, a short list is the correct answer.

Apply here ⬅️

Next to a suggested rewrite, [Apply here] changes just that spot: one place at a time, undone with a single Cmd/Ctrl+Z, and if your text changed in the meantime it touches nothing. The same button works on genre-editing cards.

Just the selection 🖱️

Select a sentence and use right-click ▸ Fix with AI ▸ Polish in my voice (proofread, simplify, shorten and expand work the same way). A suggestion card appears right below the selected text, with changed words highlighted, and the passage changes only when you press [Replace] — press [Keep mine] and the original stays exactly as it was. After replacing, one ⌘Z brings it back. If you edit that passage yourself while waiting, the suggestion is dropped (so it never overwrites your own change). It isn't available inside tables, code or properties, and a suggestion that drops the original's links isn't applied.

Translation uses the same card — select text and press right-click ▸ Translate; language chips (한국어 · English · 日本語 · 中文) sit at the top of the card so you can switch instantly. On iPhone/iPad (iOS 18+), Android and Chrome an on-device translator goes first: no key, no charge, and your text never leaves the device (the first time, it fetches a language file — press [Get language] on the card). Without one, the on-device AI takes over, and only when you press [Translate with AI] on the card is anything sent to an external AI — the card always says whether it was handled on the device or sent to an external service. Links, code and formatting are left untouched, and a result that breaks them is never shown.

Fewer AI tells to begin with 🌱

When ObsiKO's own AI writes into a note it already avoids template openers, empty hedges and one-note rhythm, and it follows the voice of the note it is adding to. Nothing to switch on.

Re-run & fix tracking 🔁

• Re-run with the draft unchanged → it looks for new angles instead of repeating itself.
• Re-run after editing → it re-examines previous notes and only re-raises what still remains.
• Before running, a badge shows "M of the N previous notes look fixed" (the app checks quote survival locally — no AI call).

History log 📜

• The collapsible log at the bottom keeps both questions and coaching runs, newest first.
• Expand an entry to see each item's quoted-passage chip — tap to jump, and passages fixed since then get an 'edited' label.

Good to know ✅

• It never edits your text — suggestions only; you decide what to apply.
• AI questions/coaching need an API key or login. The structure check and measurements always work without one.
• Notes hidden with ai: deny are excluded from AI questions and coaching (structure check and measurements still run). → AI Sources & Privacy

← Previous: Personal Memory · Home · Next: Structuring Knowledge →