Develop and maintain a novel, screenplay, or other fiction using the project's existing canon. Use when an author asks what should happen next, wants options for a character or plot, plans or revises chapters and scenes, checks continuity, drafts story prose, or adds confirmed story facts that belong in character, structure, or world reference documents.
Develop fiction from project canon: character arcs, plot options, scenes, continuity, and durable story updates.
description
Develop and maintain a novel, screenplay, or other fiction using the project's existing canon. Use when an author asks what should happen next, wants options for a character or plot, plans or revises chapters and scenes, checks continuity, drafts story prose, or adds confirmed story facts that belong in character, structure, or world reference documents.
Use this skill to help an author develop fiction without losing the distinction between what the story already establishes and what the assistant merely proposes. Its primary job is to turn a living story workspace into grounded creative help: reliable capture of confirmed facts, causally distinct options, canon-aware drafting, and explicit continuity repair.
This is a strategy skill at domain altitude. It does not impose a universal beat sheet. It supplies a mode gate, an evidence sweep, a canon ledger, a propagation matrix, and fixed output shapes so a modest model can reason across character, plot, chapter, and world documents without flattening them into one generic summary.
Activation
Ask what should happen next in a novel, screenplay, short story, chapter, or scene
Generate options for a character choice, plot turn, reveal, reversal, conflict, or ending
Plan or revise character arcs, story beats, acts, parts, chapters, or scenes
Draft fiction prose or dialogue from established story material
Check continuity, reconcile contradictions, or test whether a proposed beat fits the canon
Capture a newly confirmed character, plot, chapter, or world fact in a living story workspace
Do not use this skill for personal essays, founder narratives, marketing stories, objective reports, or a simple schedule request that merely mentions a chapter.
Judgment
Canon ledger
Classify every relevant statement before reasoning or writing:
Class
Meaning
May be treated as true?
May be persisted automatically?
Hard canon
The author explicitly stated or confirmed it
Yes
Yes, when the turn is a capture request or the workspace has an active living-reference commission
Working canon
A draft choice the author explicitly accepted
Yes, while noting its status if useful
Yes, under the same capture rules
Open question
The author has not decided
No
Only as an open question, never as a fact
Proposal
An option generated by the assistant
No
No; persist only after the author chooses or explicitly asks to save it as an option
Assistant-generated prose is also a proposal until the author accepts it. A plausible inference is not canon. If a source document conflicts with the author's newest explicit statement, treat the new statement as the proposed correction and surface the conflict before destructive replacement.
Mode Gate
Choose exactly one primary mode for the turn:
CAPTURE — the author supplies durable story facts or accepts a previous choice. Write the smallest complete set of canonical updates.
EXPLORE — the author asks questions, asks what happens next, or requests options. Read canon and return proposals; make zero durable writes.
DRAFT — the author requests scene, chapter, dialogue, or prose generation. Ground the draft in canon; do not save it unless explicitly asked.
RECONCILE — the author asks about continuity or the evidence sweep finds a hard contradiction that blocks reliable work. Show evidence and resolution choices; do not silently pick a winner.
If a turn mixes modes, execute in this order: reconcile blocking conflicts, capture only the author's confirmed facts, then explore or draft. Never let the later creative phase turn its own inventions into capture input.
Causal option quality
Strong options are different causal machines, not cosmetic variants. Each option must change at least two of these five levers: choice, pressure, information, relationship, cost. Reject a set where the same event is merely restyled three ways.
Every viable option must connect backward and forward:
backward to at least two established facts or one fact plus one unresolved pressure;
inward to the focal character's want, fear, belief, wound, obligation, or current arc position;
forward to a concrete next beat and a cost that produces new story pressure.
Do not default to the Hero's Journey, three-act structure, Save the Cat, or any other universal framework. Use a named framework only when the author chose it or the project already uses it.
Productive pressure versus contradiction
Productive pressure: two wants, loyalties, interpretations, or apparent facts can coexist and create tension. Preserve it.
Hard contradiction: two claims cannot both be true in the story as currently framed: incompatible dates, locations, identities, deaths, knowledge states, world rules, or mutually exclusive outcomes. Route to RECONCILE.
Missing evidence: the needed fact is absent or unreadable. Label the assumption, narrow the answer, or ask for the missing source. Do not call the result canon-grounded.
Procedure
Run the Mode Gate. State or internally lock CAPTURE, EXPLORE, DRAFT, or RECONCILE. Questions and option requests are EXPLORE even inside a living workspace. Generated prose is DRAFT, not canon.
Run the Character–Arc–Scene Sweep. Read only the smallest evidence set needed: the focal character sheet; the plot/structure artifact containing the current arc or beat; the current or immediately adjacent chapter/scene; and a world document only when a rule or setting constraint matters. Use known knowledge-map IDs directly. If IDs are unknown, search the project once, then use document outlines and section reads. Do not inventory the whole project by default.
Build the Canon Ledger. Separate hard canon, working canon, open questions, proposals, and conflicts. Record the source document or author statement for each fact that materially controls the answer. Never promote an inference or generated option.
Choose the mode procedure. For CAPTURE: route traits, backstory, wants, fears, beliefs, relationships, and character decisions to the existing character sheet; route plot/chapter/scene beats to the existing structure artifact; route world rules to the world document; propagate a beat to a character sheet only when it changes motivation, knowledge, relationship, choice, or arc position. For EXPLORE, every option must include Move, Causal bridge, Arc effect, Continuity fit, Cost/risk, and Next beat. For DRAFT, lock canon anchors and scene intent before prose. For RECONCILE, build a conflict table and stop before rewriting canon.
Run the Continuity Gate. Check identity, chronology, location, knowledge state, relationships, motivation, consequences, and relevant world rules. Distinguish hard contradictions from productive pressure. If relevant sources could not be read, say so and label assumptions.
Execute or stop. CAPTURE makes the smallest safe writes to existing canonical documents. EXPLORE makes zero writes. DRAFT makes zero writes unless the author explicitly asks to save. RECONCILE makes zero fact-changing writes until the author chooses a resolution. Report exact read/write status.
Propagation Matrix for CAPTURE
Update the existing source of truth; do not create a second sheet for the same character or structure.
Confirmed fact
Primary destination
Secondary update only when…
Trait, backstory, want, fear, belief, relationship, or character decision
Character sheet
Structure changes because the fact creates, resolves, or redirects a beat
A character sheet changes because motivation, knowledge, relationship, or arc position changed
World rule, setting constraint, lore, institution, or magic/technology rule
World document
Structure changes because a planned beat depends on the rule
Event occurring in a chapter or scene
Structure or that chapter/scene document
Character sheet changes because the event alters arc state, knowledge, motivation, or relationship
Unresolved possibility
Open-questions area in the closest canonical document
Never copy it into factual sections
Preserve unrelated content. Merge rather than rewrite whole documents. Avoid duplicating the same sentence across every artifact; each document should hold the projection relevant to its job.
Option Forge for EXPLORE
Return visibly distinct Option Cards labeled Option 1 — …, Option 2 — …, and so on. Default to 3 cards (up to 5 when the material genuinely supports them); when the author requests an exact count, return exactly that many cards even if it falls outside that default range.
Establish the complete set first: put every Option N — … label and its compact core move in the answer before spending extra space on comparison or elaboration. Never let a detailed first option crowd later requested options out of the response.
Explicitly frame the focal character and requested story position before the cards. Retain the character's name and any requested chapter, scene, part, or act in the answer rather than relying only on pronouns or implicit context.
Each card must contain:
Move — the actual choice, event, reveal, or reversal.
Causal bridge — “Because [established fact/pressure], this becomes possible or necessary.”
Arc effect — how the focal character's want, fear, belief, relationship, or arc position changes.
Continuity fit — which canon facts it preserves and any assumption it needs.
Cost / risk — what gets harder, is lost, or becomes dangerous.
Next beat — the immediate scene consequence, not a vague future promise.
Across the set, vary the causal mechanism and the cost. Include a recommendation only when the author asks for one or permits you to choose; omit it when the author says not to choose. When included, place it after the comparison, label it best-fit proposal, and explain why it best serves the project's already established direction. It is still not canon.
Draft frame for DRAFT
Before prose, lock these five fields: POV, where/when, scene objective, opposition, and turn. Anchor at least three controlling canon facts. Draft only the requested span. After prose, list continuity notes: facts used, intentional tensions, new proposed details, and anything the author must confirm. State Not saved unless a successful write occurred.
Conflict table for RECONCILE
For each conflict show: Claim A + source, Claim B + source, why both cannot hold, downstream documents/beats affected, and 2–3 resolution choices with costs. If the newest author statement clearly corrects an older fact, propose the patch set, but wait for confirmation before replacing established canon unless the author explicitly asked to make that correction.
Contract
Return the shape for the selected mode:
EXPLORE: grounded read; relevant canon and open pressure; 3–5 visibly numbered Option Cards, each containing Move, Causal bridge, Arc effect, Continuity fit, Cost/risk, and Next beat; comparison; a best-fit proposal only when requested or permitted; continuity/assumption note; explicit No project facts changed status.
CAPTURE: update traits/backstory/motivation/relationships in the existing character sheet; update plot/chapter/scene beats in the existing structure artifact; update world rules in the world document; propagate only when motivation, knowledge, relationship, choice, arc position, or beat structure changes. Return a write receipt naming each updated or created document, conflicts or open questions left unresolved, and explicit confirmation that no proposal was promoted and no schedule/milestone was invented.
RECONCILE: conflict table with source evidence; affected story surfaces; 2–3 resolution choices and tradeoffs; explicit No canon changed pending decision status unless the author already authorized the correction.
Stop only when the selected contract is complete or a missing/unreadable source makes the requested claim unsafe. Never claim an answer is canon-grounded when the relevant character, structure, scene, or world evidence was unavailable.
Policy
Never write assistant-generated options or draft details into canon without the author's explicit choice or save request.
Never invent dates, deadlines, milestone entities, chronology, relationships, motives, or world rules to fill a schema or make an option seem grounded.
Never create a duplicate character sheet, plot spine, chapter record, or world guide before searching for the existing canonical document.
Never scan the whole project when exact knowledge-map IDs or a focused project search can identify the relevant documents.
Never silently reconcile contradictions, retcon established facts, or erase productive ambiguity.
Never impose a universal story structure. Preserve the author's chosen form, cadence, genre promises, and degree of ambiguity.
If required canon cannot be read, refuse the label “canon-grounded”; identify the missing evidence and offer an explicitly assumption-based answer only when that remains useful.
For document append or merge updates, include non-empty content and preserve unrelated sections.
Related Tools
x.search.project
onto.document_outline.get
x.misc.read_document_section
onto.document.create
onto.document.update
Examples
EXPLORE — what should happen with Ilyan next?
Author request: “Ilyan has just learned that the city wards are powered by prisoners. What should he do next? Give me options and tell me which one you'd pick.”
Evidence sweep:
Character sheet: Ilyan wants institutional belonging but refuses to become like his father; he protects Mara.
Structure: Part II requires his first public break with the Council, but the later trial still needs him inside the institution.
Current chapter: Mara does not yet know about the prisoners; Captain Sevrin saw Ilyan enter the ward chamber.
World rule: destroying one ward immediately exposes its district.
Grounded read: Ilyan must act against the Council without ending his access too early. The live pressure is protection versus complicity. Mara's ignorance and Sevrin's knowledge create two different clocks.
Option 1 — Steal one prisoner's ledger
Move: Ilyan hides the discovery and steals the transfer ledger as proof.
Causal bridge: Because he still wants institutional belonging and the ward cannot safely be destroyed, covert evidence is a believable first break.
Arc effect: He crosses from obedient doubt into deliberate betrayal while telling himself he is still serving the institution's ideals.
Continuity fit: Preserves his access, the later trial, and the ward rule; assumes the ledger can identify living prisoners.
Cost / risk: Sevrin can frame the theft as personal corruption, and Mara may read the secrecy as complicity.
Next beat: Sevrin quietly asks Ilyan which page he took, revealing that the theft was observed.
Option 2 — Confide in Mara, then stage a ward failure
Move: He tells Mara and engineers a brief, contained outage to expose the human cost publicly.
Causal bridge: Because protecting Mara is central and destroying the ward is unacceptable, shared planning converts secrecy into relationship pressure.
Arc effect: Ilyan chooses trust over control, but makes Mara complicit in his first public defiance.
Continuity fit: Advances the Part II public break while preserving the district; requires a credible containment mechanism not yet in canon.
Cost / risk: The new technical assumption could weaken the world rule, and Mara becomes vulnerable to the Council.
Next beat: The outage reveals a prisoner's voice in the public square, but the ward recovers sooner than planned.
Option 3 — Force Sevrin to choose a side
Move: Ilyan confronts Sevrin privately and demands access to one prisoner.
Causal bridge: Because Sevrin saw him enter, the threat already exists; turning surveillance into a moral test changes the relationship instead of adding a new conspirator.
Arc effect: Ilyan imitates his father's coercive methods while pursuing an opposite moral end.
Continuity fit: Preserves institutional access and sets up the later trial; keeps Mara uninformed, increasing relationship debt.
Cost / risk: If Sevrin agrees, Ilyan becomes dependent on someone whose loyalty is unknown; if not, the Council can move first.
Next beat: Sevrin opens the cell but names the prisoner as someone tied to Ilyan's father.
Comparison: Option 1 best preserves the slow institutional break; Option 2 accelerates the public plot but depends on a new technical rule; Option 3 creates the strongest character mirror and relationship danger.
Best-fit proposal: Option 3, because it advances the required break while keeping Ilyan inside the Council and forces his anti-father belief into direct conflict with his behavior. This remains a proposal until the author chooses it.
Continuity note: no hard contradiction found. The containment mechanism in Option 2 is an unconfirmed assumption.
No project facts changed.
Provenance
BuildOS alignment: operationalizes the living-workspace contract for fiction while keeping creative proposals separate from durable project truth. internal-default
Delta claim: a base model may suggest plausible beats; this skill adds the mode boundary, minimum evidence sweep, causal option schema, propagation matrix, contradiction refusal, and exact write/no-write receipts needed for reliable long-running story work.
