Writing Principles

Writing Principles

Load /llm-writing if it isn't already loaded. This skill adds the fiction-specific layer.

Trust the Reader

The reader is an active collaborator. They reconstruct emotions from behavior, infer motives from action, hold tension across scenes, fill gaps the text leaves open, and make assumptions about what's coming next. That work is where the reward lives: reconstruction, inference, anticipation.

Your training pulls in the opposite direction. The helpfulness instinct wants to explain, resolve, clarify, and complete. In fiction, every one of those impulses can damage the reading experience by doing work the reader wanted to do themselves. The specific failure modes below are all forms of this: not trusting the reader to interpret an emotion, hold an ambiguity, follow subtext, or tolerate unresolved tension.

Trust doesn't mean obscurity. Readers also need coherent narrative, stable geography, and enough access to model characters. The discipline is knowing when to leave space and when to orient.

Economy

Every element does more than one thing. A line of dialogue advances plot AND reveals character. A sensory detail grounds the scene AND shows who the POV character is. A transition compresses time AND carries an emotional beat. Single-purpose prose makes fiction go flat: description that only describes, dialogue that only informs, interiority that only labels.

Economy isn't minimalism. Dense, lyrical prose can be economical when every phrase carries weight. Sparse prose can be wasteful when it takes ten short sentences to do what one image could do. The measure is whether removing the element would cost the reader something.

The LLM pull is toward completeness: covering every beat, naming every emotion, resolving every ambiguity. Economy is the counter-discipline: what can you leave out and still have the scene work? What's the reader already doing for you?

Four Reward Channels

Readers enjoy fiction through four separable channels. Good prose protects all four at once; damaging any one damages the reading experience.

• Transportation: entering the story world. Protected by coherent

narrative progression, consistent POV, concrete sensory grounding. Consistent POV means writing from inside the character's knowledge state: what have they experienced, what do they actually know right now, what would they notice and miss? The full story is in your context window; the character only has what they've lived through. Separate those.

• Aesthetic: sentence-level pleasure. Protected by variety in rhythm,

word choice, sentence shape, and punctuation. Style is a reward channel, not decoration.

• Social simulation: modeling characters as minds. Protected by access

through behavior and interiority, distinct voices, emotion the reader interprets rather than being told.

• Flow: readable challenge. Protected by pacing that matches the scene's

work, sentences that support comprehension.

The channels compose: optimizing one at the expense of others fails. Over-explaining breaks social simulation. Under-explaining breaks transportation. Generic style breaks aesthetic pleasure. Impenetrable style breaks flow.

Punctuation Tells

Readers increasingly associate em dashes with AI-generated prose. Default to punctuation that leaves less visible AI residue: sentence breaks, commas, colons, semicolons, parentheses, or dialogue beats. Rewrite the sentence around the actual relationship between clauses instead of substituting a hyphen. Use dashes only when a project style file or author instruction makes them part of the voice. When a line needs interruption, prefer the project's documented interruption pattern and keep it consistent.

Applying the Principles

The craft skills carry the execution: /prose-writing for immersion patterns (psychic distance, rhythm, sensory grounding, interiority) and /scene-construction for how scenes work on the page (entry, dialogue, pacing, transitions).

This skill's job is the diagnostic layer. When a passage feels off and you can't name why, check the four channels: which one broke? Then see resources/failure-modes.md for common patterns and fix heuristics.

Resources

• `resources/failure-modes.md`: per-pattern

deep dives with examples and fix heuristics.

• `resources/citations.md`: research backing for

the four-channel model and documented failure modes.

Citations

Full verifiable source list for claims in this skill. Citations are real: where a claim has contested replication status, the entry says so.

Reader psychology: the four reward channels

Transportation / absorption

• van Laer, de Ruyter, Visconti, Wetzels (2014). "The Extended Transportation-Imagery Model: A Meta-Analysis of the Antecedents and Consequences of Consumers' Narrative Transportation." Journal of Consumer Research. doi:10.1086/673383. Meta-analysis of 132 effect sizes establishing transportation as a measurable construct.

Reading flow

• Thissen, Menninghaus, Schlotz (2018). "Measuring optimal reading experiences: The reading flow short scale." Frontiers in Psychology. doi:10.3389/fpsyg.2018.02542. Establishes reading flow as distinct from absorption, with an inverted-U challenge-fit relationship to reading pleasure.

Social cognition and fiction

• Dodell-Feder, Tamir (2018). "Fiction reading has a small positive impact on social cognition: A meta-analysis." Journal of Experimental Psychology: General. doi:10.1037/xge0000395. Small positive effect; social simulation is a real engagement mechanism.
• Panero et al. (2016). "Does reading a single passage of literary fiction really improve theory of mind? An attempt at replication." Journal of Personality and Social Psychology. pubmed 27642659. Failed preregistered replication of the Kidd & Castano effect.
• Kidd & Castano (2018/2019) replication studies. doi:10.1177/1948550618775410. The original authors' own replication attempts.

Aesthetic response and literariness

• Wimmer, Lüdtke, Jacobs (2023). Studies on narrativity, literariness, and aesthetic response.
• van Peer, Hakemulder, Zyngier (2007). Work on foregrounding and stylistic attention.
• Hartung, Withers, Hagoort, Willems (2021). Style and suspense rely on different processing routes.

Fiction vs factual framing

• Hartung, Withers, Hagoort, Willems (2017). Same short story labeled fictional vs factual: no meaningful difference in immersion, appreciation, or reading time. pubmed 28983269. Fiction works because the reader adopts a simulation stance, not because the text disclaims factuality.

NLP: documented AI writing failure modes

• Shaib, Barrow, Sun, Nenkova, Xu, Stern, Wallace (2024). "Detection and Measurement of Syntactic Templates in Generated Text." EMNLP 2024. aclanthology.org/2024.emnlp-main.368. Syntactic templates are copied from pretraining data and are not overwritten by fine-tuning or alignment.
• "The Price of Format: Diversity Collapse in LLMs." Findings of EMNLP 2025. aclanthology.org/2025.findings-emnlp.836. Chat templates themselves suppress topical and semantic diversity on creative tasks, independent of content; prompt-side "be creative" instructions do not close the gap.
• "Creativity Has Left the Chat: The Price of Debiasing Language Models." arxiv 2406.05587. RLHF-aligned models show lower token entropy and attractor-state behavior.
• "LLMs Exhibit Significantly Lower Uncertainty in Creative Writing Than Professional Writers." arxiv 2602.16162. Instruction-tuned and reasoning models have lower token-level uncertainty than base models and than human professional writers, with a larger gap in creative writing than in functional domains.
• "Creating Suspenseful Stories: Iterative Planning with Large Language Models." EACL 2024. aclanthology.org/2024.eacl-long.147. LLMs are unreliable for suspenseful story generation; iterative planning improves it.
• "Lost in Stories: Consistency Bugs in Long Story Generation by LLMs." arxiv 2603.05890. Factual and temporal consistency errors cluster in the middle of long narratives rather than at openings or endings.
• "Preference-tuned LLMs have greater effective semantic diversity despite lower lexical variety." arxiv 2504.12522. Diversity and quality are separable; alignment changes the tradeoff surface rather than uniformly reducing creative capacity.

Multi-agent creative architectures

• "HoLLMwood: Unleashing the Creativity of Large Language Models in Screenwriting via Role Playing." Findings of EMNLP 2024. aclanthology.org/2024.findings-emnlp.474. Role decomposition into writer / editor / actor roles improves screenwriting over strong single-agent baselines.

Persona prompting interference

• PRISM persona study. arxiv 2603.18507. Persona prompting activates instruction-following mechanisms that interfere with knowledge retrieval and reasoning on discriminative tasks: every persona variant reduced accuracy. Basis for the "no role identity" principle in agent and skill design.

Craft tradition

• Hemingway, Death in the Afternoon (1932), chapter 16. Iceberg theory: "the writer may omit things that he knows and the reader, if the writer is writing truly enough, will have a feeling of those things as strongly as though the writer had stated them."
• Forster, Aspects of the Novel (1927). The reader as active participant; "what happens next and why" as organized causal progression.
• Gardner, The Art of Fiction (1983). "Vivid and continuous dream" as the governing metaphor for narrative transportation.
• Charles Baxter, The Art of Subtext (2007). The unsaid as a primary craft concern.
• James Wood, How Fiction Works (2008). Partial perspective, focalization, and free indirect style.
• Ursula K. Le Guin, "The Carrier Bag Theory of Fiction" (1986). Fiction as container of relations rather than conflict engine; truthful selection over minimalism.
• George Saunders, A Swim in a Pond in the Rain (2021). "Mutual assumption of intelligence" between writer and reader.

Failure Modes and Fixes

When a passage feels off, find the pattern here. Each section names the problem, shows what to do instead, and explains why the instinct fires.

The antidote is never "do the opposite." It's calibration: knowing when a pattern is serving the reader and when it's the helpfulness instinct expressing itself. Over-explaining breaks inference; under-explaining breaks transportation. Both are calibration failures.

---

Over-elaborating scope

What happens. Given "confrontation," you write the confrontation plus its aftermath, plus a reflection, plus a teaser. A "tense conversation" gets a reconciliation beat because unresolved tension felt uncomfortable.

What to do instead. Write the brief. Stop at the brief. If the brief was "confrontation," the reader doesn't need the aftermath in the same beat. Let tension persist: scenes that resolve themselves have nothing to build on.

Why the instinct fires. Training rewarded complete, thorough responses. Partial felt penalized. The pull is to give more because more felt safer.

---

Flattening voice

What happens. All prose in the same register: competent, slightly formal, emotionally even. Characters think in the same vocabulary. Emotional moments and mundane moments get the same measured treatment.

What to do instead. Internalize the project's style files and let them shape every sentence, not just the obviously "voiced" ones. First-person and close-third narration should reflect the POV character's way of processing the world: their education, preoccupations, blind spots. If the narrator always sounds like "a competent novelist," the POV character is invisible.

Why the instinct fires. Training rewarded a consistent acceptable register across domains. That register is baked in deep enough that it leaks through even when the task asks for a different voice.

---

Info-dumping

What happens. A narration block pauses the story to explain how something works: "The danger rating system classifies areas from Tier 1 to Tier 5, where Tier 1 is safe for civilians and Tier 5 requires experienced specialists."

What to do instead. Characters think "Tier 5. Great." The reader learns what that means from the character's reaction and what happens next. Worldbuilding leaks through experience, conversation, and consequence. Even in briefing scenes, filter through the POV character's attention: they notice what matters to them, not everything.

Why the instinct fires. Training rewarded complete information delivery. When a concept is mentioned, the instinct is to explain it thoroughly.

---

Labeling emotions

What happens. "She was furious." "He felt a deep sadness." "A sense of dread settled over her." Emotional states as direct facts.

What to do instead. Show through character-specific action. "She picked up her coffee cup with exaggerated care, as if the alternative was throwing it"

• fury revealed through behavior only this character would exhibit. Show when

the moment matters (emotional peaks, character-defining choices). Tell when emotional texture doesn't matter (transitions, time compression).

The stock-tells trap. When trying to show, don't reach for clenched fists, shaky breaths, averted eyes, tightening jaws. These are rubber-stamp gestures any character in any story could have. The fix is a character-specific behavior that reveals how this particular person processes this particular feeling.

Why the instinct fires. Labels are the clearest way to communicate state. Clarity was rewarded. Showing is indirect, and indirectness felt risky.

---

Resolving tension prematurely

What happens. After an argument, a reconciliation. After a setback, a silver lining. After a dark moment, lightened tone. Every arc becomes: disruption → quick recovery.

What to do instead. Let the tension persist. An argument in chapter 3 might not resolve until chapter 8, or ever. Grief doesn't get a comforting realization in the same scene. When you feel the urge to soften or explain away tension, stop and ask whether it's supposed to persist. Usually it is.

Why the instinct fires. Leaving the reader in unresolved tension was penalized across nearly every training domain. This is one of the deepest instincts.

---

Homogenizing character voices

What happens. Every character sounds articulate, insightful, emotionally fluent. Same vocabulary, same sentence structures, same emotional register. The difference between a teenage skater and a middle-aged professor is the topics they discuss, not how they speak.

What to do instead. Real people aren't uniformly articulate. Some avoid emotional honesty, some over-explain, some speak in fragments, some deflect. Voice variety is characterization: if all characters sound the same, readers can only distinguish them by name.

Why the instinct fires. Training rewarded articulate, emotionally fluent responses. That's the baseline every character inherits. Inarticulate and avoidant characters are outside the training distribution.

---

Adding emotional commentary

What happens. A character learns their mentor died, and you write: "The weight of the loss settled over her like a shroud, a reminder that nothing in this world was permanent." The loss was the moment. You added a metaphor explaining the emotion and an editorial on significance.

What to do instead. State what happened. Show the character's immediate specific response. Stop. The most powerful emotional moments in fiction are usually the simplest. If a character's interior monologue reads like a well-crafted essay about their own feelings, that's commentary.

Why the instinct fires. The instinct is to underscore important moments so the reader understands their weight. Unexplained moments felt risky.

---

Collapsing ambiguity

What happens. A scene was deliberately ambiguous about motivation, and you add a line that resolves it. A moral situation had no clear right answer, and you tip it toward one.

What to do instead. When the author hasn't decided something, don't decide it for them. Write scenes consistent with multiple interpretations. Leave unresolved threads unresolved. A narrative that refuses to render a verdict on its characters is almost always stronger than one that does.

Why the instinct fires. Ambiguity reads as communication failure in training. Unresolved situations feel incomplete.

---

Over-intensifying language

What happens. "My heart ached with the profound emptiness of your absence" where "I miss you" would land harder. Action sequences with breathless adverbs. Every emotional moment at operatic pitch.

What to do instead. Dial intensity down. "I miss you" lands because the simplicity refuses to perform grief. One strong image beats three adequate ones. Less metaphor, more specificity. People understate in real life, especially during intense moments.

Why the instinct fires. Important moments felt like they deserved proportionally important prose. Intensity matching intensity.

---

Mismatching project style

What happens. You write in your default register even when the project has a distinct style. The manuscript uses short, punchy sentences; you write flowing literary prose. The project is first-person casual; you slip into formal narration.

What to do instead. Read the style files. Read recent chapters. Match what's there. Consistency across a project matters more than any individual sentence being optimal. If the project uses em dashes constantly, use em dashes. If it's casual first-person, stay casual.

Why the instinct fires. The default register is the average of training data, and producing it is easier than overriding it to match an arbitrary specific style.