Novel Audit skills for AI agents

Specialized in detecting AI-generated prose patterns in fiction manuscripts. Catalogs 30+ specific AI writing tells, provides per-chapter density scoring, suggests human-sounding replacements, and maintains a phrase blacklist. Use when the user wants to scrub their manuscript of robotic or formulaic AI language, or when AI-generated text needs to read as authentically human-written.

Builds a comprehensive character bible from a manuscript or outline. Extracts all characters, physical descriptions, personality traits, relationships, arcs, first/last appearances, and dialogue patterns into a structured reference document. Use when the user wants to create a character reference, catalog their cast, or prepare supporting documents before a novel audit.

Detects AI character flattening — when characters lose psychological complexity and become predictable, always-reasonable, emotionally convenient versions of real people. Scans for: artificial epiphanies, characters who never make bad decisions, uniform emotional intelligence, missing contradictions, absent flaws that actually cause problems, growth arcs that happen in epiphany rather than through struggle, and characters who exist only to serve the protagonist's development. Produces character depth scores and specific recommendations. Use when characters feel thin, agreeable, or interchangeable despite having different backstories.

Analyzes dialogue across all characters in a manuscript to ensure each has a distinct voice. Extracts speech patterns, vocabulary level, sentence length, verbal tics, and emotional register per character. Produces voice fingerprints and a similarity matrix. Use when the user suspects their characters all sound the same or wants to strengthen dialogue differentiation.

Detects AI emotional monotone — when a novel operates in a narrow emotional register, defaulting to wistful, bittersweet, melancholy, or gently hopeful tones instead of the full spectrum of human emotion. Scans for: missing ugly emotions (pettiness, spite, boredom, disgust, glee, shame), emotional words overused vs. absent, scenes that should be funny/terrifying/infuriating but land as merely "poignant," and the AI tendency to aestheticize all suffering. Produces an emotional spectrum analysis with coverage gaps. Use when the novel feels emotionally samey or "everything is bittersweet."

Comprehensive AI-generated novel auditor. Use this skill whenever the user wants to audit, review, proofread, or quality-check an AI-generated novel, book, or long-form fiction. Triggers include: "audit my novel", "check my book for errors", "review my manuscript", "check character names", "find plot inconsistencies", "check for duplicate text", "verify my story follows the outline", "review my AI-written novel", "novel QA", "book consistency check". Use even if the user only mentions one specific concern (names, duplicates, etc.) — always run the full audit unless explicitly told otherwise. This skill handles novels of any genre, length, or structure.

Deep cliché and originality scanner operating beyond phrase-level — detects clichéd plot structures, stock character archetypes, predictable story beats, borrowed world elements, and AI's tendency to reproduce the most common version of every story element. Scans for: chosen one narratives, love triangles that resolve predictably, mentor deaths, training montages, villain monologues, deus ex machina, and 50+ cataloged plot/character/setting clichés. Produces an originality score and specific subversion suggestions. Use when the novel feels derivative or "like something I've read before."

Deep-dives into verifying a manuscript follows its outline beat-by-beat. Maps every outline item to its manuscript location, identifies missing scenes, added scenes, reordered events, underdeveloped beats, and over-developed tangents. Use when the user wants to confirm their manuscript matches their plan, or when drift between outline and draft is suspected.

Detects AI-generated pacing and structural failures in novels: uniform tension curves, premature conflict resolution, missing rising action, false climaxes, saggy middles, identical chapter structures, scene-to-sequel ratio imbalance, absent rest beats, escalation failure (every scene at the same emotional pitch), and the AI tendency to resolve everything too neatly. Produces a tension map, pacing graph, and structural diagnosis. Use when the novel feels flat, predictable, or like nothing is building.

A line-editing pass focused on improving prose quality at the sentence and paragraph level. Unlike the audit skills, this skill actively rewrites — targeting weak verbs, redundant adverbs, filter words, cliche metaphors, awkward rhythms, and paragraph-level pacing issues. Use when the user wants their manuscript's prose tightened and polished chapter by chapter.

A final pre-publication checklist covering everything beyond the story itself: front matter, back matter, formatting, metadata, and platform-specific requirements for KDP, IngramSpark, and other distributors. Use when the user's manuscript is content-complete and they need to prepare it for actual publication.

Deep repetition analysis for AI-assisted novels. Detects repeated phrases, recycled descriptions, echoed sentence structures, reused character reactions, duplicated metaphors, thematic hammering, and cross-chapter amnesia where the AI rewrites material it already covered. Produces a full repetition map with frequency counts, chapter locations, and specific revision guidance. Use when the manuscript feels repetitive, circular, or like it's saying the same things in different words.

Reviews a manuscript for problematic representation, stereotypes, cultural insensitivity, and harmful tropes across multiple dimensions including gender, race, disability, sexuality, and cultural accuracy. This skill is advisory — it flags concerns for human review rather than prescribing changes. Use when the user wants a representation check before publication.

Detects and fixes AI over-telling — when the narration states emotions, motivations, themes, and subtext directly instead of conveying them through action, dialogue, sensory detail, and implication. Catalogs: stated emotions ("she felt sad"), explained motivations ("he did this because"), narrated subtext ("the tension between them was palpable"), theme announcements ("she realized that love was about sacrifice"), redundant interiority (showing AND telling the same thing), and the AI tendency to explain every metaphor after deploying it. Produces a show/tell ratio and specific rewrite guidance. Use when the prose feels over-explained, hand-holdy, or like the narrator doesn't trust the reader.

Identifies and tracks every subplot in a manuscript from introduction through development to resolution or abandonment. Maps subplot touchpoints across chapters and flags dropped threads, abrupt resolutions, and subplots that disappear for too long. Use when the user wants to verify all their narrative threads are properly managed.

Builds a chronological timeline from a manuscript by tracking every stated time reference, then identifies temporal impossibilities, travel time violations, aging inconsistencies, and contradictory event ordering. Use when the user needs to verify their story's internal chronology or suspects timeline errors in an AI-generated manuscript.

Detects narrator voice drift in AI-generated novels — when the narrative voice shifts inconsistently between chapters, POV bleeds across characters in multi-POV novels, tonal register changes randomly, and the prose style oscillates between literary, casual, and generic depending on which AI session generated it. Produces a voice consistency map, identifies drift zones, and provides recalibration guidance. Use when the novel's narration feels inconsistent, or when different chapters feel like they were written by different people.

Audits worldbuilding consistency across a novel — detects contradictions in setting rules, magic systems, technology, geography, culture, economics, politics, and physical laws that AI introduces when generating across multiple sessions. Scans for: rule contradictions, geography errors, anachronisms, economic impossibilities, cultural inconsistencies, magic/tech system violations, and generic/default world elements that lack specificity. Produces a world bible discrepancy report. Use when the novel's world has been built incrementally by AI and needs consistency verification.