APODICTIC: anotherpanacea's Development Editor

A development editor for fiction and argument-shaped nonfiction. Reads what you wrote, diagnoses structure, never rewrites unless you ask — stories or arguments.

flowchart TD
    S(["Type /start"]) --> Q1{"1 · What do you have?"}
    Q1 -->|A manuscript — novel, story, draft| Q2
    Q1 -->|An argument-shaped piece — op-ed, brief, testimony, essay| Q2NF
    Q1 -->|An idea or fragments| Q2
    Q1 -->|A series| Q2
    Q2NF{"2NF · What do you want with this argument?"} --> Q3NF{"3 · Anything that changes how we work?<br/>deadline · sensitive content · editing for someone else"}
    Q3NF --> LNF["Nonfiction Argument Engine<br/>→ Dialectical Clarity letter + marked-up piece"]
    Q2{"2 · What do you want?<br/>(your options depend on what you have)"} --> Q3{"3 · Anything that changes how we work?<br/>deadline · nonfiction · sensitive content · editing for someone else · writing group"}
    Q3 --> L["Your workflow<br/>→ editorial letter or structural plan"]
    L -. the Firewall:<br/>never .-> X["rewrite your prose"]
    LNF -. the Firewall:<br/>never .-> X
    style X stroke-dasharray: 5 5,color:#9ca3af

Every edit starts at /start: three plain-language questions — the second's options depend on the first — route you to the right workflow. Nonfiction & argument work is reachable either as an up-front choice or by flagging it at the third question; both land in the Nonfiction Argument Engine. (Walk the specific routes in the interactive route explorer.)

Why This Exists

Every AI writing tool I tried wanted to rewrite my prose. I didn't need a co-writer. I needed what a good developmental editor does — read my manuscript, figure out what it's trying to be, and tell me where it's working and where it isn't. Structurally. Without touching my sentences.

So I built one. Then I kept building — because the same discipline applies to arguments.

How It Works

For fiction: APODICTIC reads a manuscript, predicts its "contract" (genre, reader promise, controlling idea) from the text alone, then measures the work against that contract. When the inferred contract doesn't match what you intended, that's the diagnostic signal — it means the text isn't communicating what you think it is. It runs 11 analytical passes: reverse outline, reader experience mapping, structural diagnosis, character architecture, reveal economy, pacing analysis, and more — all calibrated to your genre. Genre modules adjust what counts as a problem. A slow opening is a feature in literary fiction, a defect in a thriller.

For argument-shaped nonfiction (op-eds, policy briefs, testimony, essays): APODICTIC infers the argument contract — the claim you're making, the audience you're addressing, the burden you've accepted, the stakes you've invoked — then measures the piece against it. The engine flags missing warrants, scope drift, and unmet strongest objections via Dialectical Clarity and its Red-Team, Persuasion, and Evidence companions. Same Firewall. Same diagnose → letter → marked-up-piece spine. Different front end.

The Firewall: APODICTIC diagnoses problems and names classes of solution — including which structural elements are missing or mis-weighted. Its diagnostic workflows never draft your prose (drafting happens only when you ask for it: /reconstruct, from claims you approved, or a scene-level handoff, which drafts only after you confirm it), and never script the specific content that fills those needs — the events, characters, dialogue, or evidence. Naming that a climax needs a cost paid on the page is diagnosis; writing that scene is yours. You're the writer; it's the analyst.

See It in Action

You don't have to install anything to see what APODICTIC produces. These open as live, rendered pages:

Fiction samples:

• A sample editorial letter → — the main deliverable: a structural diagnosis with severity-ranked findings and classes of solution, prose left untouched. (a second example)
• An annotated manuscript → — the marked-up copy: each finding anchored in the manuscript where the problem lives, severity-tagged, and cross-linked to the letter. This is the working surface you revise at the line — comments only, never tracked prose changes. (Also exports to Obsidian, and to Word / Google Docs as native anchored comments.)
• A targeted audit letter → — a focused single-audit deep dive.
• Pre-writing output → — what you get starting from an idea instead of a draft.

Nonfiction & argument samples:

• A Dialectical Clarity letter → — the argument editorial letter: claim ladder, audience burden, warrant gaps, and the strongest unmet objection, with classes of solution. Never rewrites your argument. (A whole-book development edit of the spine argument in Mark A. R. Kleiman's When Brute Force Fails.)
• Marked-up op-ed (forthcoming) — the argument equivalent of the annotated manuscript: each finding (missing warrant, scope drift, unmet objection) anchored in the piece, severity-tagged, cross-linked to the Dialectical Clarity letter.

Two interactive maps of the tool itself:

• Overview dashboard → — the whole system at a glance: the router, what each pass analyzes, genre modules, audits, and the 50 plot spines.
• Route explorer → — answer the same three questions /start asks and watch where it routes you.

(Tip: these are rendered links. Opening the raw .html files directly in the GitHub file browser shows source code, not the page.)

Beyond Full Edits

APODICTIC isn't just for finished drafts. Both pillars — fiction and argument-shaped nonfiction — share the same diagnostic spine.

Fiction

• Partial manuscript diagnostic — stuck mid-draft? APODICTIC runs on what exists without penalizing missing structure. Adds momentum tracking, stall detection, and a setup inventory showing what your draft has committed to. Synthesis focuses on what's working, what's stalling, and where to go next
• Fragment synthesis — scattered scenes and notes but no continuous narrative? Fragment Synthesis clusters your material, maps connections, and produces a candidate structure showing what your fragments add up to
• Revision Coach (/coach) — post-diagnostic coaching that helps you plan revision sessions without doing the revision for you. Four modes: session planning (leverage-ranked priorities matched to your available time), stuck-point coaching (8-type block diagnosis with structurally-informed writing experiments), momentum tracking (session-over-session progress), and deadline coaching (honest triage with revision calendar). Includes 7 structural prompt families, a stuck-point exercise library, nocebo inoculation, and no-prompt zones for states where more structure would make things worse
• Series Continuity (/audit series-continuity) — cross-volume consequence tracking for multi-book series. Five diagnostic channels: character state, world rules, unresolved threads, hope calibration, and intentional discontinuities. Rolling Series_State.md persists across volumes
• Pre-writing pathway for writers who have an idea but no manuscript — takes you from seed to draftable structure
• Plot coach with 50 structural spines across 12 families (not just three-act)
• 38 available audits (3 universal, 20 craft, 10 genre, 5 tag) including scene function, shelf positioning, emotional craft, prose-pattern and voice calibration, worldbuilding integration, force architecture, reception risk, and intimacy/consent coverage
• Genre calibration across literary fiction, horror, mystery, thriller, SF/F, romance, and hybrids

Nonfiction & Argument

• Nonfiction Argument Engine — for persuasive, argument-shaped nonfiction (policy briefs, op-eds, testimony, essays). Infers the argument contract (claim, audience, burden, stakes) from the text, then diagnoses where the argument breaks down: missing warrants, scope drift, unmet strongest objections. Produces an argument editorial letter and a marked-up piece with anchored findings
• Dialectical Clarity (/audit dialectical) — deep-dive argument structure audit: thesis–antithesis balance, claim ladder, rhetorical fairness, straw-position detection
• Red-Team, Persuasion, and Evidence companions — three focused argument companions that stress-test the argument against its strongest opposition (Red Team), audit audience fit and rhetorical force (Persuasion), and verify the evidence base under the claim ladder (Evidence)
• Adversarial Evidence Review (/audit adversarial-evidence-review) — asks, for each claim that carries weight, whether its link to the evidence would survive a hostile expert who knows the sources
• Argument coaching — revision coaching for arguments, with the same session-planning and stuck-point modes as the fiction coach, calibrated to argument structure
• Citation Verification and Field Reconnaissance research modes — internet-enabled verification of sources and scouting for counterevidence and literature gaps; especially valuable for argument-shaped nonfiction
• Nonfiction pre-draft (/pre-writing) — for writers who have an argument but not yet a draft: captures the thesis, the claim ladder, and the strongest opposing view, and seeds the Argument_State.md the audits later read
• Author decisions and export — /adjudicate records your approve / reject / revise decisions on a reconstructed claim graph, resumably, without drafting anything; an optional AIF export turns Argument_State.md into a standard argument-map graph (claims, warrants, typed objections) for mapping tools, listing every construct it could not map

Forms it knows: op-eds, policy briefs, legislative and agency testimony, academic arguments, open letters, white papers, advocacy pieces, legal briefs, regulatory comments, and expert affidavits, plus lens or exploratory essays when you confirm that register. Narrative nonfiction and memoir route to their own genre audits instead.

Shared across both pillars:

• 6 internet-enabled research modes for citation verification, comp validation, fact-checking, field reconnaissance (counterevidence + literature gaps), genre currency, and representation context
• Legal Risk Register (/legal-risk) — flags possible defamation, privacy, and rights exposure for a lawyer's review. It flags, never adjudicates — not legal advice
• Feedback triage and beta-reader instrument (/triage-feedback, /reader-questions) — sort and prioritize beta-reader/editor feedback, and turn a diagnosis into targeted reader questions
• Project addressability (/projects) — list, resume, and tidy editing projects from saved diagnostic state, with Retcon Planning and State Cards
• Manuscript-structure visualizations plus Diagnostic-Vocabulary and Editor-Scaffolding operator modes

Install

APODICTIC works best on strong frontier models with enough context for large-manuscript analysis.

Which install do I need?

Pick the row for the app you'll actually run APODICTIC in — that's the only thing that decides your path:

You're using…	Go to	Fastest path
Claude (Cowork, claude.ai)	Claude plugin directory	Find APODICTIC in the plugin directory and install it
Antigravity	Antigravity (Native)	Workspace Isolation — download the zip, open the folder, /start
Codex	Codex	Download the zip, open the codex/ folder, install from the local marketplace
ChatGPT	ChatGPT	Open the APODICTIC plugin listing and add it
Claude Code (CLI / terminal)	Claude Code (CLI)	Two /plugin commands
Cowork (desktop app)	Cowork (Desktop App)	Add the marketplace from GitHub, then install

For Claude, Antigravity, Claude Code, and Cowork, start a fresh session and type /start. In Codex, run apodictic-start; in ChatGPT, select or mention the installed APODICTIC plugin.

Claude plugin directory

APODICTIC is listed in the Claude plugin directory. In Claude, open Customize, browse the plugin directory, search for APODICTIC, and install it. Then start a fresh session and type /start. The marketplace routes below still work if you prefer them.

Antigravity (Native)

APODICTIC supports two modes of execution in Antigravity: Workspace Isolation (recommended) and Global Installation.

Method 1: Workspace Isolation (Recommended)

This method keeps APODICTIC isolated to its own project context, safely containing manuscript analysis.

1. Download apodictic-antigravity.zip from the latest release and unzip it. (No clone or build step — the workspace is prebuilt.)
2. Open the unzipped antigravity/ folder as your workspace root.
3. Start a fresh thread and type /start.

Prefer to build from source? Clone the repo, run node scripts/build-antigravity.mjs, and open the generated antigravity/ folder instead.

Method 2: Global Installation

If you want APODICTIC available globally across all of your Antigravity workspaces.

1. Download and unzip apodictic-antigravity.zip from the latest release (or clone the repo and run node scripts/build-antigravity.mjs).

2. Symlink the generated plugin into your global Antigravity data directory:

mkdir -p ~/.gemini/antigravity/plugins
ln -s "$(pwd)/antigravity/plugins/apodictic" ~/.gemini/antigravity/plugins/apodictic

3. Copy the workflow definitions into the .agents/workflows directory of whatever active workspace you want to run APODICTIC from:

cp -r antigravity/.agents/workflows/* /path/to/your/workspace/.agents/workflows/

4. Start a fresh thread and type /start.

Codex

1. Download apodictic-codex-marketplace.zip from the latest release and unzip it. (No clone or build step — the workspace is prebuilt.)
2. Open the unzipped codex/ folder as your Codex workspace root.
3. In Codex, open the Plugins view and install APODICTIC from the local marketplace.
4. Start a fresh thread and run apodictic-start.

If APODICTIC does not appear, the usual cause is opening the wrong folder. Codex must be opened on the generated codex/ directory, not the repo root.

Prefer to build from source? Clone the repo, run node scripts/build-codex.mjs, and open the generated codex/ folder instead.

ChatGPT

APODICTIC is published in ChatGPT's plugin directory: open the APODICTIC plugin listing and add it. Then, in a new ChatGPT conversation, mention @APODICTIC or select it from + > More and ask for a developmental edit or manuscript intake.

Workspace import (admins)

Workspaces that prefer to pin a release can import this repository's marketplace instead. A workspace admin:

1. Opens Workspace settings > Plugins > Add > Import marketplace.
2. Selects GitHub as the source and enters https://github.com/anotherpanacea-eng/apodictic. Leave the path blank so ChatGPT finds the root .claude-plugin/marketplace.json. To install a fixed release, select tag v2.12.0 or a newer release tag when available; otherwise the marketplace follows the default branch and syncs daily.
3. Reviews the import result, opens APODICTIC, and makes it Available or Installed for the intended members or roles.

Members can then install APODICTIC from Plugins if it was made Available. In a new ChatGPT conversation, mention @APODICTIC or select it from + > More when those controls are shown, and ask for a developmental edit or manuscript intake. Plugin availability and controls vary by plan, workspace, and role. APODICTIC's script-dependent checks have inline instructions for hosts without a shell; test a representative edit in the target workspace before relying on full workflow parity.

See OpenAI's plugin installation guide and GitHub marketplace import guide for current controls.

Claude Code & Cowork

APODICTIC also installs into Claude Code and Cowork through the plugin marketplace — pick your host below.

Claude Code (CLI)

/plugin marketplace add anotherpanacea-eng/apodictic
/plugin install apodictic@anotherpanacea-eng-apodictic

Cowork (Desktop App)

Go to Customize > Browse > Personal > + and select Add marketplace from GitHub. Enter anotherpanacea-eng/apodictic, then install the plugin.

Or download apodictic.plugin from the latest release and upload it through Cowork. (Mac users: if your browser auto-unzips the download, re-zip the folder and rename it to apodictic.plugin.)

Updating to a new version

claude plugin marketplace update apodictic

Or /plugin marketplace update apodictic from inside Claude Code (CLI). For Cowork Desktop, use the marketplace's update control in the Customize panel.

Then fully quit and relaunch Claude Code / Cowork — a restart is required to apply the new version. The marketplace cache will refresh on update, but the running process keeps the previous version loaded until relaunched.

---

Your First Five Minutes

1. Install APODICTIC for your host (see above), then start a fresh session.
2. Type /start. It asks three plain-language questions — what you have (an idea, fragments, a partial draft, a complete draft, a series, or an argument-shaped piece like an op-ed or brief), what you want, and anything that should change how it works.
3. Give it your manuscript when it asks — paste it, or point it at the file.
4. Read the editorial letter. You get a structural diagnosis like the samples above: what's working, what isn't, ranked by severity, each with a class of solution. For an argument, the letter works from the claim, audience, and burden it inferred, and names missing warrants, scope drift, and unmet objections. It never rewrites your prose; drafting happens only in commands you run on purpose, such as /reconstruct.
5. Decide what's next. The letter ends by pointing you onward — a focused /audit, revision planning with /coach, or a submission-readiness check with /ready. For arguments, the usual next steps are /audit argument-red-team or /audit argument-evidence, and /research citation-verifier before anything goes out.

Not sure where to begin? Just /start. When in doubt, that's the front door.

Commands

Start here:

• /start — I have a manuscript or argument — what should I do with it?
• /apodictic — What can this plugin do? Where do I start?

Diagnostic workflows:

• /ready — Is this ready to submit?

Focused tools:

• /reconstruct — How do I draft from only the claims and relationships I approved?
• /adjudicate — Which claims and relationships do I authorize for reconstruction?
• /audit — Run a specific deep-dive analysis — including argument audits (dialectical, argument-decision).
• /research — I need internet-assisted verification.
• /coach — I have a diagnosis — how do I revise?
• /plot-coach — Is my plot structure working?
• /legal-risk — Flag legal exposure (defamation, privacy, rights) for a lawyer's review.
• /world-bible — Check my worldbuilding bible for self-contradiction (rules, magic costs, geography/timeline).
• /triage-feedback — Sort and prioritize beta-reader / editor feedback.
• /reader-questions — Turn my diagnosis into targeted beta-reader questions.

Setup:

• /pre-writing — I have an idea but no manuscript yet.
• /new-project — Set up a new editing project.
• /projects — List, resume, and tidy my editing projects.

Key Terms

A few words you'll meet in the README and in your first editorial letter:

Fiction terms:

• Contract — what APODICTIC infers your manuscript is trying to be: its genre, the promise it makes to readers, and its controlling idea. It predicts this from the text alone, then measures the book against it. When the inferred contract doesn't match what you intended, that mismatch is the core diagnostic signal.
• Controlling idea — the central meaning or argument the story makes through its events. Part of the contract.
• The Firewall — APODICTIC diagnoses structure and names classes of solution (what category of fix the problem needs, and which structural elements are missing or mis-weighted), but its diagnostic workflows never draft your prose or script the specific content that fills them (drafting happens only when you ask: /reconstruct, from claims you approved, or a scene-level handoff, which drafts only after you confirm it) — the events, characters, dialogue, or evidence. Naming a needed element is diagnosis; inventing what fills it is yours. You're the writer; it's the analyst. Applies equally to fiction and argument-shaped nonfiction.
• Pass — one analytical lens over the manuscript (reverse outline, reader experience, reveal economy, and so on). A development edit runs the passes your question requires.
• Macro block — a group of passes organized around a writer question (e.g. "Structure Map," "Reader Dynamics"). Eight blocks in all.
• Audit — a specialized deep-dive beyond the core passes (genre, craft, or tag), run via /audit.
• Genre module — recalibrates what counts as a problem for your genre. A slow opening is a feature in literary fiction, a defect in a thriller.
• Editorial letter — the main deliverable: a structural diagnosis with severity-ranked findings, each paired with a class of solution.
• Severity tiers — every finding is ranked Must-Fix, Should-Fix, or Could-Fix. Severity is locked before any charity reframing, so it can't be quietly softened (the "Deficit Lock").
• Spine — a plot-structure paradigm (Hero's Journey, Mystery, Spiral, …). APODICTIC works with 50 spines across 12 families, not just three-act.
• Reverse outline — reconstructing what each scene actually does, as opposed to what you intended — the starting point for structural diagnosis.

Argument terms (for nonfiction & argument-shaped work):

• Argument contract — the nonfiction equivalent of the fiction contract: the claim you're making, the audience you're addressing, the burden of proof you've accepted, and the stakes you've invoked. APODICTIC infers this from the text and measures the piece against it.
• Claim ladder — the hierarchy of claims in an argument: the core thesis at the top, supporting sub-claims below it, and the evidence tiers beneath those. Dialectical Clarity maps this structure and flags where rungs are missing or unsupported.
• Warrant — the logical connector between a claim and its evidence. A missing warrant is the most common structural failure in argument-shaped nonfiction — the evidence is there, but the link to the claim is implicit, absent, or broken.
• Burden — what you've committed to proving by making your central claim. If the claim is strong, the burden is high; if the piece doesn't meet it, that's the core diagnostic finding.
• Dialectical Clarity — APODICTIC's argument audit: maps the claim ladder, identifies missing warrants, checks rhetorical fairness (does the piece engage the strongest version of the opposition?), and flags scope drift.
• Scope drift — when the argument quietly shifts its claim partway through, so the evidence at the end is supporting a different (usually weaker) claim than the one the piece opened with.

Project Docs

• Skills-only OpenAI plugin — build and test the standalone directory submission ZIP, with host-dependent limitations
• ROADMAP.md — what's planned after publication
• BIBLIOGRAPHY.md — sources and influences (~155 works cited)
• CONTRIBUTING.md — how to contribute, changelog policy
• CODE_OF_CONDUCT.md

License

This work is licensed under CC BY-NC-SA 4.0.

You can use, adapt, and share this framework for non-commercial purposes, with attribution and under the same license.

Author

Joshua A. Miller, PhD

Focused audit ownership and consumer migration: Canonical audit skill families.

Validator tooling uses Python 3.10+: run python plugins/apodictic/scripts/validate.py --check-all from a repository checkout, or python scripts/validate.py <command> ... from an installed plugin. Children reuse that interpreter. The POSIX validate.sh launchers remain optional compatibility entrypoints. Missing Python or helper code cannot produce a clean validation result. CI also runs the full gate on Windows with only Python on PATH; see Native validator migration.