agentic-scriptorium 0.2.0

An agentic scriptorium for scholarly writing — coordinated AI capabilities for manuscripts, grants, and reviews.

scriptorium

An agentic scriptorium for scholarly writing.

Scriptorium is an agentic operating system for scholarly writing. It coordinates AI capabilities — citation audit, reviewer simulation, structural analysis — around shared editorial state, so manuscripts, grants, and reviews are improved as a workflow rather than a series of one-shot prompts.

Why this exists

A medieval scriptorium was a coordinated workspace where multiple scribes copied, annotated, glossed, and corrected manuscripts under shared conventions. The modern scholarly-writing process — drafting, citing, reviewing, revising, publishing — has the same shape, but most AI tools sit outside of it: they generate prose in isolation, lose context between sessions, and offer no way to compose improvements.

Scriptorium puts the agentic-AI layer inside the workflow:

• A single source of editorial truth (MANUSCRIPT_STATE.yaml) that every skill reads.
• A collection of conservative, single-responsibility skills (citation audit, reviewer simulation, argumentative-flow analysis) that emit structured outputs another skill can consume.
• A discipline of inspectable transformations: semantic diffs over unconstrained rewriting; preserved citations and statistics by default; no hidden state.

A deliberate scope. Scriptorium operates on prose the author has written or scaffolding the author has declared. It does not produce prose from blankness — pre-draft ideation, "help me figure out what to study", or section-from-scratch generation sit outside what scriptorium does. That cut is structural, not aspirational: it's what the existing skills already enforce (citation-audit refuses to invent citations; argumentative-flow refuses to add claims; venue-fit refuses to recommend without declared content), and it's what keeps behaviours auditable rather than vibes-based. The full rationale, grounded in Hayes' 2012 writing-process model and the AI-writing failure-modes literature, lives in knowledge/conventions/declared-work-scope.md.

The medieval scriptorium had an operating model. The agentic one is just now being built.

Install

Claude Code (recommended)

Scriptorium hosts its own plugin marketplace at seandavi/scriptorium. Easiest install — no Python toolchain needed:

/plugin marketplace add seandavi/scriptorium
/plugin install scriptorium@scriptorium

(seandavi/scriptorium is the GitHub owner/repo shorthand for https://github.com/seandavi/scriptorium. Append @v0.2.0 or any ref to pin.) Skills appear as scriptorium:tour, scriptorium:init, scriptorium:citation-audit, scriptorium:reviewer-simulation, scriptorium:argumentative-flow, scriptorium:explain.

Once installed, run:

/scriptorium:tour

This is a short conversational walk-through — three or four turns — that orients you to scriptorium and ends with one concrete next command. It writes nothing and invokes no other skill; it's the single entry point we recommend pointing new users at instead of linking them to documentation.

Updates: /plugin marketplace update then /plugin update scriptorium@scriptorium.

If you also want the Python CLI (scriptorium init, scriptorium validate, scriptorium trace), or you want a live-linked dev install, see INSTALL.md.

Other agents (Codex, Gemini, Hermes, ChatGPT, …)

Each skill ships with a platform-neutral prompt.md you can paste into any LLM directly. To get all skills concatenated into a single prompt-pack file:

scriptorium prompt-pack -o scriptorium-prompts.md

See INSTALL.md for per-platform recipes.

Quick start

# 1. Scaffold editorial state for your manuscript
scriptorium init /path/to/your/manuscript

# 2. Populate it. Two ways:
#    a) Edit /path/to/your/manuscript/MANUSCRIPT_STATE.yaml directly.
#    b) In Claude Code inside the manuscript repo, run the conversational
#       bootstrap that walks you through core_claims, known_weaknesses,
#       terminology, audience, and tone:
#       /scriptorium:init

# 3. Validate the result
scriptorium validate /path/to/your/manuscript/MANUSCRIPT_STATE.yaml

# 4. Run the leaf skills in Claude Code, inside the manuscript repo:
/scriptorium:citation-audit
/scriptorium:reviewer-simulation
/scriptorium:argumentative-flow

A fully-populated reference manuscript lives at templates/MANUSCRIPT_STATE.example.yaml (an imaginary biomedical paper that exercises every field in the schema). Use scriptorium init --example /tmp/example to drop a copy anywhere as a learning aid.

What's in the box (v0.1)

Four bundled skills, exposed through both Claude Code and the platform-neutral prompt-pack:

Skill	Phase	What it does
init	bootstrap	Conversational pass that populates MANUSCRIPT_STATE.yaml (core claims, known weaknesses, terminology, audience, tone) and routes to the right next skill for the document phase. Pairs with the scriptorium init CLI subcommand, which scaffolds the file.
citation-audit	leaf	For each scientific claim: identify citation support, evaluate evidence strength, flag overreach. Outputs a structured claim/citation/assessment/recommendation table. No text modification.
reviewer-simulation	leaf	Simulates four reviewer personas (methodological skeptic, domain expert, translational reviewer, statistical reviewer). Outputs Major/Minor Critiques, Fatal Concerns, Enthusiasm Drivers, Suggested Revisions, Acceptance Risk.
argumentative-flow	transformative	Improves logical and argumentative coherence while preserving citations, statistics, and terminology. Outputs Structural Diagnosis / Logical Gaps / Proposed Outline / Revised Text / Remaining Weaknesses.

The scriptorium CLI exposes six subcommands: install, init, validate, prompt-pack, list, and trace. scriptorium trace extracts skill invocations from Claude Code transcripts as structured records conforming to schemas/trace.schema.json — useful for self-inspection and (eventually) eval-loop work. See scriptorium --help for the full surface.

Deliberately not in v0.1 (see DESIGN.md for the build order): orchestrators, drafting skills, additional knowledge-layer contributions beyond what skills currently cite, Codex/Gemini adapters. These earn their way in once the v0.1 skills have been used on real manuscripts.

Companion tools

Scriptorium is the editorial layer; these tools are useful neighbors for the rest of the scholarly-writing workflow:

• Semantic Scholar MCP and PubMed MCP — bibliographic lookup and full-text retrieval as MCP servers, consumable by Claude Code (and any MCP-aware agent). Pair well with scriptorium:citation-audit when claims need to be checked against primary literature.
• quartobot — resolves persistent-identifier cite keys (@pmid:, @doi:) in Quarto documents before citeproc renders them. Used inside the scriptorium docs build for the knowledge/ evidence base; useful standalone for any Quarto-based scholarly writing.

These are external projects, not bundled with scriptorium. Install them separately following each project's own instructions.

Design principles

The system separates:

• generation (write new prose)
• critique (assess existing prose)
• validation (check for structural / factual issues)
• normalization (terminology, style, journal conventions)

And prefers:

• inspectable transformations over opaque rewrites
• structured outputs over freeform rambles
• semantic diffs over surface rewrites
• explicit checkpointing over hidden state

See DESIGN.md for the full design philosophy and the roadmap.

Status

v0.1 in flight. Four bundled skills (init, citation-audit, reviewer-simulation, argumentative-flow), the consolidated scriptorium CLI (six subcommands, including trace for transcript analysis), the shared state schema, the trace schema, and the Venice example. Orchestrators, additional skills, and platform adapters follow once the leaves prove out on real manuscripts.

License

Dual-licensed by category:

• Code — source under src/, tests, schemas, scripts, and configuration files — is MIT.
• Documentation and knowledge layer — the docs/ site, the knowledge/ evidence base, top-level prose files (README.md, DESIGN.md, INSTALL.md, CONTRIBUTING.md), and per-skill README.md files — is CC BY 4.0.