Install

Several install paths depending on the agent you use.

Claude Code (recommended)

Scriptorium ships as a Claude Code plugin. Two ways to install:

Live-linked (personal / development use)

git clone https://github.com/seandavi/scriptorium ~/Documents/git/scriptorium
cd ~/Documents/git/scriptorium
uv sync
uv run scriptorium install --mode dev-link

This symlinks the source repo into ~/.claude/plugins/scriptorium/. Edits to the source propagate immediately. Best for the maintainer or anyone iterating on skills.

Copy-installed (clean / shared use)

uv tool install agentic-scriptorium
scriptorium install --mode copy

This copies the plugin into ~/.claude/plugins/scriptorium/. Updates require re-running the install command. Cleaner for users who don’t plan to edit the skills.

Verifying

Restart Claude Code, then in any session:

/help

You should see entries like scriptorium:citation-audit, scriptorium:reviewer-simulation, scriptorium:argumentative-flow once the first skill PRs land.

npx skills (any supported agent)

The skills CLI installs scriptorium’s skills into the skills directory of Claude Code, Codex, Cursor and the other agents it supports. It needs Node.js (for npx) and no Python. If you use Claude Code, the plugin install above is still the recommended path.

Install all skills (the CLI asks which skills and agents you want):

npx skills add seandavi/scriptorium

Install one skill, or a few, with --skill. --list shows what is available without installing anything:

npx skills add seandavi/scriptorium --skill citation-audit
npx skills add seandavi/scriptorium --skill init --skill explain --skill tour
npx skills add seandavi/scriptorium --list

Pick agents with -a (-a '*' for every supported agent), and add -y to skip the prompts:

npx skills add seandavi/scriptorium -a claude-code codex -y

By default skills go into the current project (for Claude Code, .claude/skills/). -g installs them for your user instead, so every project sees them:

npx skills add seandavi/scriptorium -g

When you install for more than one agent, the CLI keeps one copy under .agents/skills/ and symlinks it into each agent’s directory. --copy writes a separate copy into each agent’s directory instead.

Update and remove:

npx skills update
npx skills remove citation-audit

npx skills list shows what is installed. The CLI records each skill’s source in skills-lock.json in the project.

Each skill is installed on its own, without the rest of the repo. init, explain and tour carry the files they read most (the state schema, the example state file, the knowledge-layer README and the two knowledge/conventions/ notes) under their own assets/ directory. What the plugin and CLI add on top:

• The scriptorium CLI — scriptorium validate, scriptorium init and scriptorium prompt-pack. Without it, init checks MANUSCRIPT_STATE.yaml against its bundled schema in-conversation, which is best-effort; scriptorium validate is the authoritative check. Install it with uv pip install agentic-scriptorium.
• The full knowledge layer, locally. A plugin install or a checkout has every note under knowledge/ and every skill’s SKILL.md. A standalone explain fetches any note or skill it does not bundle from GitHub main, which may be newer than your installed copy, and needs network access to do so.

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

Each skill ships with a platform-neutral prompt.md you can paste into any LLM directly. The CLI’s prompt-pack subcommand emits these prompts in a shape your agent can use.

Per-skill files (default)

scriptorium prompt-pack --output prompts/

This writes one .md per shipped skill into prompts/ plus a README.md manifest listing each skill, its category, and a one-line description:

prompts/
  README.md
  citation-audit.md
  reviewer-simulation.md
  argumentative-flow.md
  ...

Paste the specific prompt you want into your agent’s context — most non-Claude-Code users only need one skill at a time, not all of them.

If you’d like the filenames to carry the project namespace (useful when you’re dropping skills from multiple sources into the same directory), add --prefix scriptorium-:

scriptorium prompt-pack --prefix scriptorium- --output prompts/
# writes prompts/scriptorium-citation-audit.md, etc.

Single concatenated file

If you want every skill prompt in one document — for archival, for an agent that loads its whole system prompt from a single file, or just to scan everything at once — use --single-file:

scriptorium prompt-pack --single-file --output scriptorium-prompts.md

Without any agent — manual use

Every skill’s prompt.md is human-readable and can be used as a checklist. The skills double as editorial guidance.

Requirements

• Claude Code path: Claude Code installed and on PATH.
• npx skills path: Node.js, for npx.
• Prompt-pack path: Python 3.10+ to run the CLI (or just read the individual prompt.md files directly with no tooling).
• State validation: the CLI bundles jsonschema and pyyaml for scriptorium validate.

Uninstall

rm -rf ~/.claude/plugins/scriptorium

For the live-linked install, this removes the symlink only; the source repo is untouched.