Skip to content

Getting Started

This page takes you from nothing to a working research assistant with persistent memory and a paper you can build, in about 15 minutes.

0. Install and open Claude Code

The kit is driven entirely from Claude Code, Anthropic's terminal-based agentic coding tool. If you don't have it yet:

  1. Install it following the instructions at claude.com/claude-code.
  2. Sign in with your Anthropic account.
  3. Confirm it runs by opening a terminal and typing claude in any folder.

Everything below — the scaffolder, the paper workflow, every /skill — runs inside a Claude Code session in your project folder.

1. Scaffold

npm create wren@latest my-research
cd my-research

You'll answer a few prompts (name, project namespace, database backend, reference manager). Everything is editable later in research-config.yml.

Prefer to clone the repo directly?

cp research-config.example.yml research-config.yml   # edit a handful of fields
./setup.sh

The scaffolder then renders your templates, installs the hooks, and runs a verification pass. A green summary means you're ready.

2. (Optional) add secrets

Only if you enabled an integration:

cp .env.example .env    # then paste keys for whatever you turned on

The Core needs none of these.

3. Open in Claude Code

Open the folder in Claude Code. On start, the SessionStart hook surfaces any recent /retro learnings (none yet — that's expected) and, on a fresh project, nudges you to run the guided onboarding.

4. Run the guided onboarding

Run:

/setup wren

This is the recommended first step. It gives you a short tour of the system, asks a few questions about your research, sets up the right tools (database backend, references) to match, and starts your first paper — or plans a paper series. See Welcome for the same tour in writing.

Prefer to jump straight in? Run /paper, or duplicate the example by hand:

cp -r papers/example-paper papers/my-first-paper
echo "my-first-paper" > .current_paper

Edit papers/my-first-paper/STATUS.md frontmatter (title, journal, phase). From now on the UserPromptSubmit hook injects that paper's status into every turn so the assistant always knows what you're working on.

5. See memory capture

After doing some work, run:

/retro

It proposes a short learning and, on your approval, writes a learning_<date>_*.md into your Claude memory directory and indexes it. Next session, that learning resurfaces automatically. That's the self-improving loop — see Concepts for the full memory model.

6. Build a paper to Word

Put some Markdown in papers/my-first-paper/drafts/, then:

npm run paper:setup                     # one-time: installs python-docx + markdown
npm run paper:status
npm run paper:build my-first-paper      # Markdown → .docx
npm run paper:extract my-first-paper    # .docx → Markdown (round-trip)

Note

paper:setup is a one-time step — the Word build/extract engine needs python-docx and markdown (in requirements.txt). Scaffolding itself doesn't install them, so run this once before your first build. More detail in Papers & the build engine.

Where things live

You want to… Go to
Change your config research-config.yml (re-run ./setup.sh) — see Configuration
Read how memory is organised Concepts
Set up a real database Databases
Connect Zotero / other MCPs References & integrations
Toggle corpus memory research-config.yml → tiers.research_memory — see Corpus memory

Troubleshooting

  • Hooks not firing? python3 scripts/setup/install_hooks.py (makes them executable). Confirm .claude/settings.json exists.
  • {{TOKENS}} still in CLAUDE.md? Re-run ./setup.sh — the renderer didn't complete.
  • PyYAML error? python3 -m pip install pyyaml.
  • paper:build says "Missing a dependency" / No module named 'docx'? Run npm run paper:setup once.
  • Verify anytime: npm run setup:verify.