Journey Docs

Runs between preflight and your first build stage to front-load the right Butterbase documentation into context. It reads the plan, figures out which capabilities you're actually using (auth, storage, functions, whatever), fires off one butterbase_docs call per topic, then writes a summary cache file that downstream stages can reread without burning tokens. The anti-patterns section is honest: don't call topic "all", don't skip the cache, don't prime docs you won't need. It's mechanical glue code that keeps later stages from hallucinating API shapes because they started cold.

GitHub stars: 532
last updated: Jun 11, 2026

View on GitHub

Install

$ npx -y skills add butterbase-ai/butterbase-skills --skill journey-docs --agent claude-code

Installs into .claude/skills of the current project.

Files

Journey Docs Gate

A short, mechanical priming step. The goal is that every downstream stage starts with the relevant docs already in the conversation, so the model doesn't invent API shapes.

When to use

Invoke automatically after journey-preflight returns. Also invoke standalone (/butterbase-skills:journey-docs) any time the user changes the plan or you realize a stage is using a capability you haven't refreshed this session.

Procedure

Read the plan. Open docs/butterbase/02-plan.md. Identify every Butterbase capability in use (Tables → schema; RLS → auth; Auth → auth; Storage → storage; Functions → functions; AI → ai; RAG → rag; Realtime → realtime; Durable → functions; Frontend → frontend; Integrations → integrations; Substrate → substrate; Payments → billing).
Call butterbase_docs per topic. For each unique topic from step 1, call butterbase_docs with that topic argument. Skip duplicates.

Write the cache file. Create docs/butterbase/03b-docs-cache.md with this structure:

---
primed_at: <ISO timestamp>
topics: [<comma-separated list>]
---

# Docs Cache

SKILL.md on GitHub

Details

Repository: github.com/butterbase-ai/butterbase-skills
Categories: DevOps & CI/CD, Documentation, Productivity & Planning
License: MIT
First seen: Jun 11, 2026