Documentation for the AI agent era. Structured, verifiable, and ready to be consumed by your robot overlords. π€
PAVED is a documentation framework designed for the human+agent workflow. When AI agents need to understand, author, and maintain documentation, prose isn't enough. They need ground truth, constraints, examples, and verification.
What is this thing? What problem does it solve? 1-3 sentences. Include non-goals. Agents need to know boundaries. π―
How do you use it? CLI commands, HTTP endpoints, config keys. Agents thrive on tables and schemas. π‘
How do you know it's working? Golden commands, expected outputs, healthchecks. Without verification, agents hallucinate correctness. β
Concrete copy-paste examples. Happy path, realistic path, failure path. Agents use examples as shape matching. π
Why this design? What must not change? Tradeoffs. This prevents agents from refactoring your intent away. π
"Write docs like you're training a careful junior engineer who can run commands but can't read minds."
Scaffold new documentation with pave new. Components, runbooks, ADRs - all with the right sections already in place.
Generate AI agent prompts with pave prompt. Your project's rules, templates, and context - packaged for LLMs.
Keep your docs navigable with pave index. Automatically scans, categorizes, and creates a doc map.
Enforce doc quality with pave check. Max lines, required sections, verification steps - all configurable.
# Initialize pave in your project
pave init
# Create a new component doc
pave new component my-service
# Generate an index of all docs
pave index
# Validate your documentation
pave check