# Rye - Agent Setup Instructions

You are an agent setting up Project Rye for a user. Rye is an open source,
agent-native temporal knowledge graph schema that installs INTO a PostgreSQL
database. It is a schema, not a separate database, application, or framework.

Follow the steps in order. Ask the user only the questions listed here.

## 1. Check prerequisites

Rye needs ONE of:

- Docker with Docker Compose (local trial - recommended first run), or
- a PostgreSQL 15+ connection string (Rye installs alongside existing tables
  in its own `rye` schema and never resets remote data).

Verify Docker with `docker info` (the daemon must be running, not just
installed). Otherwise ask the user for a connection string. If neither is
available, stop and tell the user which one to provide.

## 2. Install

Local (Docker):

    curl -fsSL https://projectrye.dev/onboard | sh

Remote (existing database):

    curl -fsSL https://projectrye.dev/onboard | sh -s -- --remote "$DATABASE_URL"

This clones the repo into ./project-rye, installs the Rye schema, syncs
plugin/skill/capability metadata, and writes .rye.env (local connection
config - generated for the user, never hand-edited). It adds no example data
unless you pass --seed.

## 3. Verify

    cd project-rye
    ./scripts/rye doctor
    ./scripts/rye status

Success looks like: doctor reports the schema and metadata present; status
returns node/scope counts without errors. On failure, read .rye/install.log
before retrying.

## 4. Create the first scope - required before any data intake

Ask the user:

    "What is the one limited workflow Rye should assist first?"

Name the scope after that organizational purpose, never after a data source
or retrieval channel ("Wholesale account follow-up", not "Slack ingest").

    ./scripts/rye onboard create \
      --label "<workflow name>" \
      --purpose "<one sentence: what should be remembered, for whom>"

## 5. Hand yourself context

    ./scripts/rye context --json

Returns the catalog, active scope, source inventory, pending confirmations,
and compiled policy. Read it before writing anything to the graph.

## Rules - non-negotiable

- Scope first. Never ingest sources or promote facts before an active scope,
  boundary, and review policy exist.
- Connectors collect; Rye classifies. Register sources neutrally, mark them
  needs_confirmation, and ask the user to confirm a source's purpose before
  inferring business meaning from it.
- Candidates before truth. New facts enter as candidates with provenance.
  Promotion to accepted knowledge follows the review policy - usually a human
  decision, never yours alone.
- Stay in your schema. Rye lives in the `rye` schema. Do not modify the
  user's other tables.

## Going deeper

- Agent skills (onboarding, source intake, agent ops):
  npx skills add BLTGV/project-rye --skill rye-onboarding
- CLI reference: https://projectrye.dev/docs/reference/cli/
- Onboarding scopes: https://projectrye.dev/docs/getting-started/onboarding/
- Machine-readable site map: https://projectrye.dev/llm.txt