How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

ralph-setup - Prompt - OpenSmartRoute

Name: ralph-setup
Author: AdrianRogowski

Prompt file imported from AdrianRogowski/auto-sdd (.cursor/commands/ralph-setup.md). Copyright stays with the author.

/ralph-setup — Interactive Environment Setup Wizard

Set up .env.local for the SDD build loop with interactive prompts and auto-detection.

What This Command Does

Detect project — framework, test runner, build tool, package manager
Ask key questions — CLI provider, branch strategy, integrations, parallelism
Generate .env.local — sensible defaults with auto-detected commands pre-populated
Validate tooling — check that agent or claude CLI exists, gh installed if PRs enabled
Print summary — "you're ready" with the exact command to run next

Instructions

Step 1: Check for Existing Config

Read .env.local if it exists. If it does:

Tell the user what's currently configured
Ask if they want to reconfigure or just update specific sections
Preserve any values they don't want to change

If .env.local doesn't exist, copy from .env.local.example as a starting point.

Step 2: Detect Project Type

Scan the project directory to determine:

- Framework: Check for package.json (Next.js, Vite, etc.), Cargo.toml, go.mod, pyproject.toml, etc.
- Package manager: pnpm-lock.yaml, yarn.lock, bun.lockb, package-lock.json
- Test runner: vitest, jest, pytest, cargo test, go test
- Build tool: tsc, vite build, cargo build, go build
- Lint tool: eslint, biome, ruff, clippy
- ORM/DB: drizzle, prisma, alembic, diesel
- E2E: playwright, cypress

Report what was detected.

Step 3: Ask Configuration Questions

Use the AskQuestion tool for each section. Group related questions together:

Core Settings:

CLI provider: Cursor (agent) or Claude Code (claude)?
Branch strategy: chained (default), independent, sequential, parallel, or both?
If parallel:
- How many concurrent features? (default: 3) → PARALLEL_FEATURES
- Rebuild conflicted features on integrated code instead of blocking them? (default: true) → REBUILD_ON_CONFLICT. When a feature branch can't be merged (or breaks build/tests/drift after merge), it gets rebuilt sequentially in a fresh worktree forked from the integration branch, then merged again. Set false to restore the old behavior (mark ⏸️ and preserve the branch for a human).
- Model for the merge-resolution agent? (default: same as AGENT_MODEL) → MERGE_MODEL. This agent resolves source-file merge conflicts semantically during integration; a stronger model here pays off since bad resolutions cost a full rebuild.
Max features per run? (default: 50 for local, 4 for overnight)

Integrations (optional):

Do you use Slack for feature requests? → configure SLACK_* vars
Do you use Jira for tracking? → configure JIRA_* vars
Want draft PRs on completion? → requires gh CLI

Build Validation: Show auto-detected commands and ask if they look right:

Build: {detected} — correct? Override?
Test: {detected} — correct? Override?
Lint: {detected} — correct? Override?
Migration: {detected} — correct? Override?
E2E: {detected} — correct? Override?

Model Selection (optional):

Default model for all steps? (recommend composer-1.5 for Cursor)
Want different models per phase? (show the phase list: spec, build, refactor, drift, compound, review, merge)

Overnight Automation (optional):

Want to set up overnight scheduled runs? → guide through setup-overnight.sh

Step 4: Generate .env.local

Write the .env.local file with:

All configured values
Comments explaining each section
Auto-detected commands filled in (not empty strings)
Unconfigured optional sections commented out

Step 5: Validate

Run these checks:

CLI exists: which agent or which claude (depending on provider)
If PRs enabled: which gh
If mapping generation needed: which yq (optional, has fallback)
Build command works: quick dry-run of BUILD_CHECK_CMD
Test command works: quick dry-run of TEST_CHECK_CMD

Report any issues with fix instructions.

Step 6: Dev Ports Config

Ask about dev server ports:

What ports does your dev server use? (auto-detect from package.json scripts, vite config, etc.)
Set DEV_PORTS and DEV_CMD in .env.local for /clean-slate

Step 7: Summary

Print a clear summary:

✓ ralph-setup complete!

  CLI: cursor (agent)
  Strategy: chained
  Build: npx tsc --noEmit
  Test: pnpm test
  Lint: pnpm lint
  Models: composer-1.5 (all phases)
  Dev ports: 3000,5173
  Dev cmd: pnpm dev

  Next steps:
    /ralph-run          — launch the build loop
    /clean-slate        — kill dev servers and restart
    /spec-first {feat}  — build a single feature

If Things Go Wrong

If auto-detection fails for any command, set it to empty ("") and tell the user:

The build loop will try to auto-detect after the first feature creates project infrastructure
They can always run /ralph-setup again to reconfigure

Install ralph-setup and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

# /ralph-setup — Interactive Environment Setup Wizard Set up `.env.local` for the SDD build loop with interactive prompts and auto-detection. ## What This Command Does 1. **Detect project** — framework, test runner, build tool, package manager 2. **Ask key questions** — CLI provider, branch strategy, integrations, parallelism 3. **Generate `.env.local`** — sensible defaults with auto-detected commands pre-populated 4. **Validate tooling** — check that `agent` or `claude` CLI exists, `gh` installed if PRs enabled 5. **Print summary** — "you're ready" with the exact command to run next --- ## Instructions ### Step 1: Check for Existing Config Read `.env.local` if it exists. If it does: - Tell the user what's currently configured - Ask if they want to reconfigure or just update specific sections - Preserve any values they don't want to change If `.env.local` doesn't exist, copy from `.env.local.example` as a starting point. ### Step 2: Detect Project Type Scan the project directory to determine: ``` - Framework: Check for package.json (Next.js, Vite, etc.), Cargo.toml, go.mod, pyproject.toml, etc. - Package manager: pnpm-lock.yaml, yarn.lock, bun.lockb, package-lock.json - Test runner: vitest, jest, pytest, cargo test, go test - Build tool: tsc, vite build, cargo build, go build - Lint tool: eslint, biome, ruff, clippy - ORM/DB: drizzle, prisma, alembic, diesel - E2E: playwright, cypress ``` Report what was detected. ### Step 3: Ask Configuration Questions Use the AskQuestion tool for each section. Group related questions together: **Core Settings:** 1. CLI provider: Cursor (`agent`) or Claude Code (`claude`)? 2. Branch strategy: chained (default), independent, sequential, parallel, or both? 3. If parallel: - How many concurrent features? (default: 3) → `PARALLEL_FEATURES` - Rebuild conflicted features on integrated code instead of blocking them? (default: true) → `REBUILD_ON_CONFLICT`. When a feature branch can't be merged (or breaks build/tests/drift after merge), it gets rebuilt sequentially in a fresh worktree forked from the integration branch, then merged again. Set false to restore the old behavior (mark ⏸️ and preserve the branch for a human). - Model for the merge-resolution agent? (default: same as `AGENT_MODEL`) → `MERGE_MODEL`. This agent resolves source-file merge conflicts semantically during integration; a stronger model here pays off since bad resolutions cost a full rebuild. 4. Max features per run? (default: 50 for local, 4 for overnight) **Integrations (optional):** 1. Do you use Slack for feature requests? → configure SLACK_* vars 2. Do you use Jira for tracking? → configure JIRA_* vars 3. Want draft PRs on completion? → requires `gh` CLI **Build Validation:** Show auto-detected commands and ask if they look right: - Build: `{detected}` — correct? Override? - Test: `{detected}` — correct? Override? - Lint: `{detected}` — correct? Override? - Migration: `{detected}` — correct? Override? - E2E: `{detected}` — correct? Override? **Model Selection (optional):** 1. Default model for all steps? (recommend `composer-1.5` for Cursor) 2. Want different models per phase? (show the phase list: spec, build, refactor, drift, compound, review, merge) **Overnight Automation (optional):** 1. Want to set up overnight scheduled runs? → guide through `setup-overnight.sh` ### Step 4: Generate .env.local Write the `.env.local` file with: - All configured values - Comments explaining each section - Auto-detected commands filled in (not empty strings) - Unconfigured optional sections commented out ### Step 5: Validate Run these checks: 1. CLI exists: `which agent` or `which claude` (depending on provider) 2. If PRs enabled: `which gh` 3. If mapping generation needed: `which yq` (optional, has fallback) 4. Build command works: quick dry-run of BUILD_CHECK_CMD 5. Test command works: quick dry-run of TEST_CHECK_CMD Report any issues with fix instructions. ### Step 6: Dev Ports Config Ask about dev server ports: - What ports does your dev server use? (auto-detect from package.json scripts, vite config, etc.) - Set `DEV_PORTS` and `DEV_CMD` in `.env.local` for `/clean-slate` ### Step 7: Summary Print a clear summary: ``` ✓ ralph-setup complete! CLI: cursor (agent) Strategy: chained Build: npx tsc --noEmit Test: pnpm test Lint: pnpm lint Models: composer-1.5 (all phases) Dev ports: 3000,5173 Dev cmd: pnpm dev Next steps: /ralph-run — launch the build loop /clean-slate — kill dev servers and restart /spec-first {feat} — build a single feature ``` --- ## If Things Go Wrong If auto-detection fails for any command, set it to empty (`""`) and tell the user: - The build loop will try to auto-detect after the first feature creates project infrastructure - They can always run `/ralph-setup` again to reconfigure

{ "prompt": "# /ralph-setup — Interactive Environment Setup Wizard\n\nSet up `.env.local` for the SDD build loop with interactive prompts and auto-detection.\n\n## What This Command Does\n\n1. **Detect project** — framework, test runner, build tool, package manager\n2. **Ask key questions** — CLI provider, branch strategy, integrations, parallelism\n3. **Generate `.env.local`** — sensible defaults with auto-detected commands pre-populated\n4. **Validate tooling** — check that `agent` or `claude` CLI exists, `gh` installed if PRs enabled\n5. **Print summary** — \"you're ready\" with the exact command to run next\n\n---\n\n## Instructions\n\n### Step 1: Check for Existing Config\n\nRead `.env.local` if it exists. If it does:\n- Tell the user what's currently configured\n- Ask if they want to reconfigure or just update specific sections\n- Preserve any values they don't want to change\n\nIf `.env.local` doesn't exist, copy from `.env.local.example` as a starting point.\n\n### Step 2: Detect Project Type\n\nScan the project directory to determine:\n\n```\n- Framework: Check for package.json (Next.js, Vite, etc.), Cargo.toml, go.mod, pyproject.toml, etc.\n- Package manager: pnpm-lock.yaml, yarn.lock, bun.lockb, package-lock.json\n- Test runner: vitest, jest, pytest, cargo test, go test\n- Build tool: tsc, vite build, cargo build, go build\n- Lint tool: eslint, biome, ruff, clippy\n- ORM/DB: drizzle, prisma, alembic, diesel\n- E2E: playwright, cypress\n```\n\nReport what was detected.\n\n### Step 3: Ask Configuration Questions\n\nUse the AskQuestion tool for each section. Group related questions together:\n\n**Core Settings:**\n1. CLI provider: Cursor (`agent`) or Claude Code (`claude`)?\n2. Branch strategy: chained (default), independent, sequential, parallel, or both?\n3. If parallel:\n - How many concurrent features? (default: 3) → `PARALLEL_FEATURES`\n - Rebuild conflicted features on integrated code instead of blocking them? (default: true) → `REBUILD_ON_CONFLICT`. When a feature branch can't be merged (or breaks build/tests/drift after merge), it gets rebuilt sequentially in a fresh worktree forked from the integration branch, then merged again. Set false to restore the old behavior (mark ⏸️ and preserve the branch for a human).\n - Model for the merge-resolution agent? (default: same as `AGENT_MODEL`) → `MERGE_MODEL`. This agent resolves source-file merge conflicts semantically during integration; a stronger model here pays off since bad resolutions cost a full rebuild.\n4. Max features per run? (default: 50 for local, 4 for overnight)\n\n**Integrations (optional):**\n1. Do you use Slack for feature requests? → configure SLACK_* vars\n2. Do you use Jira for tracking? → configure JIRA_* vars\n3. Want draft PRs on completion? → requires `gh` CLI\n\n**Build Validation:**\nShow auto-detected commands and ask if they look right:\n- Build: `{detected}` — correct? Override?\n- Test: `{detected}` — correct? Override?\n- Lint: `{detected}` — correct? Override?\n- Migration: `{detected}` — correct? Override?\n- E2E: `{detected}` — correct? Override?\n\n**Model Selection (optional):**\n1. Default model for all steps? (recommend `composer-1.5` for Cursor)\n2. Want different models per phase? (show the phase list: spec, build, refactor, drift, compound, review, merge)\n\n**Overnight Automation (optional):**\n1. Want to set up overnight scheduled runs? → guide through `setup-overnight.sh`\n\n### Step 4: Generate .env.local\n\nWrite the `.env.local` file with:\n- All configured values\n- Comments explaining each section\n- Auto-detected commands filled in (not empty strings)\n- Unconfigured optional sections commented out\n\n### Step 5: Validate\n\nRun these checks:\n1. CLI exists: `which agent` or `which claude` (depending on provider)\n2. If PRs enabled: `which gh`\n3. If mapping generation needed: `which yq` (optional, has fallback)\n4. Build command works: quick dry-run of BUILD_CHECK_CMD\n5. Test command works: quick dry-run of TEST_CHECK_CMD\n\nReport any issues with fix instructions.\n\n### Step 6: Dev Ports Config\n\nAsk about dev server ports:\n- What ports does your dev server use? (auto-detect from package.json scripts, vite config, etc.)\n- Set `DEV_PORTS` and `DEV_CMD` in `.env.local` for `/clean-slate`\n\n### Step 7: Summary\n\nPrint a clear summary:\n\n```\n✓ ralph-setup complete!\n\n CLI: cursor (agent)\n Strategy: chained\n Build: npx tsc --noEmit\n Test: pnpm test\n Lint: pnpm lint\n Models: composer-1.5 (all phases)\n Dev ports: 3000,5173\n Dev cmd: pnpm dev\n\n Next steps:\n /ralph-run — launch the build loop\n /clean-slate — kill dev servers and restart\n /spec-first {feat} — build a single feature\n```\n\n---\n\n## If Things Go Wrong\n\nIf auto-detection fails for any command, set it to empty (`\"\"`) and tell the user:\n- The build loop will try to auto-detect after the first feature creates project infrastructure\n- They can always run `/ralph-setup` again to reconfigure", "variables": [], "notes": "Set up `.env.local` for the SDD build loop with interactive prompts and auto-detection.", "metadata": { "source": { "provider": "github-cursor-commands", "repository": "https://github.com/AdrianRogowski/auto-sdd", "path": ".cursor/commands/ralph-setup.md", "ref": "69ff0aaa63496bdbf484f562965e69ee50153c22", "url": "https://github.com/AdrianRogowski/auto-sdd/blob/69ff0aaa63496bdbf484f562965e69ee50153c22/.cursor/commands/ralph-setup.md", "key": "AdrianRogowski/auto-sdd/.cursor/commands/ralph-setup.md" } } }

ralph-setup

About

/ralph-setup — Interactive Environment Setup Wizard

What This Command Does

Instructions

Step 1: Check for Existing Config

Step 2: Detect Project Type

Step 3: Ask Configuration Questions

Step 4: Generate .env.local

Step 5: Validate

Step 6: Dev Ports Config

Step 7: Summary

If Things Go Wrong

Install ralph-setup and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews