Ralph

Prepare files for Ralph execution with exhaustive pre-flight checks. Creates PRD.md, plan.md, activity.md. BLOCKS until all prere…

vindows 3 updated 3mo ago
Claude CodeGeneric
View source ↗
---
name: ralph
description: Prepare files for Ralph execution with exhaustive pre-flight checks. Creates PRD.md, plan.md, activity.md. BLOCKS until all prerequisites are cleared.
---

# /ralph - Ralph Execution Preparation

You are preparing files for autonomous Ralph execution. Quality and completeness are paramount.

## CRITICAL RULES

1. **Exhaustive pre-flight checks** - Identify EVERYTHING that needs to happen first
2. **BLOCK until cleared** - Do not create files until prerequisites are handled
3. **Super clear instructions** - Every step must be unambiguous
4. **Phase X.0 → X.1 TDD** - Tests before implementation in all tasks
5. **FILES GO IN PROJECT ROOT** - ALWAYS put PRD.md, plan.md, activity.md, PROMPT_build.md in the project root directory. NEVER create subfolders. ALWAYS overwrite existing files.
6. **SELF-VERIFY EVERYTHING POSSIBLE** - Run commands yourself to verify. Only ask humans for things that REQUIRE human judgment (design decisions, business logic, external service setup). Never ask humans to run simple verification commands.
7. **CITE SOURCES FOR ALL CLAIMS** - When recommending approaches, cite: codebase patterns (file:line), external best practices (URLs), or prior decisions (alignment.md, retros). If citing your own reasoning, label it explicitly.
8. **SURFACE ALL ASSUMPTIONS** - Before proceeding past any phase, state assumptions explicitly and get user confirmation. Example: "I'm assuming the database schema already has X. Is that correct?"

---

## Self-Verification Protocol

**ALWAYS verify these yourself (run the command, don't ask human):**
- `npx tsc --noEmit` - TypeScript compiles
- `npm run build` - Build succeeds
- `npm test` / `npm run test:run` - Tests pass
- `npm run lint` - Linting passes
- File exists checks - Use Glob/Read tools
- Import verification - Run TypeScript compiler
- Package installation - Run `npm install` and check

**ONLY ask humans for:**
- Design decisions (how should X look?)
- Business logic choices (what should happen when Y?)
- External service setup (API keys, database credentials, OAuth config)
- Deployment decisions
- Anything requiring account access you don't have

**Example - BAD (don't do this):**

"Please run npx tsc --noEmit and let me know if it compiles."


**Example - GOOD (do this):**

Running TypeScript compiler to verify... [runs npx tsc --noEmit] ✓ TypeScript compiles successfully


---

## Phase 1: Alignment Check

### Step 1: Check for alignment.md

Use Glob/Read tools to look for `alignment.md` in the project root.

**If alignment.md EXISTS:**
1. Read the file
2. Check the `Last Updated` timestamp
3. If timestamp is recent (same day): Use this alignment, proceed to Phase 2
4. If timestamp is old (different day): Warn user that alignment may be stale

Found alignment.md (Last Updated: [timestamp])

Alignment Summary

[show key points from alignment.md]

This alignment looks [fresh/stale]. Proceed to pre-flight checklist?


**If alignment.md DOES NOT EXIST:**

Present options with clear gap communication:

I don't see a /plan summary for this feature. Your options:

A) Run /plan first ⭐ Recommended Ensures requirements are clear, codebase is analyzed, risks are identified. Run: /plan [your feature description]

This gives you: ✓ Codebase pattern analysis ✓ Technical approach research ✓ Risk assessment ✓ Test strategy

B) Quick alignment now (3 min) I'll ask 3 key questions to understand what you want.

⚠️ What you'll miss:

  • No formal codebase analysis
  • Less thorough risk assessment
  • Test patterns not researched

C) Proceed to /ralph anyway I'll create execution files based on your description.

⚠️ What you'll miss:

  • No codebase pattern analysis performed
  • No explicit requirements alignment
  • Higher risk of rework if requirements were unclear
  • Pre-flight checklist may miss dependencies
  • Tests may not follow existing conventions

If you choose this, I'll note it in the files created.

Which would you like?


**If user chooses B:**
Ask 3 key questions:
1. What exactly should this feature do? (success criteria)
2. What should it NOT do? (scope boundaries)
3. Any constraints I should know about? (tech, time, patterns)

Then proceed to Phase 2.

**If user chooses C:**
Add this header to PRD.md:

⚠️ PLANNING NOTE: This PRD was created without /plan workflow.

  • Codebase analysis was NOT performed
  • Risk assessment was NOT performed
  • Test patterns were NOT researched Review carefully before Ralph executes.

Then proceed to Phase 2.

---

## Phase 2: Pre-Flight Research

Launch exhaustive research to identify all prerequisites:

### Check 1: Dependencies

What the feature requires:

  • New npm packages needed?
  • New environment variables?
  • Database migrations/schema changes?
  • External API keys or credentials?
  • Configuration file changes?

### Check 2: Human Prerequisites

What must the HUMAN do before Ralph runs:

  • Supabase SQL to execute?
  • Manual file creation needed?
  • Service configuration?
  • Third-party setup (API keys, OAuth, etc.)?

### Check 3: Integration Points

How this integrates with existing code:

  • Which files will be modified?
  • Any breaking changes?
  • Any deprecations to handle?

---

## Phase 3: Pre-Flight Checklist (BLOCKING)

Present checklist that MUST be cleared before files are created:

```markdown
# ⚠️ PRE-FLIGHT CHECKLIST

These items MUST be completed before Ralph can run successfully.
Do NOT skip this - Ralph will fail if prerequisites aren't met.

## Database/Backend
- [ ] Run SQL: `[exact SQL statement]`
- [ ] Create Supabase function: [name]
- [ ] Update RLS policies: [description]

## Dependencies
- [ ] Install: `npm install [package]`
- [ ] Add to package.json: [what]

## Environment
- [ ] Add to .env: `NEW_VAR=[description of what value should be]`
- [ ] Configure: [service/setting]

## Manual Setup
- [ ] Create file: [path] with [cont

Maintain Ralph?

Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.

[Ralph on getagentictools](https://getagentictools.com/loops/vindows-ralph-ralph-execution-preparation?ref=badge)