Imported from miguelfiguera/miguelfiguera (AGENTS.md). Install upstream with npx skills add miguelfiguera/miguelfiguera. Copyright stays with the author.

Ralph Pipeline Agent

You are an agent executing ONE task in the Ralph Loop pipeline.

Project Context

Replacing Next.js portfolio with Astro for Miguel Figuera Quintero (Full-Stack Developer, Venezuela). Branch: feat/astro-migration. Astro project at repo root.

NO blog, NO dashboard, NO formspree. Pages: Home, Services, Portfolio, Products (ERPNext-ready). Must include: Odoo Partner badge, Shopify experience, real certifications only.

Design System (preserve EXACTLY)

• shadcn/ui React components as Astro islands
• HSL CSS variables (neutral base) — light + dark mode
• Dark mode: --background: 0 0% 3.9%, --foreground: 0 0% 98%, etc.
• Custom silver accent: #C0C0C0
• Font: Arial, Helvetica, sans-serif
• Plugin: tailwindcss-animate | Icons: Lucide React

Stack

Astro 5.x + @astrojs/react + @astrojs/tailwind | TypeScript strict | Vitest | clsx + tailwind-merge + class-variance-authority

Existing Images (REUSE)

public/Logos/mfLogo.png (+ variants), public/myPhoto/mySelf.png, public/placeholder*, public/favicon.ico

YOUR EXACT WORKFLOW — Do these 8 steps in order. Skip nothing.

Step 1: Read learnings

cat .ralph/learnings.md

Read what previous agents learned. Don't repeat their mistakes.

Step 2: Read status and find your task

cat .ralph/status.json

If a task is in_progress, that's yours (previous agent died) — verify existing work first. Otherwise take the first pending task. Note the task ID (e.g., 0c, 1a-adv).

Step 3: Mark in_progress IMMEDIATELY

Do this BEFORE any other work. Replace TASK_ID with your actual task ID:

python3 -c "
import json
with open('.ralph/status.json','r') as f: data=json.load(f)
data['tasks']['TASK_ID']['status']='in_progress'
with open('.ralph/status.json','w') as f: json.dump(data,f,indent=2)
print('Marked TASK_ID as in_progress')
"

Step 4: Read the PRD for your task

cat PRD-ralph-pipeline.md

Find the section ## Tarea TASK_ID: and follow its instructions exactly.

Step 5: Do the work

Execute what the PRD says. Verify the acceptance criteria (AC) are met.

Step 6: Write learnings

Always write what you learned, even if the task was trivial. Append to .ralph/learnings.md:

cat >> .ralph/learnings.md << 'LEARNING'

### Task TASK_ID — Completed
- What I did: (brief summary)
- What I learned: (gotchas, surprises, tips for next agents)
- Files created/modified: (list)
---
LEARNING

Step 7: Mark completed

THIS IS THE MOST IMPORTANT STEP. If you skip this, the pipeline considers your work a FAILURE and will redo it.

python3 -c "
import json
with open('.ralph/status.json','r') as f: data=json.load(f)
data['tasks']['TASK_ID']['status']='completed'
with open('.ralph/status.json','w') as f: json.dump(data,f,indent=2)
print('Marked TASK_ID as completed')
"

Verify it worked:

python3 -c "
import json
with open('.ralph/status.json','r') as f: data=json.load(f)
print('Status:', data['tasks']['TASK_ID']['status'])
"

The output MUST say Status: completed. If it doesn't, something is wrong — fix it before stopping.

Step 8: Git commit and STOP

git add -A
git commit -m "your conventional commit message"

Then STOP. Do not continue to the next task.

If blocked

If you can't complete the task:

python3 -c "
import json
with open('.ralph/status.json','r') as f: data=json.load(f)
data['tasks']['TASK_ID']['status']='blocked'
with open('.ralph/status.json','w') as f: json.dump(data,f,indent=2)
"

Write WHY in learnings.md, then STOP.

Rules

• One task per agent. Do your task, mark completed, stop.
• Steps 3, 6, 7 are non-negotiable. Checkpoint, learnings, mark completed.
• Don't modify code outside your task scope.
• Read prior learnings — they exist for a reason.
• Commits use conventional commits. Never mention AI/Claude/Anthropic.

Mandatory Timeouts

timeout 120 npm run build 2>&1
timeout 60 npx vitest run 2>&1
timeout 30 npx tsc --noEmit 2>&1

NEVER run build/test/tsc without timeout.