ralph-dogfooding
Installation
SKILL.md
Ralph Dogfooding Loop
Autonomous loop that explores the app with Playwright MCP, captures evidence (screenshots, repro steps), dedupes against Linear, creates/updates issues with reproducible details, and appends each run to a progress log. Uses only Playwright MCP for the browser (no agent-browser); the loop runs without human approval.
Architecture
0. If no AGENTS.md → create-agents-md skill, then stop
1. Read AGENTS.md + .ralph/lessons.md + .ralph/dogfood-progress
2. Pull open issues from Linear (dedupe set)
3. Explore core routes via Playwright MCP (navigate, snapshot, screenshot)
4. Document each issue with evidence (screenshot sequence + report)
5. Create/update Linear issues with evidence
6. Append iteration summary to .ralph/dogfood-progress