Same prompt. Same model. One had your references.
Two real runs of Claude Code for each screen. The only change on the right: Moodfile handed it the references below and MOODFILE.md. Drag to compare, or open the raw HTML each run wrote.
Prompt, both runs
Build the landing page for Ledger, a money app for freelancers. It tracks who owes you, sorts your spending and sets aside tax. Collect waitlist emails.
What Moodfile handed over
MOODFILE.md, 5 references
Check our work
Both files are exactly what the agent wrote. The references are designs we made for this page.
Prompt, both runs
Build the home dashboard for Ledger: cash on hand, money clients owe and who's late, spending this month, six months of cash flow, recent transactions.
What Moodfile handed over
MOODFILE.md, 5 references
Check our work
Both files are exactly what the agent wrote. The references are designs we made for this page.
Watch a real session.
Claude Code with the Moodfile MCP, sped up 7×. It reads your MOODFILE.md, opens four references by their IDs, builds the page, then says which reference each section follows.
Save it once.
Your agent remembers.
Free to join. Every friend you bring moves you up the line. Or skip it for $9.