Lovable prompting playbook

Lovable does best with small, specific, verifiable prompts. This skill helps you (or another agent) write prompts that ship.

Anatomy of a good Lovable prompt

[Context]   What part of the app, which files or routes.
[Goal]      What "done" looks like in user-visible terms.
[Approach]  Optional. Pick stack / pattern if it matters.
[Limits]    What NOT to change.
[Verify]    How you'll check it works.

Example:

On the /dashboard route, add a "Revenue (30d)" KPI card to the top row, using useRevenue30d (already exists). Show value, % delta, and a sparkline. Do not touch the other cards. I'll verify by loading the dashboard and comparing the number to Stripe.

Sequence work into prompts

Break a feature into 3–7 prompts, each shippable on its own:

1. Schema — tables, RLS, types (migration-playbook).
2. API — fetchers + hooks (typed-api-hooks-forms).
3. UI shell — empty/loading/error layout (error-states-and-empty-ui).
4. Wiring — connect UI to hooks.
5. Edge cases — auth, permissions, validation.
6. Polish — copy, a11y, mobile.

After each prompt, review the chat diff before moving on.

When Lovable goes off course

• "Revert to the previous version and try again, keeping X unchanged."
• "Show me only the diff for path/to/file.tsx."
• "Stop. Summarize what you changed in the last 3 messages before continuing."
• Use a skill invocation (/refactor-safe-diff) to constrain scope.

Pair prompts with skills

Phrases that work well

• "Implement X. Do not change Y."
• "Add a checklist comment in the PR description listing every file you touched."
• "If you are unsure between two approaches, ask me before writing code."
• "Reuse the existing useFoo hook; do not create a duplicate."

Phrases that cause trouble

• "Improve the app." — undefined target.
• "Refactor everything." — invites scope creep; use /refactor-safe-diff with explicit files.
• "Add a few small fixes." — bundle multiple things; you'll lose track of regressions.

Workspace setup recommendations

• Add knowledge entries for: stack, brand, banned patterns (any, inline Supabase in components).
• Add this repo's skills to your workspace and let descriptions trigger them automatically.
• Keep prompts in a note (Linear, Notion, or prompts.md) so they can be re-run on a new branch.

Avoid

• Mega-prompts that mix schema, UI, billing, and copy.
• Trusting Lovable to "remember" earlier sessions — assume zero memory across chats.
• Letting failing diffs accumulate; revert early.