Advanced Lovable prompting: Skills, goals, Plan mode and prompt chains
Advanced prompting in Lovable is using the controls around the chat box: the mode picker (Build, Chat, Plan), the slash menu with Skills and /goal, the @ menu for other projects and files, and chained follow-up prompts against a build that works. Each has a different cost and a different job. The library here carries 245 chained refinement prompts, counted from the data, and one rule changes how you write a goal: Lovable stops asking questions once a message is set as a goal.
By Marco Kohns · Every description re-read in Lovable's docs on 23 October 2026
10hours
The longest Lovable works on one Build message or goal; the last half hour wraps up.
20credits
The default credit check-in: a long message or goal pauses and asks before going on.
245refinements
Chained prompts across the 8 library categories, each run after a working build.
The controls
What makes a Lovable prompt advanced?
Length does not. These five mechanisms do, and each is documented by Lovable.
| Mechanism | What it does | Use it when | Cost or limit |
|---|---|---|---|
| Plan mode | Investigates and writes a plan; changes no code until you approve | A fix missed twice, or a feature needs scoping | 1 credit per message plus subagent research |
| Chat mode | Discusses the open app without editing | You want to understand before deciding | A fraction of a credit, daily chat allowance first |
| Skills | Workspace-level instructions loaded on demand when a request matches, or picked from the slash menu | A task repeats: a release checklist, a content format | Created by workspace owners and admins; Lovable ships built-in ones |
| /goal | A Build message Lovable keeps working on until the goal is achieved | A large task you want finished in one message | Up to 10 hours; pauses at the credit check-in |
| @ references | Reads another project's files, patterns and chat history; the referenced project stays unchanged | Reusing a brand, a component or a solved problem | Read-only on the source project |

Sources, read 23 October 2026: Plan mode, project chat, Skills, goals, cross-project references.
Chaining
How do you chain prompts without breaking what already works?
Run the next prompt against a build that works, name the one change, and say what stays untouched.
Lovable's best practices define a change prompt in three parts: what to build, where it goes, and what must stay untouched. Every Pro variant in the library ends with a numbered Refinement Prompts section written that way, to run after the base build. The counts below come from the data.
| Category | Refinement prompts |
|---|---|
| App Builder | 60 |
| API & Integration | 32 |
| Backend Logic | 30 |
| SaaS App | 30 |
| UI/UX | 28 |
| Landing Page | 25 |
| CRM App | 20 |
| Admin Dashboard | 20 |
Chain
A refinement that holds
- Names the thing that already exists
- Asks for one change
- States the rule that must still hold
- Runs after you checked the last build in the preview
Break
A refinement that regresses
- Re-sends the whole original prompt
- Bundles three features with and
- Leaves the boundary unstated, so Lovable may touch working parts
- Runs before the previous build was checked
Reuse
When do you use a Skill, project knowledge or a goal?
Three mechanisms for three jobs. The docs draw the lines.
Always on
Project knowledge
On demand
Skills
One message
A goal
Goals
How do you turn a library prompt into a goal Lovable can finish alone?
Replace the question line with the answers. Lovable does not pause to ask questions during a goal, so the decisions have to be in the message.
Every library prompt ends by asking Lovable for its questions, because in a normal Build message that saves a correction round: on my runs Lovable asked up to four. A goal works the other way. The goal docs ask for three things in the message: what a finished result looks like, the decisions you care about, and access to other services set up before you send it. The Done When section of a library prompt already describes the finished result, so the conversion is one section.
A library prompt, converted to a goal
Build the team habit tracker described below and keep working until every item under Done When passes. /goal [the library prompt, Context to Safe-Guard Instructions, unchanged] ## Decisions Already Made (instead of Before Building) - Audience: team leads of remote teams of 3 to 20 people - Sign-in: email and password; no social login in this goal - A week starts on Monday in each user's own time zone - Teams are private: nobody sees another team's habits - Stop and report if anything needs a service I have not connected
A goal still pauses at the credit check-in, 20 credits by default, and costs what the work costs. Use it for a foundation you have already scoped; for a first build where you want to steer, keep the question line.
Plan mode
Which advanced prompts does Lovable recommend for debugging?
Its debugging guide ships three, to run in Plan mode so you get analysis before anything changes.
Plan mode
Codebase audit
Plan mode
Performance audit
Build mode
Fragile change
Summarised from Lovable docs, Debug and improve your app, read 23 October 2026; the fix order is in the common mistakes guide.
Meta prompting
What are meta prompting and reverse meta prompting?
Two techniques from Lovable's Prompting Bible, published 16 January 2025, that still hold.
Before you run
Meta prompting
- Ask a model to rewrite your prompt to be more concise and detailed
- Then check the rewrite against the sections a library prompt carries: states, access rules, Done When
- Run the rewrite, not the draft
After you fix
Reverse meta prompting
- When a debugging session ends, ask Lovable to summarise what broke and how it was fixed
- Save that summary as a prompt or to project knowledge
- The next project starts with the lesson
Chains, already written
Disclosure: if you create a Lovable account through links on this page, we may earn a commission. It costs you nothing and does not change what we recommend. Create a free Lovable account
Frequently asked
What counts as an advanced Lovable prompt?
One that uses Lovable beyond a single Build message: Plan mode to scope or debug, a Skill, a goal set with /goal, an @ reference to another project or file, or a chained refinement that names what must stay untouched.
What is a Lovable goal and how long can it run?
A Build message Lovable keeps working on until the goal is achieved, for up to 10 hours, using the last half hour to wrap up. It does not pause to ask questions about the work, stops early only for something only you can provide, and pauses at your credit check-in level, 20 credits by default.
Should I keep the question line when I set a prompt as a goal?
No. A goal does not pause for questions, so replace the Before Building line with the decisions themselves: audience, sign-in method, privacy rules, edge cases. The Done When section of a library prompt already tells Lovable what finished looks like.
What is the difference between a Skill and project knowledge?
Knowledge is always in context and suits rules that apply to everything. A Skill loads on demand when a request matches its description, or when you pick it from the slash menu, and suits one kind of task. Workspace owners and admins create Skills; Lovable also ships built-in ones.
When should I switch from Build mode to Plan mode?
After one or two failed fix attempts. Lovable's debugging guide says repeated blind fixes pile up code that hides the cause, and Plan mode investigates without changing code. A Plan message costs 1 credit plus any subagent research.
Can Lovable read another of my projects?
Yes, with an @ reference in the chat: it can explore the other project's files, read its code and chat history, and reuse assets such as images and fonts. The referenced project stays unchanged.
What we did not publish
- A measured goal run. I have not set a library prompt as a goal yet; the conversion above follows the goal docs.
- Credit costs for Skills, which run inside normal messages and are billed as such.
- The 26 September description of Skills as available on all plans: the 23 October docs did not repeat it, so it is out.
Sources
- Lovable docs: Skills, read 23 October 2026
- Lovable docs: Goals, read 23 October 2026
- Lovable docs: Plan mode and Subagents, read 23 October 2026
- Lovable docs: Cross-project referencing, read 23 October 2026
- Lovable docs: Debug and improve your app, read 23 October 2026
- Lovable blog: The Lovable Prompting Bible, 16 January 2025
- The library's Pro variants, counted at build time
Written by

Marco Kohns
Founder of ProtoBites - Venture Growth Studio
Growth PM at a Silicon Valley scale-up (a16z and General Catalyst backed), ex-Techstars where he consulted 13 early-stage startups, Reforge-trained. Every prompt on this site comes out of shipping ProtoBites' own portfolio products.
Related reading
- every Lovable guide, by build stage · Choose a tool, plan, prompt, build safely, launch.
- the prompting guide · Start here if the first prompt is not structured yet.
- the SaaS prompts whose refinement prompts are quoted above · Fifteen chained briefs.
- app builder prompts · Thirty apps with a build order each.
- the mode picker and Chat actions menu · Screenshots of both.
- common mistakes, priced · What the debugging order avoids.
- Lovable pricing and credits · Plans read today and what one prompt costs in credits.
- AI app ideas for Lovable · Fifteen AI prompts with the rule that keeps each honest.