Liberate: Force AI to use always the same wording for data-liberation flows - #4122
Conversation
| 1. **Migrate content into blocks + products** — WordPress-native blocks + navigation + WooCommerce product pages. Best launchpad for a redesign. (Reconstructs product pages.) | ||
| 2. **Theme replication** — carry-and-scope: highest-fidelity replica of the source, raw-HTML-editable (not block-editable). (Imports product *data*; product pages fall back to default WooCommerce, not a carried replica.) | ||
| - question: `How should the site be reconstructed in WordPress?` | ||
| - option label: `Migrate content into blocks + products` |
There was a problem hiding this comment.
I decided to keep the original wording, but WDYT about changing it to just "Blocks", so that we have:
- Blocks
- Theme replication
sejas
left a comment
sejas
left a comment
There was a problem hiding this comment.
I like the intent of this PR but I couldn't verify it as I got a different questions each time.
I executed npm run cli:build && node apps/cli/dist/cli/main.mjs code and then /liberate https://neicss.wixsite.com/sunflower-valley-exp, and I got this questions asked:
First time
After repeating a second time I got the expected questions but the (Recommended) appears cut. Could we change the wording to fit the entire title?
Third time
Yeah, I menthioned it the PR description - "JFYI you can see that we trim label and description. It will be improved in this PR #4119" 4119 is already merged, but this branch is not updated. So in trunk we have the title and description on a separate lines. |
|
Ah, ok. I saw the PR was merged and thought it was already applied to this PR too. |
|
Hm, weird, I just tested and I received the right question: Anyway, it's AI and determinism is not about AI 😁 At least mostly it shows the same question and options, so I think it's still better than always random question and options. |
There was a problem hiding this comment.
The question is consistent, but the options jump between two options. I suggest applying this change to make them more consistent in order and recommended label.
I tested my suggestions in three conversations, and they worked well and consistently for the same site.
| - "I'll extract first so the operator has more data when they decide." → Discovery already gives platform · counts · features — enough to choose. Don't burn extraction to defer the question. | ||
|
|
||
| Show the discovery inventory (pages · archetypes · products · platform features) and a scope/cost/time estimate. Make a **platform-informed recommendation** (mark it `(Recommended)` as the first option), then call **AskUserQuestion** with these two options: | ||
| Show the discovery inventory (pages · archetypes · products · platform features) and a scope/cost/time estimate as a plain chat message, then make a **platform-informed recommendation** and call **AskUserQuestion**. The question, labels, and descriptions below are **fixed copy — use them verbatim**: do not paraphrase, shorten, or restate the inventory inside them. List the recommended option first and append ` (Recommended)` to its label — that is the only allowed variation. |
There was a problem hiding this comment.
| Show the discovery inventory (pages · archetypes · products · platform features) and a scope/cost/time estimate as a plain chat message, then make a **platform-informed recommendation** and call **AskUserQuestion**. The question, labels, and descriptions below are **fixed copy — use them verbatim**: do not paraphrase, shorten, or restate the inventory inside them. List the recommended option first and append ` (Recommended)` to its label — that is the only allowed variation. | |
| Show the discovery inventory (pages · archetypes · products · platform features) and a scope/cost/time estimate as a plain chat message, then call **AskUserQuestion**. The question, labels, and descriptions below are **fixed copy**: the question field must contain EXACTLY this string — nothing prepended or appended (the inventory stays in the chat message, never in the question). Exactly one option is recommended: blocks+products when the site has a store/products, theme replication otherwise. ALWAYS list the recommended option FIRST and append ` (Recommended)` to its label — order and that suffix are the only allowed variations. |
There was a problem hiding this comment.
The problem now is - how AI determines what the recommended option is. So, we should improve this instruction, which is responsible specifically for determining what is the recommended option.
In the proposed changes, I see just duplicating blocks+products when the site has a store/products from there, nothing what can help AI more to understand what is the recommended way
There was a problem hiding this comment.
Sure, feel free to adapt the text as you think it makes more sense. In the end what I would expect is a consistent answer and with those instructions I got something a bit more robust.
| - question: `How should the site be reconstructed in WordPress?` | ||
| - option label: `Migrate content into blocks + products` | ||
| description: `WordPress-native blocks + navigation + WooCommerce product pages. Best launchpad for a redesign. (Reconstructs product pages.)` | ||
| - option label: `Theme replication` |
There was a problem hiding this comment.
If the theme replication is the most common recommendation should we move it to the first option? I think it will help the instructions described above.
486fa07 to
70ce84b
Compare
1984e14 to
f6500b9
Compare
2 participants
Related issues
How AI was used in this PR
AI assisted
Proposed Changes
I noticed that the wording keeps changing, so even internally we’ve started using different names for the same options.
I think we should standardize the names of these flows. It would make the terminology more predictable and help users understand how we expect each flow to be named and described.
JFYI you can see that we trim label and description. It will be improved in this PR #4119
Testing Instructions
I tested, it helps, so a visual review should suffice