AI Feedback Fixes
Ask your AI to read feedback and fix what it safely can, landing changes in review.
Your AI can read the feedback left on a site and apply fixes using its normal build tools. There are no special fix tools. It reads a pin (the screenshot, the element, and the thread), makes the change the same way it builds anything else, and replies in the thread saying what it did.
This page is about the fix loop. For how feedback pins are created and managed, see Feedback.
What it works on
Your AI only touches items it is cleared to touch. It works Unassigned items and items assigned to BuildPress AI. It leaves human-assigned items alone. If you ask it to fix something assigned to a person, it does not override that assignment. It asks you to reassign the item to BuildPress AI first, then it picks it up.
- Unassigned items: fair game.
- Items assigned to BuildPress AI: fair game.
- Items assigned to a person: left alone. Your AI asks you to reassign them to BuildPress AI before it acts.
How it finishes
After your AI applies a fix, it moves the item to In review so a human can confirm the result. It never marks anything Complete on its own. The only path to Complete is an explicit human instruction. So the lifecycle on a fix looks like Open or In progress, then In review once the AI is done, then Complete only when you say so.
In review is not done
In review means the AI applied a change and is waiting for a human to check it. The item stays in the list until you confirm and mark it Complete, or reopen it with another comment.
When it defers
Your AI does not force a fix when it cannot do one cleanly. It reads two signals on every pin and defers when either one says so.
First, how well the pin can still be located. A pin can be resolved (the exact element is still there), selector_fallback (a near match was found), or detached (the original target is gone). A detached pin is never silently dropped. Your AI tells you the target moved or no longer exists instead of guessing at a different element.
Second, whether the target is something it can edit. Your AI will not touch a theme template, a third-party embed, or another plugin's output. When the target is read-only, it explains why and defers, rather than reaching outside what it safely controls.
- Pin still located cleanly and the target is editable: it applies the fix and moves the item to In review.
- Pin detached: it tells you the original target is gone and asks how to proceed, instead of editing the wrong element.
- Target is a theme template, a third-party embed, or another plugin's output: it explains why it cannot edit there and defers.
Trigger the AI
In v1 the model is pull, not push: you take the first pass and ask your AI to work the feedback. Refer to any item by its number, for example #5. The canonical prompt is below, followed by other useful ones.
Check this site's feedback and fix what you can.Your AI reads the items it is cleared to work, applies each change, replies in the thread noting what it changed, moves those items to In review, skips human-assigned items, and defers on read-only or detached targets.
Show me everything assigned to BuildPress AI. Assign feedback #5 to BuildPress AI and fix it. Fix the high-priority feedback that's still Open. Feedback #12 looks good, mark it Complete.Cloud feedback agent (Soon)
A Cloud feedback agent is on the way. You will trigger BuildPress AI from inside WordPress, and it works the feedback on its own in BuildPress Cloud, with no external MCP client in the loop. Until it ships, you trigger fixes from your own MCP client as described above.