> For the complete documentation index, see [llms.txt](https://docs.scribeware.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.scribeware.com/into-the-weeds/ai-assistant-mcp/reviewing-ai-changes.md). # Reviewing AI Changes The Changes bar is your undo stack for anything the AI touched — see every edit, jump to it, keep it or roll it back. Every edit the AI makes is recorded, item by item, in the **Changes** bar. Open it from the icon on the right-hand toolbar — the one with the red count badge on it (its tooltip reads *Pending MCP changes*). ## Changes Are Already Applied This is the one thing worth getting straight before anything else. When the AI edits an observation, that edit is **live immediately**. Open the chapter and it's there. The Changes bar is not a queue of things waiting for your permission — it's a list of things that have already happened, each with an undo button attached. So the two buttons on every row mean: | Button | What it does | | ------------- | --------------------------------------------------------------------------------------------------- | | ✅ **Confirm** | You're happy with it. The undo record is cleared and the row disappears. The report doesn't change. | | ❌ **Revert** | Put it back how it was. The change is undone in the report. | Nothing expires or auto-confirms. A change you never look at just sits there with its undo intact. {% hint style="info" %} If you've been reading the AI's own output, you may see it describe a change as *"queued"*. That's the same thing — the change is live, and "queued" means an undo record was stored for you. {% endhint %} ## Finding Your Way Around Changes are grouped the way the app is, so you can scan them fast: * **Chapters** come first, broken down by chapter and then by section, with each field, observation and section change listed underneath. * Below that: **Report Settings**, **Summary Tables**, **Modifiers**, **Library**, the **Settings** groups (tag groups, smart tags, boilerplate, PDF mappings) and the **Gallery**. A few things to look for on a row: * **Click the row itself** to jump straight to that item in the app — the right chapter, scrolled to the right observation, or the right settings tab. This is the fastest way to actually read a change before deciding on it. * **Small blue chips** summarise what changed on that item. * **A padlock** means the change is part of a set that has to be kept or reverted together. * **A red "stale" chip** means the change can no longer be applied or undone — usually because the thing it touched has since been deleted. Those rows only offer **Dismiss**, which clears the row and does nothing else. Every group header has its own confirm and revert buttons, so you can take a whole section, a whole chapter or a whole library category in one click. **Confirm All** and **Revert All** at the top do the lot. The **refresh** icon recalculates everything and drops rows that are no longer relevant — handy after a big AI session, or if the list looks out of date. ## Changes in Other Reports The bar shows the report you have open. If the AI has touched others, you'll see a **View all changes** link listing them by report, with a **Confirm All** button for each. You can confirm from there, but to *revert* a change you have to open that report first — the undo needs the report loaded to put things back properly. ## Auto-Approve At the top of the bar is a checkbox: **Auto-approve report changes**. It's set **per report**, not globally. Tick it and the AI's edits to *that report* stop generating undo records. They just happen, like your own typing does, and the Changes bar stays quiet. That's a genuinely reasonable way to work once you trust the setup — reviewing forty rows of "fixed a typo" gets old. But be clear about the trade: with auto-approve on, the one-click undo is gone. Your safety net becomes the normal one — you noticing, and fixing it yourself. {% hint style="warning" %} Changes that affect your whole company — library categories, tag groups, boilerplate — **always** get an undo record, whatever the auto-approve setting says. Those edits reach every report and every inspector on your team, so they're never made silently. {% endhint %} ## A Sensible Way to Work 1. Leave auto-approve **off** to start with. 2. Give the AI a job, then open the Changes bar and click through a few rows to see what it actually did. 3. Confirm the good ones, revert anything you don't like, and tell the AI why — that's what makes the next batch better (see [Teaching It Your Style](/into-the-weeds/ai-assistant-mcp/teaching-it-your-style.md)). 4. Once a particular kind of job is boringly reliable, turn auto-approve on for the reports where you're doing it. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.scribeware.com/into-the-weeds/ai-assistant-mcp/reviewing-ai-changes.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.