## TL;DR
The Files Changed tab and the list-files API both cap at 3000 files; anything past that is invisible in the browser. Fetch the PR locally and enumerate the full change set with `git diff --name-only`. Then review the files GitHub hid, one directory at a time. Nothing is lost, the web UI just will not show it.

```text
Sorry, we could not display the entire diff because it was too large.
```

1. Check out the PR: `gh pr checkout 123`. Expected: the PR branch is active.
2. Count the changed files: `git diff --name-only main...HEAD | wc -l`. Expected: the true count, for example `4217`, even though the web tab stopped at 3000.
3. List the files the tab hid: `git diff --name-only main...HEAD | tail -n +3001`. Expected: the file paths past the browser cap.
4. Review them in chunks: `git diff main...HEAD -- path/to/dir`. Expected: per-directory hunks.
5. Prioritize the biggest changes: `git diff --numstat main...HEAD | sort -rn | head -20`. Expected: the 20 files with the most added lines, so review effort goes to the largest changes first.

## Use this when
- The Files Changed tab stops listing files at 3000
- The pull request files API caps at 3000 entries
- You need to review files the browser never rendered

## Not for this skill when
- The tab is empty with the too-large message on a modest diff (that is the diff-size cap, a different limit)
- The PR has under 3000 files (the cap is not your problem)

## Variant phrasings
- "github PR only shows 3000 files"
- "files changed tab truncated 3000 file limit"
- "how to see all files in a huge pull request"

## Why it happens
GitHub caps the web file list and the list-files API response at 3000 files to bound page cost. The cap is a display limit, not a data limit; the git history carries every file, so a local clone always sees the whole set.

## Edge cases
- Pagination on the files API will never get you past 3000; it is a hard cap, not a page size.
- `gh pr diff` also fails on oversized diffs; the `git diff` route after checkout is the reliable one.
- If most of the 4000+ files are generated (migrations, snapshots), exclude them first and review the hand-written remainder.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_pgO2sok7aX9DZqxlH2M2iQ