GitHub Copilot `file_search` truncation can cause false “file does not exist” conclusions

GitHub Copilot `file_search` result truncation consistently trips up models. **This is a recurring problem and the results misguide models.** A model such as GPT-5.6 Luna is completely capable of a simple task of loading the context window with relevant context. However if there are many files, the `file_search` tool just truncates results and **does not provide sufficient notice to the model that results were truncated and that more results are available**. Instead, the model comes away with the impression that some of the files it was looking for does not exist.

Below is the explanation in the words of GPT-5.6 Sol:

# `file_search` truncation is not reported explicitly enough

## Description

A prompt requested loading several project-context artifacts, including an active work item's description, while skipping targets that do not exist.

I searched the work item's directory using a recursive glob:

```text
path/to/work-item/**
```

I omitted `maxResults`, expecting the returned file list to be sufficient for determining which requested artifacts existed.

The tool reported the total number of matches, displayed only a subset, and ended the output with `...`. Because truncation was indicated only by this unstructured ellipsis, the omitted description file was incorrectly treated as absent.

A targeted search or a sufficiently large `maxResults` returns the file correctly, so this is not an indexing failure. It is a result-reporting problem: callers can easily mistake a partial result set for a complete inventory.

## Steps to Reproduce

1. Create a directory containing more files than `file_search` displays by default.
2. Place the target file outside the subset selected for display.
3. Call `file_search` with a recursive glob and omit `maxResults`.
4. Observe that the tool reports the total match count, displays only some paths, and appends `...`.
5. Observe that the response provides no structured truncation flag, displayed-result count, or continuation mechanism.

## Actual Behavior

Truncation is represented only by a trailing ellipsis. A matching file omitted from the displayed subset can therefore be mistaken for a nonexistent file.

## Expected Behavior

The response should explicitly state that it is truncated, for example:

```text
Showing 20 of 24 results. Results truncated; narrow the query or increase maxResults.
```

Ideally the tool result would also expose structured metadata such as:

```json
{
"totalResults": 24,
"returnedResults": 20,
"truncated": true
}
```

This would let agents reliably distinguish “no match exists” from “the match may have been omitted.”

## Environment

```
Version: 1.131.0 (user setup)
Commit: e4c7e7b1d6d060162f4aa7f8225271b67ce1df75
Date: 2026-07-28T10:51:25Z
Electron: 42.7.0
ElectronBuildId: 14710161
Chromium: 148.0.7778.280
Node.js: 24.18.0
V8: 14.8.178.38-electron.0
OS: Windows_NT x64 10.0.26200
```