Upstream model generated file and line linkification #1803

vijayupadya · 2025-11-05T18:39:23Z

Motivation

Move file path linkification upstream to the model so it emits canonical, machine‑parsable Markdown links (path), reducing reliance on heuristic fallback and enabling precise line anchors. Linkification falls back to current component if model does not handle it.

Also ensure consistent behavior across all model families (OpenAI/GPT‑5, Anthropic/Claude, etc.) while eliminating duplicated prompt text.

This is example response with line number linkification:

mjbvz · 2025-11-05T19:38:09Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+
+export class FileLinkificationInstructions extends PromptElement<{}> {
+	render() {
+		return <Tag name='file_linkification'>


Just a gut feeling but this seems pretty verbose and I'm not sure how well models will follow the good/bad method written this way

mjbvz · 2025-11-05T19:39:16Z

src/extension/prompts/node/agent/openAIPrompts.tsx

 				For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.<br />
 				<br />
 				When referring to a filename or symbol in the user's workspace, wrap it in backticks.<br />
+				<FileLinkificationInstructions />


I don't think this should break up the example

mjbvz · 2025-11-05T19:41:04Z

src/extension/prompts/node/agent/openAIPrompts.tsx

 			- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.<br />
 			- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.<br />
 			- File References: When referencing files in your response, always follow the below rules:<br />
 			* Use inline code to make file paths clickable.<br />


We should revisit these instructions too as they duplicate the linkfication stuff. I'd expect to have some base set of linkfication rules and then a (hopefully) small set of additional customizations for each model

mjbvz · 2025-11-05T19:45:05Z

src/extension/linkify/common/modelFilePathLinkifier.ts

+// Example: [src/file.ts](src/file.ts#L10-12) or [src/file.ts](src/file.ts)
+const modelLinkRe = /\[(?<text>[^\]\n]+)\]\((?<target>[^\s)]+)\)/gu;
+
+export class ModelFilePathLinkifier implements IContributedLinkifier {


Let's break up the existing FilePathLinkifier into the real links functionality and the inline code functionality. Maybe just delete the real links stuff from FilePathLinkifier and keep this class. I'd like to avoid the duplication though and make it so we just have one place that handles the markdown file links

… into vijayu/autoGen-links-3

Copilot

Pull Request Overview

This PR moves file path linkification upstream to AI models by instructing them to emit canonical Markdown links with line number anchors ([path](path#L10-12)). A new ModelFilePathLinkifier processes these structured links first, with the existing FilePathLinkifier serving as fallback for other patterns.

Key changes:

New FileLinkificationInstructions component with detailed prompting rules for models
ModelFilePathLinkifier class to parse model-generated links with #L line anchors
Integration across all model-specific prompts (OpenAI, Anthropic, Gemini, xAI, VSC)
Removal of markdown link handling from legacy FilePathLinkifier
Enhanced LinkifyLocationAnchor with title parameter for display paths

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
src/extension/prompts/node/agent/fileLinkificationInstructions.tsx	New shared prompt component with linkification instructions for models
src/extension/prompts/node/agent/openAIPrompts.tsx	Integrated FileLinkificationInstructions, updated GPT-5 examples, removed old inline instructions
src/extension/prompts/node/agent/anthropicPrompts.tsx	Added FileLinkificationInstructions to Claude prompts
src/extension/prompts/node/agent/geminiPrompts.tsx	Added FileLinkificationInstructions to Gemini prompts
src/extension/prompts/node/agent/xAIPrompts.tsx	Added FileLinkificationInstructions to Grok prompts
src/extension/prompts/node/agent/vscModelPrompts.tsx	Added FileLinkificationInstructions to VSC model prompts
src/extension/linkify/common/modelFilePathLinkifier.ts	New linkifier for parsing model-generated markdown links with line anchors
src/extension/linkify/common/filePathLinkifier.ts	Removed markdown link handling, improved directory slash handling
src/extension/linkify/common/linkifyService.ts	Registered ModelFilePathLinkifier before FilePathLinkifier for priority processing
src/extension/linkify/test/node/modelFilePathLinkifier.spec.ts	Test suite for new model linkifier functionality
src/extension/linkify/test/node/util.ts	Enhanced mock services to support directory detection and workspace methods
src/extension/prompt/node/chatParticipantRequestHandler.ts	Updated anchor-to-markdown conversion to use title field and include line numbers in labels

Comments suppressed due to low confidence (4)

src/extension/prompts/node/agent/xAIPrompts.tsx:115

[nitpick] The examples only show file references with backticks but don't demonstrate the line number linkification format introduced by <FileLinkificationInstructions />. For consistency with the GPT-5 prompt (which includes an example like [src/network/httpClient.ts](src/network/httpClient.ts#L42)), consider adding a similar example to help the model understand when and how to use the line number format:

<Tag name='example'>
  The class `Person` is in `src/models/person.ts`.<br />
  The function `calculateTotal` is defined in [lib/utils/math.ts](lib/utils/math.ts#L42).<br />
  You can find the configuration in `config/app.config.json`.
</Tag>

				<Tag name='example'>
					The class `Person` is in `src/models/person.ts`.<br />
					The function `calculateTotal` is defined in `lib/utils/math.ts`.<br />
					You can find the configuration in `config/app.config.json`.
				</Tag>

src/extension/prompts/node/agent/geminiPrompts.tsx:110

[nitpick] The examples only show file references with backticks but don't demonstrate the line number linkification format introduced by <FileLinkificationInstructions />. For consistency with the GPT-5 prompt (which includes an example like [src/network/httpClient.ts](src/network/httpClient.ts#L42)), consider adding a similar example to help the model understand when and how to use the line number format:

<Tag name='example'>
  The class `Person` is in `src/models/person.ts`.<br />
  The function `calculateTotal` is defined in [lib/utils/math.ts](lib/utils/math.ts#L42).<br />
  You can find the configuration in `config/app.config.json`.
</Tag>

				<Tag name='example'>
					The class `Person` is in `src/models/person.ts`.<br />
					The function `calculateTotal` is defined in `lib/utils/math.ts`.<br />
					You can find the configuration in `config/app.config.json`.

src/extension/prompts/node/agent/openAIPrompts.tsx:110

[nitpick] The examples only show file references with backticks but don't demonstrate the line number linkification format introduced by <FileLinkificationInstructions />. For consistency with the GPT-5 prompt (which includes an example like [src/network/httpClient.ts](src/network/httpClient.ts#L42)), consider adding a similar example to help the model understand when and how to use the line number format:

<Tag name='example'>
  The class `Person` is in `src/models/person.ts`.<br />
  The function `calculateTotal` is defined in [lib/utils/math.ts](lib/utils/math.ts#L42).<br />
  You can find the configuration in `config/app.config.json`.
</Tag>

				<Tag name='example'>
					The class `Person` is in `src/models/person.ts`.<br />
					The function `calculateTotal` is defined in `lib/utils/math.ts`.<br />
					You can find the configuration in `config/app.config.json`.
				</Tag>

src/extension/prompt/node/chatParticipantRequestHandler.ts:463

The markdown link format on line 463 appears incorrect. When generating markdown links for Location anchors with line numbers, the line number should be part of the URL fragment (e.g., path#L10), not as a separate title attribute in the markdown link syntax.

Currently: [text](path "title") where text includes #L10 but path doesn't
Expected: [text](path#L10) where the line number is part of the URL

Consider updating to:

if (isLocation(anchor.value)) {
  const lineFragment = `#L${anchor.value.range.start.line + 1}${anchor.value.range.start.line === anchor.value.range.end.line ? '' : `-${anchor.value.range.end.line + 1}`}`;
  path = `${getWorkspaceFileDisplayPath(workspaceService, anchor.value.uri)}${lineFragment}`;
  const label = anchor.title ?? path;
  text = `\`${label}\``;
}

This ensures the URL includes the line fragment so it's linkifiable by the ModelFilePathLinkifier.

	} else if (isLocation(anchor.value)) {
		path = getWorkspaceFileDisplayPath(workspaceService, anchor.value.uri);
		const label = anchor.title ?? `${path}#L${anchor.value.range.start.line + 1}${anchor.value.range.start.line === anchor.value.range.end.line ? '' : `-${anchor.value.range.end.line + 1}`}`;
		text = `\`${label}\``;
	} else if (isSymbolInformation(anchor.value)) {
		path = getWorkspaceFileDisplayPath(workspaceService, anchor.value.location.uri);
		text = `\`${anchor.value.name}\``;
	} else {
		// Unknown anchor type
		return '';
	}

	return `[${text}](${path} ${anchor.title ? `"${anchor.title}"` : ''})`;

src/extension/prompts/node/agent/anthropicPrompts.tsx

src/extension/linkify/common/modelFilePathLinkifier.ts

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

src/extension/linkify/common/modelFilePathLinkifier.ts

mjbvz · 2025-11-14T18:59:13Z

src/extension/linkify/common/filePathLinkifier.ts

+ * foo.ts
 * ```
 */
 export class FilePathLinkifier implements IContributedLinkifier {


Do we still need this class at all? Maybe just delete it and call new ModelFilePathLinkifier FilePathLinkifier instead

It seems like the new linkifier should handle everything that the current one does. If there are any gaps we can fix those instead of keeping the fallback around

Right now ModelFilePathLinkifier handles structured markdown links with anchors texttext, while FilePathLinkifier handles inline code and plain text paths.
They're registered with clear separation of concerns, where FilePathLinkifier acts as a fallback for informal references.

That said, I'm happy to explore merging them if you think that would be better! Let me know your preference and I can make it work either way.

mjbvz · 2025-11-14T19:01:20Z

src/extension/linkify/common/modelFilePathLinkifier.ts

+
+// Matches markdown links where the text is a path and optional #L anchor is present
+// Example: [src/file.ts](src/file.ts#L10-12) or [src/file.ts](src/file.ts)
+const modelLinkRe = /\[(?<text>[^\]\n]+)\]\((?<target>[^\s)]+)\)/gu;


Make sure this can support the case where the path text is code too as this is fairly common and models have likely been trained on a lot of it:

[`file.ts`](file.ts)

Especially true if we get rid of the old linkifier

Yes, works. Added unit test to validate this.

mjbvz · 2025-11-14T19:02:18Z

src/extension/linkify/common/modelFilePathLinkifier.ts

+				continue;
+			}
+
+			const resolved = await this.resolveTarget(parsed.targetPath, workspaceFolders, parsed.preserveDirectorySlash);


Let's try to run these type of calls in parallel instead of awaiting each iteration

Valid point. Updated

mjbvz · 2025-11-14T19:06:00Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+			- `See [path/to/file.ts](path/to/file.ts#L10-12) for the range.`<br />
+			- `Configuration is defined in [path/to/file.ts](path/to/file.ts).` (whole file)<br />
+			When you need a bullet list of references, write a short description before the link, for example:<br />
+			- Await chat view: [path/to/chatQuick.ts](path/to/chatQuick.ts#L142)<br />


Is this the syntax we want or should we support:

[Await chat view](path/to/chatQuick.ts#L142)

That feels more natural if we can get the model to generate these links properly

Agree, updated

mjbvz · 2025-11-14T19:07:18Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+			- Show widget: [path/to/chatQuick.ts](path/to/chatQuick.ts#L321)<br />
+			Examples:<br />
+			❌ `The function is in exampleScript.ts at line 25.`<br />
+			✓ `The function is in [exampleScript.ts](exampleScript.ts#L25).`<br />


Is the ❌ how we provided counter examples to the model? Seems like it's mostly used in our docs, not in the actual prompts

roblourens · 2025-11-14T22:09:56Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+		return <Tag name='file_linkification'>
+			ALWAYS convert file paths to markdown links with 1-based line numbers whenever you cite specific code locations. Use links inside normal sentences, not as the entire answer.<br />
+			Format examples:<br />
+			- `The handler lives in [path/to/file.ts](path/to/file.ts#L10).` (single line)<br />


What's the advantage of this over a path in backticks? We're asking the model to repeat the path twice, it could be long.

In every other path scenario, we ask it to use absolute paths. How do you handle relative paths within multiroot workspaces?

And right now I think we support linkifying many different patterns, does this mean we would no longer support those patterns? (I strongly support simplifying the patterns we handle)

For multi root workspace, currently the first match is selected.

roblourens · 2025-11-14T22:12:10Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+			- `The handler lives in [path/to/file.ts](path/to/file.ts#L10).` (single line)<br />
+			- `See [path/to/file.ts](path/to/file.ts#L10-12) for the range.`<br />
+			- `Configuration is defined in [path/to/file.ts](path/to/file.ts).` (whole file)<br />
+			When you need a bullet list of references, write a short description before the link, for example:<br />


model was generating responses like below, without any description on what those lines represent.

chatListRenderer.ts#857-857
chatListRenderer.ts#1001-1001
chatListRenderer.ts#1477-1477
chatListRenderer.ts#1486-1486

That's interesting. I wonder if all the rules about links caused it to hyper-focus on links specifically and forget what is was actually trying to communicate 😅

roblourens · 2025-11-14T22:16:40Z

src/extension/prompts/node/agent/fileLinkificationInstructions.tsx

+
+export class FileLinkificationInstructions extends PromptElement<{}> {
+	render() {
+		return <Tag name='file_linkification'>


This is a lot, I just don't think all these rules and examples are necessary for a modern model. What you're asking for is extremely simple, I would expect that a decent model can pick it up with just a couple sentences.

We could do it in a scientific way and write some evals to test all models and how far we can trim the rules while still getting good results.

Valid point, I started with simple few sentence prompts, but I got inconsistent results across models. Within the model also it was inconsistent between different files. I will try pruning it offline separately (with evals) and simplify without loosing the intent.

… into vijayu/autoGen-links-3

vijay upadya added 2 commits November 5, 2025 10:21

Upstream model generated file/line Linkification

a3c7e9e

CP feedback updated

cd34558

mjbvz reviewed Nov 5, 2025

View reviewed changes

vijay upadya added 2 commits November 5, 2025 15:07

simplify prompt

a25735f

Optimization and consolidation

490a486

vijayupadya marked this pull request as ready for review November 12, 2025 17:46

vijayupadya requested a review from karthiknadig as a code owner November 12, 2025 17:46

Copilot AI review requested due to automatic review settings November 12, 2025 17:46

vijayupadya requested review from bhavyaus, pwang347 and roblourens as code owners November 12, 2025 17:46

Copilot started reviewing on behalf of vijayupadya November 12, 2025 17:47 View session

vs-code-engineering bot assigned vijayupadya Nov 12, 2025

vs-code-engineering bot added this to the November 2025 milestone Nov 12, 2025

Merge branch 'main' of https://github.com/microsoft/vscode-copilot-chat…

25ca856

… into vijayu/autoGen-links-3

Copilot finished reviewing on behalf of vijayupadya November 12, 2025 17:53

Copilot AI reviewed Nov 12, 2025

View reviewed changes

vijay upadya added 3 commits November 12, 2025 11:00

GCP feedback updates

e7f3c3e

Update prompt snapshot.

94547bc

few prompt tweaks

9059f78

mjbvz reviewed Nov 14, 2025

View reviewed changes

roblourens previously approved these changes Nov 14, 2025

View reviewed changes

PR feedback updates

278f698

vijayupadya dismissed roblourens’s stale review via 278f698 November 14, 2025 23:11

vijay upadya added 3 commits November 14, 2025 17:06

Move fileLinkifiction instruction to correct location

e4869a3

Remove from default

6fd00ab

Merge branch 'main' of https://github.com/microsoft/vscode-copilot-chat…

62d47ce

… into vijayu/autoGen-links-3

Upstream model generated file and line linkification #1803

Are you sure you want to change the base?

Upstream model generated file and line linkification #1803

Conversation

vijayupadya commented Nov 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vijayupadya Nov 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vijayupadya Nov 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

vijayupadya commented Nov 5, 2025 •

edited

Loading

vijayupadya Nov 14, 2025 •

edited

Loading

vijayupadya Nov 14, 2025 •

edited

Loading