Clickable symbol / line references in Kimi Code chat panel #2440

Summary

Currently, the Kimi Code chat panel renders inline-code file paths (e.g. tools/example_module.py) as clickable links that open the corresponding file. However, it does not support clicking on a function/method name to jump to its definition or declaration line. This makes it harder to navigate large agent responses that reference many symbols.

Current behavior

• tools/example_module.py → click opens the file. ✅
• build_pipeline() or build_pipeline() at line 52 → click does nothing / is not a link. ❌
• Markdown links such as [build_pipeline()](tools/example_module.py#L52) open the file but do not scroll to line 52. ❌

Desired behavior

When the agent mentions a symbol, the user should be able to click it and VS Code should navigate directly to the symbol's definition (or at least to the correct file + line). For example:

• build_pipeline() in chat → opens tools/example_module.py at line 52.
• DataProcessor.transform() → opens the class file at the method definition (line 31).

Prior art / reference

• Continue.dev already implements this in its chat panel: it uses VS Code's symbol provider (DocumentSymbolProvider) to resolve a function name to a location and creates a clickable link that jumps to the exact line.
• VS Code webviews support postMessage to the extension, which can then execute vscode.commands.executeCommand('vscode.open', Uri, { selection: ... }) or revealTextDocument with a Range.

Suggested implementation approaches

1. Symbol-provider based (preferred)
   When rendering inline-code that looks like an identifier (def foo(), class Bar, foo()), ask the extension host to resolve the symbol via vscode.executeDocumentSymbolProvider or vscode.executeWorkspaceSymbolProvider, then generate a vscode:// or command-backed link that opens the resolved Range.

2. Explicit markdown link support
   Allow the agent (or post-processor) to emit markdown links in the form [funcName](path/to/file.py#L123). Parse the #L123 fragment on the extension side and open the file with the corresponding line selected.

3. Hybrid
   Use approach (2) as a lightweight fallback when approach (1) cannot resolve the symbol unambiguously.

Why this matters

• Agent responses often enumerate multiple functions / classes (e.g. code review, refactoring plans, trace analysis).
• Clicking each symbol saves the user from manually searching (Ctrl+T / Ctrl+Shift+F) and reduces context-switching.
• Competing tools already provide this; adding it improves parity and daily usability.

Environment (where the limitation was observed)

• OS: Windows 10
• VS Code: latest stable
• Extension: moonshot-ai.kimi-code
• Agent: Kimi Code CLI via VS Code extension chat panel

Summary

Currently, the Kimi Code chat panel renders inline-code file paths (e.g. tools/example_module.py) as clickable links that open the corresponding file. However, it does not support clicking on a function/method name to jump to its definition or declaration line. This makes it harder to navigate large agent responses that reference many symbols.

Current behavior

• tools/example_module.py → click opens the file. ✅
• build_pipeline() or build_pipeline() at line 52 → click does nothing / is not a link. ❌
• Markdown links such as [build_pipeline()](tools/example_module.py#L52) open the file but do not scroll to line 52. ❌

Desired behavior

When the agent mentions a symbol, the user should be able to click it and VS Code should navigate directly to the symbol's definition (or at least to the correct file + line). For example:

• build_pipeline() in chat → opens tools/example_module.py at line 52.
• DataProcessor.transform() → opens the class file at the method definition (line 31).

Prior art / reference

• Continue.dev already implements this in its chat panel: it uses VS Code's symbol provider (DocumentSymbolProvider) to resolve a function name to a location and creates a clickable link that jumps to the exact line.
• VS Code webviews support postMessage to the extension, which can then execute vscode.commands.executeCommand('vscode.open', Uri, { selection: ... }) or revealTextDocument with a Range.

Suggested implementation approaches

1. Symbol-provider based (preferred)
   When rendering inline-code that looks like an identifier (def foo(), class Bar, foo()), ask the extension host to resolve the symbol via vscode.executeDocumentSymbolProvider or vscode.executeWorkspaceSymbolProvider, then generate a vscode:// or command-backed link that opens the resolved Range.

2. Explicit markdown link support
   Allow the agent (or post-processor) to emit markdown links in the form [funcName](path/to/file.py#L123). Parse the #L123 fragment on the extension side and open the file with the corresponding line selected.

3. Hybrid
   Use approach (2) as a lightweight fallback when approach (1) cannot resolve the symbol unambiguously.

Why this matters

• Agent responses often enumerate multiple functions / classes (e.g. code review, refactoring plans, trace analysis).
• Clicking each symbol saves the user from manually searching (Ctrl+T / Ctrl+Shift+F) and reduces context-switching.
• Competing tools already provide this; adding it improves parity and daily usability.

Environment (where the limitation was observed)

• OS: Windows 10
• VS Code: latest stable
• Extension: moonshot-ai.kimi-code
• Agent: Kimi Code CLI via VS Code extension chat panel