Description
VS Code offers a “Fix using Copilot” link in the hover UI for diagnostics. When selected, this triggers an in-editor chat session with the /fix command, invoking an AI-driven code fix based on the diagnostic.
However, extension authors who contribute diagnostics via languages.createDiagnosticCollection() currently have no way to enrich that fix flow with domain-specific context - e.g. remediation hints, surrounding logs, links to documentation, or synthetic analysis results. This often limits the quality and relevance of AI-generated fixes, especially for domain-specific tooling (e.g. linters, policy checkers, custom analysis engines).
Proposal
Introduce a new optional hook for diagnostic collections:
provideFixInformation?: (diagnostic: Diagnostic) => ProviderResult<string | undefined>;
This could be added either:
- as a method on the
DiagnosticCollection, or - as part of a new options object to
languages.createDiagnosticCollection(name: string, options?: {...}).
The returned string would be included in the AI model prompt when fixing that diagnostic—providing structured, dynamic, and extension-specific guidance.
Use Cases
- A security extension detects a risky pattern and adds an explanation of why it’s insecure and how to safely rewrite it.
- A custom DSL linter includes auto-generated fix rationale or references to the project’s style guide.
- An extension analyzing build errors includes metadata like full macro expansions or evaluated conditions that don’t appear directly in the source.
Design Considerations
- This does not prescribe a specific UX or interaction model - how the Copilot chat or any agent uses the string is up to them.
- The term “model” is intentionally used here instead of “agent” because it is unclear whether
/fixflows operate in agent mode or not. - This hook could also be reused in future diagnostic flows that involve automated analysis or remediation, beyond just
/fix. - Whenever pertinent, the same pattern could be added to other "editor overlay providers"