AI Error Fixer (VS Code extension)
Watches your integrated terminal and debugger for errors, sends the error +
the relevant file to a GitHub Models AI model, and applies (or previews) a fix.
Plain JavaScript — no build/compile step needed.
1. Get a GitHub token
You need a fine-grained personal access token with the models: read
permission (no other permissions are required):
- https://github.com/settings/personal-access-tokens/new
- Under Permissions → Account permissions, set Models to Read-only.
- Generate the token and copy it (starts with
github_pat_...).
GitHub Models is free to use with normal GitHub accounts, subject to rate limits.
2. Run it
No dependencies to install — this extension only uses the built-in vscode
API and Node's global fetch. Just open this folder in VS Code and press
F5. This opens an "Extension Development Host" window with the extension
loaded — that's where you'll actually test it (open a Python/Node/etc.
project there).
In the Extension Development Host:
- Run command "AI Error Fixer: Set GitHub Token" (Ctrl+Shift+P / Cmd+Shift+P) and paste your token.
- Open a script and run it in the integrated terminal (e.g.
python app.py), or start a debug session.
- If it errors, the extension sends the file + error to the model and:
- shows a diff with Apply Fix / Discard, or
- applies the fix immediately and re-runs the command — if you've turned on auto-apply.
Toggle auto-apply any time with "AI Error Fixer: Toggle Auto-Apply Fixes",
or manually re-trigger a fix on the last captured error with
"AI Error Fixer: Fix Last Captured Error".
Settings (Settings UI → search "AI Error Fixer")
| Setting |
Default |
Description |
aiErrorFixer.autoApply |
false |
Apply fixes without showing a diff first |
aiErrorFixer.model |
openai/gpt-4.1 |
Any model ID from the GitHub Models catalog |
aiErrorFixer.autoRerun |
true |
Re-run the same command after a fix is applied |
aiErrorFixer.watchTerminal |
true |
Watch terminal exit codes |
aiErrorFixer.watchDebugger |
true |
Watch debug sessions for exceptions |
aiErrorFixer.maxAutoAttempts |
3 |
Consecutive auto fix+rerun cycles before pausing to ask you |
You can list available model IDs with:
curl -H "Authorization: Bearer YOUR_TOKEN" https://models.github.ai/catalog/models
How file detection works
The extension parses the error output for File "...", line N (Python) or
(path:line:col) (Node.js) stack frames and edits that file — not just
whatever tab happens to be open. If nothing matches, it falls back to your
currently active editor.
Known limitations / good next steps
- Terminal watching requires VS Code's shell integration (works in the
default integrated terminal with bash/zsh/PowerShell/pwsh; not all shells
support reading command output).
- File detection currently covers Python and Node.js stack-trace formats;
add more regexes in
src/fileResolver.js for other languages (e.g. Java,
Go, Rust) as needed.
- The model edits one file at a time. Multi-file fixes (e.g. error caused by
a bug in an imported helper module from a different file than the
traceback's deepest frame) aren't handled yet.
- There's no diffing against multiple candidate model responses — it's a
single request/response per error. You could add retries with the error
fed back in if the rerun fails again.
Packaging as a .vsix (optional)
If you want to install it permanently rather than always running via F5:
npm install -g @vscode/vsce
vsce package
This produces an .vsix file you can install via the Extensions panel's
"Install from VSIX..." command.
