🤖 fix: infer tool failure from error field

[ethanndickson]

Problem

When in Plan Mode, file edit tools (file_edit_*) are disabled via tool policy. However, the model doesn't know about this policy and may still attempt to call these tools*.

When this happens:

1. The AI SDK emits a tool-error event with the error message
2. The error output was { error: "Model tried to call unavailable tool..." }
3. The frontend's hasFailureResult() only checked for success === false
4. Since there was no success field, the tool call showed "completed" status instead of "failed"

This made it appear the edit succeeded when it actually failed, and the error wasn't visible in the dropdown.

Solution

Update hasFailureResult() to infer failure from the presence of an error field, not just explicit success: false:

function hasFailureResult(result: unknown): boolean {
  if (typeof result !== "object" || result === null) return false;
  // Explicit failure
  if ("success" in result && result.success === false) return true;
  // Implicit failure - error field present
  if ("error" in result && result.error) return true;
  return false;
}

This is more defensive and handles any tool error that includes an error field, regardless of whether success is explicitly set.

*This only appears to happen when the context includes usage of those tools. I assume this isn't an easily solvable problem as an agent author, the model sees the tool is disabled, but it also sees it being used dozens of times in the context. How does it know it's really disabled?

---

Generated with mux

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.