fix(claude): use used_percentage instead of cumulative total tokens

[ahmetb]

Prerequisites

• I have read and understood the contributing guide.
• The commit message follows the conventional commits guidelines.
• Tests for the changes have been added (for bug fixes / features).
• Docs have been added/updated (for bug fixes / features).

Description

TokenUsagePercent() had a fallback path that calculated context window usage from TotalInputTokens + TotalOutputTokens. These are cumulative session totals that never reset on /clear or compact. In long sessions (e.g. 3.7M+ input tokens), this caused the percentage to always show 100% even though actual context usage was only ~20%.

FormattedTokens() had the same fallback and would display inflated cumulative token counts instead of the actual current context size.

This PR removes the fallback to cumulative total tokens from both functions. They now:

1. Use used_percentage / CurrentUsage from Claude's statusline JSON (correctly resets on /clear and compact)
2. Return 0 if neither is available

Cumulative session totals are still accessible via .ContextWindow.TotalInputTokens / .TotalOutputTokens for users who want them in custom templates.

Fixes #7286

[Copilot]

Pull request overview

Fixes the Claude segment’s context-window percentage calculation so long-running sessions don’t incorrectly show 100% usage by avoiding cumulative session totals and preferring Claude’s used_percentage (or current usage).

Changes:

• Updates TokenUsagePercent() to use context_window.used_percentage when available and otherwise compute from CurrentUsage only.
• Removes the previous fallback path that used cumulative TotalInputTokens + TotalOutputTokens for percentage calculation.
• Updates Claude segment tests to reflect the new fallback behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/segments/claude.go	Adjusts token usage percentage logic and removes cumulative-total fallback logic (also affects FormattedTokens() behavior).
src/segments/claude_test.go	Updates unit tests to match the new token usage percent behavior and the updated FormattedTokens() expectations.

Comments suppressed due to low confidence (1)

src/segments/claude.go:159

• PR description says FormattedTokens() should retain a fallback to cumulative TotalInputTokens + TotalOutputTokens, but this change removes that fallback entirely (and returns 0 when CurrentUsage is nil/zero). Please either reintroduce the fallback in FormattedTokens() or update the PR description to match the new behavior (and consider whether this is a breaking change for users relying on cumulative totals).

// FormattedTokens returns a human-readable string of current context tokens.
// Uses CurrentUsage (which represents actual context and resets on compact/clear).
func (c *Claude) FormattedTokens() string {
	var currentTokens int

	if c.ContextWindow.CurrentUsage != nil {
		currentTokens = c.ContextWindow.CurrentUsage.InputTokens +
			c.ContextWindow.CurrentUsage.CacheCreationInputTokens +
			c.ContextWindow.CurrentUsage.CacheReadInputTokens
	}

	if currentTokens < int(thousand) {
		return fmt.Sprintf("%d", currentTokens)
	}

	if currentTokens < int(million) {
		return fmt.Sprintf("%.1fK", float64(currentTokens)/thousand)
	}

	return fmt.Sprintf("%.1fM", float64(currentTokens)/million)

[Copilot]

These test updates assert FormattedTokens() returns "0" when CurrentUsage is nil/zero and no longer validate the cumulative total fallback. This conflicts with the PR description which states FormattedTokens() should keep falling back to TotalInputTokens + TotalOutputTokens. Align the tests with the intended behavior (either restore fallback expectations or update the PR description if the behavior change is intentional).

[ahmetb]

@JanDeDobbeleer please take a look, assuming you have a Claude setup probably easier to verify on your side as well.

[JanDeDobbeleer]

@ahmetb are you sure this wasn't already fixed on latest?

[ahmetb]

Indeed, my bad for not testing HEAD.