[FEATURE] Token usage visibility and threshold hooks in Claude Code

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When using the Claude Code desktop app (Pro/Max plans), there is no visibility into daily token consumption during a session. When the limit is reached, the session simply stops — no warning, no summary, no opportunity to wrap up.

This creates two real problems:

1. Work in progress is lost without any signal it was about to happen.
2. There's no way to act proactively — by the time you notice, it's already too late.

This affects any workflow involving long, multi-step sessions where context has been built up over time (debugging complex issues, scaffolding features, iterative design work). The loss of context mid-session forces a full restart the next day with no record of what was done or what remained.

Proposed Solution

When using the Claude Code desktop app (Pro/Max plans), there is no visibility into daily token consumption during a session. When the limit is reached, the session simply stops — no warning, no summary, no opportunity to wrap up.

This creates two real problems:

1. Work in progress is lost without any signal it was about to happen.
2. There's no way to act proactively — by the time you notice, it's already too late.

This affects any workflow involving long, multi-step sessions where context has been built up over time (debugging complex issues, scaffolding features, iterative design work). The loss of context mid-session forces a full restart the next day with no record of what was done or what remained.

Alternative Solutions

When using the Claude Code desktop app (Pro/Max plans), there is no visibility into daily token consumption during a session. When the limit is reached, the session simply stops — no warning, no summary, no opportunity to wrap up.

This creates two real problems:

1. Work in progress is lost without any signal it was about to happen.
2. There's no way to act proactively — by the time you notice, it's already too late.

This affects any workflow involving long, multi-step sessions where context has been built up over time (debugging complex issues, scaffolding features, iterative design work). The loss of context mid-session forces a full restart the next day with no record of what was done or what remained.

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

Use Case Example

Scenario: multi-hour feature development session

1. I open Claude Code desktop to work on a new feature. The session starts fresh.
2. Over 2–3 hours, Claude reads multiple files, runs commands, writes code, and we iterate on the implementation together.
3. Midway through a complex refactor — context is rich, decisions have been made, multiple files are in flux — the session abruptly stops. Daily token limit reached.
4. No warning was given. No summary was saved. The next day I have to reconstruct what was done, re-read files, and re-explain decisions already made.

With the proposed feature:

At ~80% usage, an alert appears. I have a few minutes to type /compact, ask Claude to write a summary of the session state to disk, or at minimum finish the current task cleanly. When the limit hits, I have a .session-summary.md with completed work, current state, and next steps — ready to paste at the start of the next session.

Additional Context

• Platform: Claude Code desktop app (macOS)
• Plan: Pro / Max (subscription-based — no API key, no external usage monitoring possible)
• The Stop hook exists but cannot solve this: it fires at session end, when no tokens remain for generation
• Similar patterns exist in other tools: VS Code shows token usage in Copilot; Cursor displays context window fill percentage inline
• A threshold hook would be a low-effort, high-impact addition for power users who already use the hooks system