[BUG] /context reports % against a 200k window on a 1M-context Opus model (false "over 100%" + spurious autocompact warning)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Title: /context reports % against a 200k window on a 1M-context Opus model (false "over 100%" + spurious autocompact warning)

Version: 2.1.153 (Claude Code), model claude-opus-4-8 (Opus 1M / "5× more context")

Summary: /context computes the usage percentage against a 200k denominator even when the active model has a ~1M context window. The meter climbs past 100%
(observed 107%, 213.8k/200k) while the session continues to function normally — which is only possible because the real window is larger than 200k. The
accompanying "Context is N% full / Autocompact will trigger soon" warning fires off the wrong denominator, so it alarms with no real pressure, and
auto-compact (correctly) never triggers.

Expected: /context percentage and the autocompact warning should be computed against the active model's actual context window (~1M for this model), so e.g.
213.8k would read ~21%, not 107%.

Actual: Denominator is pinned at 200k. /context shows 102% → 107% across multiple real model turns; warning persists; no auto-compact (consistent with there
being no real pressure).

Decisive evidence: The session accepts and answers requests at 213.8k tokens with full recall to the start of the conversation. A true 200k-limited model
would reject a 214k-token request, so the real window must exceed 200k — confirming the meter's denominator, not the window, is wrong.

Repro: Run a long Opus-4-8 (1M) session past ~200k tokens; run /context; observe >100% with the session still working and no auto-compaction.

Impact: Cosmetic but misleading — prompts users to /compact and discard history to relieve pressure that doesn't exist.

What Should Happen?

Claude should report the context window properly

Error Messages/Logs

Steps to Reproduce

Just run the context window up past 200k

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.153

Platform

AWS Bedrock

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_