fix: make Claude 200k context window selection take effect

[MaxHTu]

What Changed

Selecting the 200k context window for Claude models now actually runs the session at 200k. The adapter injects env: { CLAUDE_CODE_DISABLE_1M_CONTEXT: "1" } into the SDK's flag-settings layer whenever the resolved context window is 200k, restates or clears it on mid-thread model switches via applyFlagSettings, and applies the same opt-out in ClaudeTextGeneration (commit-message / PR-content / thread-title generation, which spawns the same CLI).

Why

Fixes #8405. Claude Code auto-enables the 1M window for claude-opus-5 / claude-fable-5 / claude-sonnet-5 unless the process opts out, so resolveClaudeApiModelId's bare slug does not mean 200k — the 200k selection was a silent no-op and every such session ran at 1M (verified empirically via modelUsage[].contextWindow; details in the issue).

Three design points, each verified against SDK 0.3.170 + CLI 2.1.247:

• Flag settings, not process env: sessions spawn with settingSources: ["user", "project", "local"], and an env block in a user/project settings file overrides the spawned process env — a queryOptions.env-based fix is silently defeated by e.g. "env": {"CLAUDE_CODE_DISABLE_1M_CONTEXT": "0"} in ~/.claude/settings.json. Flag settings (the SDK settings option) outrank those files.
• Mid-thread switches: the opt-out outranks an explicit [1m] model suffix, so without restating it on setModel, a session started at 200k (any default Sonnet session) would silently clamp a later 1M selection to 200k. sendTurn now calls applyFlagSettings alongside setModel, passing null to clear when the new selection isn't 200k.
• Same bug in text generation: ClaudeTextGeneration spawns the same CLI with the same bare slug and already builds a --settings object, so it gets the same conditional spread.

This also makes the context meter agree with reality (#5286): selectedClaudeContextWindow seeds 200k and the CLI now confirms it instead of reporting 1M.

Testing: 7 new / 2 updated unit tests across ClaudeAdapter.test.ts and ClaudeTextGeneration.test.ts; full src/provider + src/textGeneration suites pass (691 passed, 8 skipped); typecheck clean. Behavior verified end-to-end through the real SDK: bare slug → 1,000,000; with the flag-settings opt-out → 200,000 (including winning against a conflicting user-settings env block); mid-session applyFlagSettings flip verified in both directions.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes (no UI changes)
• I included a video for animation/interaction changes (no animation changes)

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes how Claude sessions are configured (context window and live model switches); wrong flag logic would silently run at 1M or clamp 1M selections, but behavior is narrowly scoped and well-tested.

Overview
Fixes 200k context window selections being ignored: Claude Code auto-enables 1M for supported models, so the adapter now sets CLAUDE_CODE_DISABLE_1M_CONTEXT through the SDK settings / flag-settings layer (not process env) when resolveClaudeContextWindow is 200k, including default-200k models like Sonnet 5.

On mid-thread model changes, sendTurn calls applyFlagSettings before setModel—setting the opt-out for 200k or env: null to clear it—so the flag stays in sync with the model id and can override an explicit [1m] suffix.

The same conditional is applied in ClaudeTextGeneration for CLI-spawned flows (commit messages, etc.). Tests cover start-session settings, merges with fast mode, and flag updates when switching between 1M and 200k.

^{Reviewed by Cursor Bugbot for commit 87eba2d. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Fix Claude 200k context window by setting CLAUDE_CODE_DISABLE_1M_CONTEXT flag

When a user selects the 200k context window, the SDK and CLI now set env: { CLAUDE_CODE_DISABLE_1M_CONTEXT: "1" } so the 1M context is actually disabled.

• makeClaudeAdapter injects the flag into createQuery settings and calls query.applyFlagSettings on model switches in sendTurn, clearing it (env: null) when the new model is not 200k.
• makeClaudeTextGeneration adds the same env flag to the CLI --settings payload for 200k selections.
• ClaudeQueryRuntime interface now exposes applyFlagSettings typed from the SDK's Query.

^{Macroscope summarized 87eba2d.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 38fd0cb7-3d62-400f-b2af-6628ead132d4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This change makes the existing Claude 200k selection alter the effective context capacity across live sessions and CLI-based text generation, including model switches. The implementation is focused and tested, but the production runtime impact of this propagated SDK/CLI configuration merits human review.

^{You can add or adjust custom eligibility rules. Learn more.}