fix(claude): ask for the summary context report so a resumed chat makes no token-count API calls (#43441)

The CLI's default 'full' report calls count_tokens once per category on a
fresh process, so resuming several chats at once burst 20-31 calls each.
'summary' answers from the last response's usage and local estimates.

Tracked in #43104.

Co-authored-by: 김윤진 <yunjin.kim@merakiplace.kr>
This commit is contained in:
Brennan Benson
2026-11-17 00:51:47 -08:00
committed by GitHub
co-authored by 김윤진
parent c073dc2feb
commit e5ddd18a5e
2 changed files with 18 additions and 1 deletions
@@ -268,7 +268,12 @@ export function createClaudeControlSurface(query: Query): ClaudeControlSurface {
initializationResult: () =>
runClaudeControl('initialize', () => query.initializationResult(), null),
getContextUsage: (options) =>
runClaudeControl('get_context_usage', () => query.getContextUsage(), options?.timeoutMs),
runClaudeControl(
'get_context_usage',
// 'full' makes one token-count API call per category on each fresh CLI; CLIs before 2.1.257 ignore detail.
() => query.getContextUsage({ detail: 'summary' }),
options?.timeoutMs
),
renameSession: (title, sessionId, options) => {
const rename = claudeQuerySessionRenamer(query)
return rename
@@ -161,6 +161,18 @@ describe('Claude stream-json connection', () => {
}
)
it('asks for the context report without per-category token-count API calls', async () => {
const scenario = scriptScenario([HOLD_OPEN])
const connection = await open(launchFor(scenario))
await connection.initializationResult()
await connection.getContextUsage()
expect(scenario.readReport().controlRequests).toContainEqual(
expect.objectContaining({
request: { subtype: 'get_context_usage', detail: 'summary' }
})
)
})
it('completes task Stop through the real SDK while semantic reading is paused', async () => {
const scenario = scriptScenario([HOLD_OPEN])
const connection = await open(launchFor(scenario))