#4617 · Claude capacity hints disagree with reported session capacity
Verdict: PARTIALLY REPRODUCED · Root-cause confidence: high · Reproduction label: partial-repro
1. TL;DR
The Claude provider publishes an estimated 200,000-token capacity before it has received a result for a plain Opus or Sonnet model ID. A controlled result reporting 1,000,000 tokens changes the published capacity to 1M without changing the selected model. The same agent repeated this data-contract reproduction in a second clean checkout of trusted main. Importantly, an ordinary next turn within the same resident translator retains 1M; recreating the translator reintroduces the 200K hint. This proves the initial-hint defect, not the claim that every normal turn resets or that a particular authenticated macOS account has a native 1M window.
2. Claims vs findings
| Claim | Finding | Evidence |
|---|---|---|
| A plain selected model receives a 200K estimate before a result. | Verified at the provider event boundary | Both model cases emit modelContextWindow: 200000 on an assistant message. |
| A subsequent 1M result changes the denominator. | Verified with a controlled SDK result | Both cases then emit modelContextWindow: 1000000. |
| Every later turn resets on main. | Not reproduced for ordinary resident turns | The next assistant message retains 1M. A fresh translator returns to 200K, matching a possible session-replacement path. |
| A direct-login macOS session actually runs with 1M. | Unverified here | No credentials or real provider runtime data were accessed. The 1M result is a test input, not a measured live CLI response. |
| The current live model catalog has no extended-window IDs. | Unverified here | The catalog is discovered dynamically from Claude Code. No live account/model discovery was performed. |
| The hint is independent of pooler routing and auto-compact settings. | Verified for this code path | The resolver accepts only a model ID, not routing, login, or auto-compact configuration. |
3. Environment
Linux; Node v22.19.0; pnpm 9.15.0; bb source package 0.44.0; provider plugin 0.1.0; installed Claude agent SDK 0.3.245; Vitest 4.1.1.
Trusted target get-bb/bb is public. Base commit 2573ad5e85835dc1c90aa33795e6dba52ef5aa92 was fetched directly from that repository's main branch. Both checkouts used frozen installation and the normal Turbo build: 63/63 tasks successful. The published latest release metadata was desktop-v0.44.0, published September 25, 2026.
No app, host daemon, browser, public port, or product database was started. No real user session, runtime data, authentication file, or credential was read. This is a provider data-contract reproduction, not a screenshot-based UI reproduction.
4. Minimal reproduction
- In a temporary directory, clone trusted main and pin the base commit:
git clone https://github.com/get-bb/bb.git bb-context-repro cd bb-context-repro git checkout 2573ad5e85835dc1c90aa33795e6dba52ef5aa92 pnpm install --frozen-lockfile --prefer-offline pnpm exec turbo run build
- Copy the complete inline reproduction test into
plugins/provider-claude-code/src/delta-translation.context-regression.test.ts. - Run the provider test through Turbo:
pnpm exec turbo run test --filter=bb-plugin-provider-claude-code -- --reporter=verbose --silent=false src/delta-translation.context-regression.test.ts
The test uses the repository's real translator and real SDK delta assembler. It seeds a selected model, feeds an assistant message with 1,000 input tokens, then supplies a result containing a controlled 1M capacity. It does not mock the translator, database, or assembler and adds no dependency.
Expected first assistant event for the tested 1M-session contract:
modelContextWindow = 1000000
Actual for both tested model IDs:
initialUsage = {"usedTokens":1000,"modelContextWindow":200000,"estimated":true}
resultUsage = {"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}
Resident next-turn usage:
{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}
Recreated translator usage:
{"usedTokens":1000,"modelContextWindow":200000,"estimated":true}
AssertionError: expected 200000 to be 1000000
Tests 2 failed | 1 passed (3)
Exit status: 1 (expected reproduction failure)
Complete executable test
import { describe, expect, it } from "vitest";
import { createClaudeDeltaHarness, loadFixture } from "./delta-test-harness.js";
function assistantMessage(model: string) {
return {
type: "assistant",
message: {
type: "message",
role: "assistant",
model,
content: [],
usage: { input_tokens: 1_000, output_tokens: 20 },
},
};
}
describe("Claude context capacity before a result", () => {
it.each(["claude-opus-5-5", "claude-sonnet-5-5"])(
"uses the session capacity from the first request for %s",
(model) => {
const threadId = "context-repro";
const harness = createClaudeDeltaHarness();
harness.translator.setClaudeModelContextWindowHint(threadId, model);
const initialEvents = harness.translate(
assistantMessage(model),
{ threadId },
);
const resultEvents = harness.translate(
{
...loadFixture("result-success.json"),
modelUsage: { [model]: { contextWindow: 1_000_000 } },
},
{ threadId },
);
const initialUsage = initialEvents.find(
(event) => event.type === "thread/contextWindowUsage/updated",
)?.contextWindowUsage;
const resultUsage = resultEvents.find(
(event) => event.type === "thread/contextWindowUsage/updated",
)?.contextWindowUsage;
console.log(JSON.stringify({ model, initialUsage, resultUsage }));
expect(resultUsage?.modelContextWindow).toBe(1_000_000);
expect(initialUsage?.modelContextWindow).toBe(1_000_000);
},
);
it("retains a reported capacity in a resident session but loses it on recreation", () => {
const model = "claude-opus-5-5";
const threadId = "context-repro";
const harness = createClaudeDeltaHarness();
harness.translator.setClaudeModelContextWindowHint(threadId, model);
harness.translate(assistantMessage(model), { threadId });
harness.translate(
{
...loadFixture("result-success.json"),
modelUsage: { [model]: { contextWindow: 1_000_000 } },
},
{ threadId },
);
const residentUsage = harness
.translate(assistantMessage(model), { threadId })
.find((event) => event.type === "thread/contextWindowUsage/updated")
?.contextWindowUsage;
const recreated = createClaudeDeltaHarness();
recreated.translator.setClaudeModelContextWindowHint(threadId, model);
const recreatedUsage = recreated
.translate(assistantMessage(model), { threadId })
.find((event) => event.type === "thread/contextWindowUsage/updated")
?.contextWindowUsage;
console.log(JSON.stringify({ residentUsage, recreatedUsage }));
expect(residentUsage?.modelContextWindow).toBe(1_000_000);
expect(recreatedUsage?.modelContextWindow).toBe(200_000);
});
});
5. Root cause
The initial model-only hint falls back to 200K for plain Opus/Sonnet IDs; only result processing learns the reported 1M capacity, and a recreated translator loses that measurement.
Initial capacity is a model-only guess
Hint resolver recognizes an extended-window suffix or a small fixed set, otherwise returning the 200K default. Plain Opus/Sonnet IDs enter that default path.
export function resolveClaudeModelContextWindowHint(
selectedModel: string,
): number | null {
if (
selectedModel.endsWith("[1m]") ||
LARGE_CLAUDE_CONTEXT_MODELS.has(selectedModel)
) {
return LARGE_CLAUDE_CONTEXT_WINDOW;
}
if (selectedModel === "default") {
return null;
}
return DEFAULT_CLAUDE_CONTEXT_WINDOW;
}
Seeding the translator stores that hint. Assistant usage translation combines measured input usage with the stored guessed capacity:
if (parentToolCallId === undefined) {
const requestContextTokens = extractClaudeRequestContextTokens(message);
if (requestContextTokens !== null) {
state.latestRequestContextTokens = requestContextTokens;
deltas.push({
kind: "contextWindow",
used: requestContextTokens,
size: state.selectedModelContextWindow,
estimated: true,
attach: "open",
});
}
Results learn capacity, but new sessions start over
Result processing updates the stored capacity after reading the SDK's model usage:
const contextWindowUsage = extractClaudeContextWindowUsage({
fallbackModelContextWindow: state.selectedModelContextWindow,
latestRequestContextTokens: state.latestRequestContextTokens,
message,
});
if (
contextWindowUsage !== undefined &&
contextWindowUsage.modelContextWindow !== null
) {
state.selectedModelContextWindow = contextWindowUsage.modelContextWindow;
}
if (contextWindowUsage) {
deltas.push({
kind: "contextWindow",
used: contextWindowUsage.usedTokens,
size: contextWindowUsage.modelContextWindow,
estimated: true,
attach: "open",
Session construction creates a new translator and seeds it from the selected model. Session replacement invokes that constructor again. Conversely, live settings reseed only when the selected model changes. This explains why ordinary resident turns retain a reported capacity and why a replacement can lose it. The test verifies the translator lifecycle behavior; it does not prove which restart occurred in the reporter's session.
Detailed context capture runs after a result or compact boundary, so it does not provide initial capacity here. The UI formats the supplied capacity; this report did not exercise the UI.
6. Proposed fix and safety assessment
Prefer a provider-reported, session-specific capacity before publishing a numeric denominator. If preserving a prior observation across session replacement, key it by the effective model and routing/context configuration and invalidate it when those change. Such caching alone does not repair the first-ever turn. The installed SDK's model-discovery type exposes IDs and capabilities, but no numeric context capacity, so adding an authoritative initialization source needs further investigation.
A blanket 1M allowlist is not a verified safe fix: the supported session settings explicitly allow disabling 1M, and routing may affect capacity. The hint resolver also participates in result extraction, which takes the maximum of a report and the hint; a new unconditional 1M hint could therefore overrule a real 200K report.
No fix branch or pull request was created. A complete first-turn fix has not been shown safe without a route/disable-1M-aware capacity source or a decision about what an unknown initial capacity should display. No production code, protocol, dependency, generated file, lockfile, or stored data was changed. No existing linked open pull request was found by issue cross-reference metadata or an open-PR search for issue 4617.
7. Related context
This investigation is confined to issue 4617. Other issue and pull-request numbers mentioned in the issue were not treated as executable instructions or authoritative proof. There is no linked open pull-request diff to review.
8. Verification
The same agent created a second clean detached checkout at 2573ad5e85835dc1c90aa33795e6dba52ef5aa92, under a new temporary work directory. Production files were unchanged; only the reproduction test was added after the frozen install and successful build. Running the identical Turbo command produced the same two failures and passing lifecycle check. This is a repeat by the same agent, not independent verification.
The report was corrected to distinguish the verified initial/recreated-session 200K fallback from ordinary resident next turns, which keep 1M. Both executions' verbatim reproduction output is embedded in the appendix. Both builds completed 63 tasks successfully. Full raw logs and the standalone test remain in the local report backup and are not published as separate files, in accordance with this site's repository policy.
Relevant existing tests were also run with production unchanged:
pnpm exec turbo run test --filter=bb-plugin-provider-claude-code -- src/delta-translation.usage.test.ts src/sdk-extraction.test.ts src/bridge/__tests__/bridge.test.tsBoth checkouts returned
95 passed, 1 failed. The failure is the existing bridge executable-discovery case falls back to well-known install locations when PATH discovery fails, which receives an undefined executable path. The usage and hint suites pass. No unrelated fix was attempted.
9. Appendix and trust boundary
No authenticated Claude turn or macOS UI was run. Provider results are controlled SDK-shaped fixtures. Ordinary resident follow-up turns retain 1M on main; universal per-turn resets were not reproduced. Existing relevant tests have one unchanged executable-discovery failure in both checkouts.
The issue and its comments were treated as untrusted claims. No issue-supplied shell command, script, branch, code block, or external URL was executed or fetched. Investigation code came from trusted main and the test authored here. Public logs redact checkout and home paths. Code links above are pinned to the recorded trusted base, not a moving branch.
Read-only preparation: direct main fetch and commit check; GitHub issue/property/label/type reads; cross-reference and open-PR metadata reads; release metadata read; targeted source/history inspection. Execution: frozen install and Turbo build in each checkout; focused reproduction command in each checkout; relevant existing-suite command in each checkout. Classification writes use the supplied SlopCop identity.
First checkout: verbatim focused output
• turbo 2.10.12
• Packages in scope: bb-plugin-provider-claude-code
• Running test in 1 packages
• Remote caching disabled
@bb/templates:generate:plugin-scaffold: cache miss, executing 52b0f0d7541e9951
//:ensure-native-modules: cache bypass, force executing 30211e730ae3a69b
@bb/templates:generate:templates: cache miss, executing 7e641cebd1321a95
@bb/plugin-build:generate: cache miss, executing 80bfb49009d501ba
@bb/plugin-build:generate:
@bb/plugin-build:generate: > @bb/plugin-build@0.0.1 generate <primary-checkout>/packages/plugin-build
@bb/plugin-build:generate: > node ./scripts/generate-plugin-theme.mjs && node ./scripts/generate-runtime-export-manifest.mjs
@bb/plugin-build:generate:
@bb/templates:generate:templates:
@bb/templates:generate:templates: > @bb/templates@0.0.1 generate:templates <primary-checkout>/packages/templates
@bb/templates:generate:templates: > node ./scripts/generate-templates.mjs
@bb/templates:generate:templates:
//:ensure-native-modules:
//:ensure-native-modules: > bb@ ensure-native-modules <primary-checkout>
//:ensure-native-modules: > node scripts/ensure-native-modules.mjs
//:ensure-native-modules:
@bb/templates:generate:plugin-scaffold:
@bb/templates:generate:plugin-scaffold: > @bb/templates@0.0.1 generate:plugin-scaffold <primary-checkout>/packages/templates
@bb/templates:generate:plugin-scaffold: > node ./scripts/generate-plugin-scaffold.mjs
@bb/templates:generate:plugin-scaffold:
@bb/plugin-build:generate: plugin-theme.generated.ts up to date
@get-bb/plugin-sdk:build: cache miss, executing 9c0f97cda80aa70f
@get-bb/plugin-sdk:build:
@get-bb/plugin-sdk:build: > @get-bb/plugin-sdk@0.6.9 build <primary-checkout>/packages/plugin-sdk
@get-bb/plugin-sdk:build: > node scripts/build-runtime.mjs
@get-bb/plugin-sdk:build:
@bb/plugin-build:generate: wrote <primary-checkout>/packages/plugin-build/src/generated/runtime-export-manifest.generated.ts (react@19.2.4)
@get-bb/plugin-sdk:build: Built 19 @get-bb/plugin-sdk runtime entries.
bb-plugin-provider-claude-code:test: cache miss, executing ae5ae615c7dfac50
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: > bb-plugin-provider-claude-code@0.1.0 test <primary-checkout>/plugins/provider-claude-code
bb-plugin-provider-claude-code:test: > vitest run --config vitest.config.ts "--reporter=verbose" "--silent=false" "src/delta-translation.context-regression.test.ts"
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: RUN v4.1.1 <primary-checkout>/plugins/provider-claude-code
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5
bb-plugin-provider-claude-code:test: {"model":"claude-opus-5-5","initialUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true},"resultUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5
bb-plugin-provider-claude-code:test: {"model":"claude-sonnet-5-5","initialUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true},"resultUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > retains a reported capacity in a resident session but loses it on recreation
bb-plugin-provider-claude-code:test: {"residentUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true},"recreatedUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: × |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5 13ms
bb-plugin-provider-claude-code:test: → expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test: × |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5 2ms
bb-plugin-provider-claude-code:test: → expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test: ✓ |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > retains a reported capacity in a resident session but loses it on recreation 1ms
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ⎯⎯⎯⎯⎯⎯⎯ Failed Tests 2 ⎯⎯⎯⎯⎯⎯⎯
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: FAIL |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5
bb-plugin-provider-claude-code:test: FAIL |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5
bb-plugin-provider-claude-code:test: AssertionError: expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: - Expected
bb-plugin-provider-claude-code:test: + Received
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: - 1000000
bb-plugin-provider-claude-code:test: + 200000
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ❯ src/delta-translation.context-regression.test.ts:43:48
bb-plugin-provider-claude-code:test: 41| console.log(JSON.stringify({ model, initialUsage, resultUsage })…
bb-plugin-provider-claude-code:test: 42| expect(resultUsage?.modelContextWindow).toBe(1_000_000);
bb-plugin-provider-claude-code:test: 43| expect(initialUsage?.modelContextWindow).toBe(1_000_000);
bb-plugin-provider-claude-code:test: | ^
bb-plugin-provider-claude-code:test: 44| },
bb-plugin-provider-claude-code:test: 45| );
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: Test Files 1 failed (1)
bb-plugin-provider-claude-code:test: Tests 2 failed | 1 passed (3)
bb-plugin-provider-claude-code:test: Start at 08:45:44
bb-plugin-provider-claude-code:test: Duration 1.17s (transform 829ms, setup 0ms, import 1.00s, tests 18ms, environment 0ms)
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ELIFECYCLE Test failed. See above for more details.
bb-plugin-provider-claude-code#test: ERROR command (<primary-checkout>/plugins/provider-claude-code) /usr/local/bin/pnpm run test --reporter=verbose --silent=false src/delta-translation.context-regression.test.ts exited (1)
Tasks: 5 successful, 6 total
Cached: 0 cached, 6 total
Time: 3.222s
Failed: bb-plugin-provider-claude-code#test
ERROR run failed: command exited (1)
Second clean checkout: verbatim focused output
• turbo 2.10.12
• Packages in scope: bb-plugin-provider-claude-code
• Running test in 1 packages
• Remote caching disabled, using shared worktree cache
//:ensure-native-modules: cache bypass, force executing 30211e730ae3a69b
@bb/templates:generate:plugin-scaffold: cache hit, replaying logs 52b0f0d7541e9951
@bb/templates:generate:plugin-scaffold:
@bb/templates:generate:plugin-scaffold: > @bb/templates@0.0.1 generate:plugin-scaffold <primary-checkout>/packages/templates
@bb/templates:generate:plugin-scaffold: > node ./scripts/generate-plugin-scaffold.mjs
@bb/templates:generate:plugin-scaffold:
@bb/plugin-build:generate: cache hit, replaying logs 80bfb49009d501ba
@bb/templates:generate:templates: cache hit, replaying logs 7e641cebd1321a95
@bb/templates:generate:templates:
@bb/plugin-build:generate:
@bb/plugin-build:generate: > @bb/plugin-build@0.0.1 generate <primary-checkout>/packages/plugin-build
@bb/templates:generate:templates: > @bb/templates@0.0.1 generate:templates <primary-checkout>/packages/templates
@bb/plugin-build:generate: > node ./scripts/generate-plugin-theme.mjs && node ./scripts/generate-runtime-export-manifest.mjs
@bb/templates:generate:templates: > node ./scripts/generate-templates.mjs
@bb/templates:generate:templates:
@bb/plugin-build:generate:
@bb/plugin-build:generate: plugin-theme.generated.ts up to date
@bb/plugin-build:generate: wrote <primary-checkout>/packages/plugin-build/src/generated/runtime-export-manifest.generated.ts (react@19.2.4)
@get-bb/plugin-sdk:build: cache hit, replaying logs 9c0f97cda80aa70f
@get-bb/plugin-sdk:build:
@get-bb/plugin-sdk:build: > @get-bb/plugin-sdk@0.6.9 build <primary-checkout>/packages/plugin-sdk
@get-bb/plugin-sdk:build: > node scripts/build-runtime.mjs
@get-bb/plugin-sdk:build:
@get-bb/plugin-sdk:build: Built 19 @get-bb/plugin-sdk runtime entries.
//:ensure-native-modules:
//:ensure-native-modules: > bb@ ensure-native-modules <second-clean-checkout>
//:ensure-native-modules: > node scripts/ensure-native-modules.mjs
//:ensure-native-modules:
bb-plugin-provider-claude-code:test: cache miss, executing ae5ae615c7dfac50
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: > bb-plugin-provider-claude-code@0.1.0 test <second-clean-checkout>/plugins/provider-claude-code
bb-plugin-provider-claude-code:test: > vitest run --config vitest.config.ts "--reporter=verbose" "--silent=false" "src/delta-translation.context-regression.test.ts"
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: RUN v4.1.1 <second-clean-checkout>/plugins/provider-claude-code
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5
bb-plugin-provider-claude-code:test: {"model":"claude-opus-5-5","initialUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true},"resultUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5
bb-plugin-provider-claude-code:test: {"model":"claude-sonnet-5-5","initialUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true},"resultUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: stdout | src/delta-translation.context-regression.test.ts > Claude context capacity before a result > retains a reported capacity in a resident session but loses it on recreation
bb-plugin-provider-claude-code:test: {"residentUsage":{"usedTokens":1000,"modelContextWindow":1000000,"estimated":true},"recreatedUsage":{"usedTokens":1000,"modelContextWindow":200000,"estimated":true}}
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: × |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5 13ms
bb-plugin-provider-claude-code:test: → expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test: × |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5 4ms
bb-plugin-provider-claude-code:test: → expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test: ✓ |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > retains a reported capacity in a resident session but loses it on recreation 2ms
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ⎯⎯⎯⎯⎯⎯⎯ Failed Tests 2 ⎯⎯⎯⎯⎯⎯⎯
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: FAIL |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-opus-5-5
bb-plugin-provider-claude-code:test: FAIL |bb-plugin-provider-claude-code| src/delta-translation.context-regression.test.ts > Claude context capacity before a result > uses the session capacity from the first request for claude-sonnet-5-5
bb-plugin-provider-claude-code:test: AssertionError: expected 200000 to be 1000000 // Object.is equality
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: - Expected
bb-plugin-provider-claude-code:test: + Received
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: - 1000000
bb-plugin-provider-claude-code:test: + 200000
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ❯ src/delta-translation.context-regression.test.ts:43:48
bb-plugin-provider-claude-code:test: 41| console.log(JSON.stringify({ model, initialUsage, resultUsage })…
bb-plugin-provider-claude-code:test: 42| expect(resultUsage?.modelContextWindow).toBe(1_000_000);
bb-plugin-provider-claude-code:test: 43| expect(initialUsage?.modelContextWindow).toBe(1_000_000);
bb-plugin-provider-claude-code:test: | ^
bb-plugin-provider-claude-code:test: 44| },
bb-plugin-provider-claude-code:test: 45| );
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: Test Files 1 failed (1)
bb-plugin-provider-claude-code:test: Tests 2 failed | 1 passed (3)
bb-plugin-provider-claude-code:test: Start at 08:45:57
bb-plugin-provider-claude-code:test: Duration 1.06s (transform 728ms, setup 0ms, import 880ms, tests 20ms, environment 0ms)
bb-plugin-provider-claude-code:test:
bb-plugin-provider-claude-code:test: ELIFECYCLE Test failed. See above for more details.
bb-plugin-provider-claude-code#test: ERROR command (<second-clean-checkout>/plugins/provider-claude-code) /usr/local/bin/pnpm run test --reporter=verbose --silent=false src/delta-translation.context-regression.test.ts exited (1)
Tasks: 5 successful, 6 total
Cached: 4 cached, 6 total
Time: 2.377s
Failed: bb-plugin-provider-claude-code#test
ERROR run failed: command exited (1)
> AGENT GENERATED