test(core): add unit tests for auth, instructions, tool callbacks, live runs, and usage metadata

[ScottMansfield]

Summary

This change ports the remaining unit tests from adk-python v0.5.0..v1.0.0 across auth, flows, and models to adk-js. Upstream adk-js already implements the runtime behavior from this range. This pull request adds 20 unit test cases covering generateAuthEvent preprocessing, instruction provider state-injection bypass, tool callback chains, SequentialAgent.runLive, and streaming usageMetadata.

Files, insertions, and deletions

5 files changed, 779 insertions(+), 2 deletions(-).

File	Insertions	Deletions	Notes
core/test/auth/auth_preprocessor_test.ts	93	0	Tests AUTH_PREPROCESSOR with generateAuthEvent output.
core/test/agents/processors/instructions_llm_request_processor_test.ts	320	0	Ports test_instructions.py state-injection and provider tests.
core/test/agents/functions_test.ts	92	2	Lines 2423-2424 widen invokeToolWithCallbacks to BeforeToolCallback and AfterToolCallback.
core/test/agents/sequential_agent_test.ts	181	0	Tests SequentialAgent.runLive tool injection and execution order.
core/test/models/google_llm_test.ts	93	0	Tests streaming usageMetadata preservation on Gemini.

Ported behavior

• fe5a5494e: AUTH_PREPROCESSOR processes adk_request_credential events produced by generateAuthEvent with snake_case argument keys.
• 9e9dfa747 and 4c4cfb74a: INSTRUCTIONS_LLM_REQUEST_PROCESSOR injects session state into string instructions and skips state injection for sync and async InstructionProvider functions.
• 2cbbf8813: handleFunctionCallsAsync executes beforeToolCallback and afterToolCallback chains in order and short-circuits when a callback returns a response.
• 4188673b0: SequentialAgent.runLive registers task_completed and appends completion instructions on LlmAgent sub-agents before running them sequentially.
• 509db3f9f: Gemini.generateContentAsync preserves usageMetadata on flushed and final aggregated streaming LlmResponse objects.

Interface discrepancies

Member	adk-python	adk-js	Why
generateAuthEvent args	Emits camelCase via model_dump(by_alias=True)	Emits function_call_id and auth_config; AUTH_PREPROCESSOR accepts both casings	Existing adk-js tests pin snake_case keys on generateAuthEvent.
BuiltInCodeExecutionTool	Removed in v1.0.0 (5115474f2)	Kept in core/src/tools/built_in_code_execution_tool.ts alongside BuiltInCodeExecutor	Strictly additive rule preserves existing exports.
GeminiCodeExecutor	Added as deprecated alias in 993f9971b	Not added; adk-js provides BuiltInCodeExecutor	adk-python later removed GeminiCodeExecutor on main.
Live audio transcription	SingleFlow runs client-side AudioTranscriber on transfer (39f78dc28)	Runner.runLive enables native model-side inputAudioTranscription and outputAudioTranscription	AudioTranscriber in adk-js requires an explicit SpeechClientLike instance.
AnthropicLlm / LiteLlm	Streaming usage_metadata updated (509db3f9f)	Omitted from core	adk-js core does not bundle Anthropic or LiteLLM model wrappers.

Reference test accounting

Reference file (adk-python v1.0.0)	Reference tests in range	Ported in adk-js	Status
tests/unittests/flows/llm_flows/test_instructions.py	7	7 + 1 global provider state test	Ported in instructions_llm_request_processor_test.ts.
tests/unittests/flows/llm_flows/test_async_tool_callbacks.py	8 (2 single + 6 parameterized)	8 (2 existing + 6 added)	Ported in functions_test.ts.
tests/unittests/models/test_google_llm.py	1 streaming test	2 streaming usageMetadata tests	Ported in google_llm_test.ts.
tests/unittests/models/test_litellm.py	1 streaming test	0	Not portable: LiteLlm is a Python-only wrapper not in adk-js.
tests/unittests/flows/llm_flows/test_functions.py	1 (test_WorkFlowAgent)	Covered by 3 runLive tests in sequential_agent_test.ts	Ported for SequentialAgent.runLive.
tests/unittests/artifacts/test_artifact_service.py	3	Already in in_memory_artifact_service_test.ts and gcs_artifact_service_test.ts	Covered upstream.
tests/unittests/events/test_event.py	9	Already in event_test.ts	Covered upstream.
tests/unittests/tools/test_langchain_tool.py	2	Already in langchain_tool_test.ts	Covered upstream.

Mutation testing confirmed the new tests fail when production logic is broken (12 failed across all 5 test files when mutating auth_preprocessor.ts, instructions_llm_request_processor.ts, functions.ts, sequential_agent.ts, and streaming_utils.ts).

Verification

• npx vitest run --project unit:core core/test/auth/auth_preprocessor_test.ts core/test/agents/processors/instructions_llm_request_processor_test.ts core/test/agents/functions_test.ts core/test/agents/sequential_agent_test.ts core/test/models/google_llm_test.ts: 5 test files passed, 174 tests passed.
• npm run build: passed.
• npm run ts:check: passed (0 errors).
• npm run lint: passed (0 errors).
• npm run format:check: passed.
• npm run docs:check: passed.

What the reviewer must decide

• Keep BuiltInCodeExecutionTool exported alongside BuiltInCodeExecutor for backward compatibility.
• Keep AnthropicLlm and LiteLlm out of core.
• Keep generateAuthEvent emitting function_call_id and auth_config while AUTH_PREPROCESSOR normalizes both snake_case and camelCase keys.

Manual testing

Executed the unit test suite, mutation checks, build, type checker, linter, formatter check, and TypeDoc check locally.

Checklist

• I have formatted my code using npm run format
• I have linted my code using npm run lint
• I have added unit tests for my changes
• I have updated the documentation (if applicable)
• I have tested my changes manually

Produced by Foundry (task 45509488-c2b3-4c92-8dcf-e9542b114287). Reviewed by an automated
reviewer before upload; still needs human review.

[ScottMansfield]

Scored 2026-10-17 01:59 UTC at acd07af, round 1. Earlier runs are in this comment's edit history.

Foundry merge critic: APPROVE

This change adds unit tests for credential preprocessing, instruction state injection, tool callback chains, live sequential agents, and streaming usage metadata.

Rubric — the default is approve; each rule is a reason a human should read this change first.

[no ] public_api — Touches only test files under core/test/; no exported symbol or production file changes.
[no ] weakened_guarantee — Adds test cases and widens the local invokeToolWithCallbacks helper in core/test/agents/functions_test.ts to accept callback arrays.
[no ] sensitive_surface — Adds a unit test in core/test/auth/auth_preprocessor_test.ts; no production auth or credential code changes.
[no ] release_surface — Modifies no CI workflow, package manifest, lockfile, or build configuration.
[no ] test_removed — Adds 20 new test cases across five test files and leaves all existing tests and assertions intact.
[no ] inexplicable — All five modified files add unit tests for existing core runtime behavior.

Posted by Foundry's merge critic as evidence rather than a request, so nothing here is waiting on you. Advisory only: Foundry has given no approval.