[gaurav0107]

Summary

• When Kiro streams a tool call that has no input (a tool with no parameters, or one whose parameters are all optional and the model passed none), the executor sends the tool-call start chunk with function.arguments: "" and nothing after it. Clients reassemble "" and strict OpenAI clients reject the call. OpenAI sends "{}" in this case.
• flush() now calls a new flushEmptyToolCallArgs helper after the buffered-args flush and before the finish chunk. It sends "{}" for every started tool call that has streamed no arguments (missing, "", or whitespace only).
• The fill runs only at the terminal flush, not at messageStopEvent, so a late argument fragment can never be appended after an injected {}.
• flushPendingWrapperToolCalls now records the arguments it emits in toolArgsEmitted. Without that, the fill would append {} to a valid tool_call wrapper payload.

Related Issues

• Closes fix(providers): kiro tool calls with no input stream function.arguments: "" instead of "{}", so strict OpenAI clients reject the call #18718
• Related to fix(providers): tool call with empty arguments is replayed as tool_use.input "" and 400s every later turn #14798 (same "" shape on the Claude translator path)

Target Branch

• Base is the active release/v3.8.x branch (release/v3.8.53), not main
• New feature? Not applicable: this is a bug fix

Validation

• Change type: provider
• Focused tests: every test file that imports the Kiro executor
• eslint + prettier --check on the changed files, and npm run check:open-sse-typecheck
• Reconciled with the current release/v3.8.53 tip
• Production-code change includes new automated tests

Without the fix, the new end-to-end test fails (the reassembled arguments are ""):

✖ Kiro streams {} arguments for a tool call that carries no input (#18718)
ℹ tests 24
ℹ pass 23
ℹ fail 1

With the fix, across the 15 test files that import open-sse/executors/kiro.ts:

$ node --import tsx/esm --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit <15 kiro test files>
ℹ tests 137
ℹ pass 137
ℹ fail 0

$ npm run check:open-sse-typecheck
openSseTypecheckErrors=0
[open-sse-typecheck] OK — 0 pre-existing error(s), all within frozen baseline.

Tests Added Or Updated

• tests/unit/executor-kiro.test.ts: a stream with three tool calls: no input, input: "", and an object input. The first two reassemble to {}, the third is unchanged, and the fill lands before the tool_calls finish chunk. A second case checks that a valid tool_call wrapper does not get {} appended.
• tests/unit/kiro-tool-args-streaming.test.ts: unit tests for flushEmptyToolCallArgs covering missing, empty, and whitespace-only arguments, a call that already streamed arguments, idempotence, and the no-tool-call case.

Coverage Notes

• The open-sse/executors/kiro.ts change is covered by the two test files above. The new helper is exercised directly and through transformEventStreamToSSE, and every branch is covered.

Reviewer Notes

• The fill only applies to calls whose emitted argument string is empty after trimming. Calls that streamed string fragments or a buffered object are untouched.
• Changelog fragment: changelog.d/fixes/18780-kiro-empty-tool-arguments.md.

---

🤖 Opened with Superhuman, an open-source contribution agent.

[coderabbitai]

Important

Review skipped

Auto incremental 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

• Source mode: Auto
• Repository: cheaper-inference/OmniRoute
• Branch: release/v3.8.53
• File: .coderabbit.yaml
• Commit: f000c11
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Advanced
• Run ID: 86d6d36c-1f68-4b89-8a3a-a5dbd6edbadd

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

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts