fix: preserve LM Studio Responses tool arguments by 849261680 · Pull Request #90593 · openclaw/openclaw

849261680 · 2026-06-05T08:19:24Z

Summary

Preserve streamed OpenAI Responses tool-call arguments when the final response.function_call_arguments.done event omits arguments or sends an empty string.
Add shared parser regression coverage for both LM Studio-style done event shapes.

Verification

node scripts/run-vitest.mjs src/llm/providers/openai-responses-shared.test.ts
node scripts/run-vitest.mjs extensions/lmstudio/src/stream.test.ts
./node_modules/.bin/oxfmt --check --threads=1 src/llm/providers/openai-responses-shared.ts src/llm/providers/openai-responses-shared.test.ts
git diff --check
.agents/skills/autoreview/scripts/autoreview --mode local -> clean, no accepted/actionable findings

Real behavior proof

Behavior addressed: LM Studio/OpenAI-compatible Responses streams can emit function-call argument deltas, then a response.function_call_arguments.done event whose arguments field is missing or empty. The parser now keeps the accumulated delta JSON instead of replacing it with an empty/missing final payload.

Real environment tested: Local OpenAI-compatible HTTP/SSE fixture exercised through streamSimpleOpenAIResponses, using a real POST to /responses and streamed SSE events matching the LM Studio failure shape. Local LM Studio itself was checked at http://127.0.0.1:1234/v1/models, but no LM Studio server was running on this machine.

Exact steps or command run after this patch: Ran a one-shot node --import tsx --input-type=module script that started an HTTP server, returned Responses SSE events with response.function_call_arguments.delta followed by response.function_call_arguments.done without arguments, and consumed it through streamSimpleOpenAIResponses.

Evidence after fix: Copied terminal output from the local HTTP/SSE fixture run:

{
  "ok": true,
  "eventTypes": [
    "start",
    "toolcall_start",
    "toolcall_delta",
    "toolcall_end",
    "done"
  ],
  "stopReason": "toolUse",
  "toolCall": {
    "type": "toolCall",
    "id": "call_read|fc_read",
    "name": "read",
    "arguments": {
      "path": "docs/gateway/local-models.md"
    }
  },
  "requestPath": "/responses",
  "requestStream": true
}

Observed result after fix: The streamed read tool call retained its path argument and the provider stream ended with stopReason: "toolUse".

What was not tested: A live LM Studio model run was not possible because 127.0.0.1:1234 was not reachable in this environment.

clawsweeper · 2026-06-05T08:22:30Z

Codex review: needs maintainer review before merge. Reviewed June 7, 2026, 3:12 AM ET / 07:12 UTC.

Summary
The PR preserves streamed OpenAI Responses tool-call arguments when the final function-call-arguments done event omits or sends empty arguments, and adds regression coverage for both shapes.

PR surface: Source +7, Tests +106. Total +113 across 2 files.

Reproducibility: yes. from source inspection: current main overwrites buffered function-call argument deltas with the final done.arguments value, so missing or empty done payloads can drop arguments or error. I did not run a live LM Studio repro in this read-only review.

Review metrics: none identified.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🐚 platinum hermit
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

[P2] Run one live LM Studio tool-call smoke before merge if maintainers want provider-specific confidence beyond the HTTP/SSE fixture.

Risk before merge

[P1] The PR proof exercises a real local HTTP/SSE fixture through streamSimpleOpenAIResponses, but not a live LM Studio model; maintainers may still want one live LM Studio tool-call smoke before landing.
[P1] The touched parser is shared by OpenAI Responses, Azure Responses, ChatGPT/Codex Responses mapping, and OpenAI-compatible proxy routes, so focused regression tests should be paired with normal CI before merge.

Maintainer options:

Decide the mitigation before merge
Land this narrow shared parser fix after maintainer review and CI, with an optional live LM Studio smoke if maintainers want provider-specific confirmation.
Pause or close
Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

No repair job is needed because the review found no concrete patch defect; the remaining action is normal maintainer review, CI, and optional live LM Studio smoke.

Security
Cleared: The diff only changes shared stream parsing logic and colocated tests, with no dependency, workflow, secret, install, or supply-chain surface changes.

Review details

Best possible solution:

Land this narrow shared parser fix after maintainer review and CI, with an optional live LM Studio smoke if maintainers want provider-specific confirmation.

Do we have a high-confidence way to reproduce the issue?

Yes from source inspection: current main overwrites buffered function-call argument deltas with the final done.arguments value, so missing or empty done payloads can drop arguments or error. I did not run a live LM Studio repro in this read-only review.

Is this the best way to solve the issue?

Yes: fixing the shared Responses stream parser is the narrow owner-boundary solution, while LM Studio-specific repair would duplicate generic OpenAI-compatible stream handling.

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against f8db47e34045.

Label changes

Label changes:

add P1: The PR addresses a shipped regression where LM Studio-backed agents can lose required tool-call arguments, breaking real tool workflows.
add proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes after-fix terminal output from a local OpenAI-compatible HTTP/SSE server exercised through streamSimpleOpenAIResponses, which directly shows the retained tool-call path argument.

Label justifications:

P1: The PR addresses a shipped regression where LM Studio-backed agents can lose required tool-call arguments, breaking real tool workflows.
rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🐚 platinum hermit and patch quality is 🦞 diamond lobster.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes after-fix terminal output from a local OpenAI-compatible HTTP/SSE server exercised through streamSimpleOpenAIResponses, which directly shows the retained tool-call path argument.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes after-fix terminal output from a local OpenAI-compatible HTTP/SSE server exercised through streamSimpleOpenAIResponses, which directly shows the retained tool-call path argument.

Evidence reviewed

PR surface:

Source +7, Tests +106. Total +113 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	11	4	+7
Tests	1	108	2	+106
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	119	6	+113

What I checked:

Current main still overwrites streamed arguments on done: On current main, processResponsesStream assigns event.arguments to currentBlock.partialJson and then calls event.arguments.startsWith(previousPartialJson), so a missing done.arguments can throw and an empty done.arguments can erase buffered delta JSON. (src/llm/providers/openai-responses-shared.ts:667, f8db47e34045)
PR guards the done payload before replacing the buffer: The patch converts done arguments to a string only when present, updates partialJson only for non-empty final arguments or an already-empty buffer, and only emits a synthetic delta when the final string extends the previous buffer. (src/llm/providers/openai-responses-shared.ts:667, 85f7f3a487a3)
Regression tests cover both LM Studio-style done shapes: The PR adds processResponsesStream tests for a done event that omits arguments and one that sends an empty string, both preserving the streamed read tool path argument. (src/llm/providers/openai-responses-shared.test.ts:355, 85f7f3a487a3)
Runtime path reaches the shared parser: runResponsesStreamLifecycle calls processResponsesStream, streamOpenAIResponses uses that lifecycle, and LM Studio wraps the registered streamSimple provider, so LM Studio models configured with openai-responses hit this shared parser. (src/llm/providers/openai-responses-shared.ts:503, f8db47e34045)
Dependency contract checked: OpenAI SDK 6.39.1 types declare ResponseFunctionCallArgumentsDoneEvent.arguments as a string, so the PR is deliberately tolerating OpenAI-compatible proxy divergence without changing the normal full-arguments done path. (package.json:1900, f8db47e34045)
History and ownership provenance: The current shared parser lines are blamed to a1f1895, while git log -S shows the older Responses function-call streaming path traces back through 4798e12. (src/llm/providers/openai-responses-shared.ts:656, a1f1895b1bf8)

Likely related people:

wsyjh8: Blame shows the current shared OpenAI Responses parser and its adjacent tests were added in a1f1895, which now owns the changed lines on main. (role: recent area contributor; confidence: high; commits: a1f1895b1bf8; files: src/llm/providers/openai-responses-shared.ts, src/llm/providers/openai-responses-shared.test.ts)
Vincent Koc: git log -S ties the earlier OpenAI Responses function-call streaming path to the provider transport adapter seam commit, which is relevant background for this parser behavior. (role: feature-history contributor; confidence: medium; commits: 4798e125f40c; files: src/agents/openai-transport-stream.ts)
Krasimir Kralev: Recent history shows adjacent OpenAI transport compatibility work in c45295c, making this a useful routing candidate for parser/runtime compatibility review. (role: recent adjacent contributor; confidence: medium; commits: c45295cc336e; files: src/agents/openai-transport-stream.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

ceo-nada · 2026-06-05T08:24:35Z

Wow this was a crazy fast PR. Thank you!

849261680 · 2026-06-05T08:40:13Z

Wow this was a crazy fast PR. Thank you!

Glad to help!

ceo-nada · 2026-06-06T04:22:20Z

Is this going to get approved to merge to main?

849261680 · 2026-06-06T06:13:21Z

Is this going to get approved to merge to main?

Maybe, up to the maintainers.

steipete · 2026-06-07T07:13:18Z

Maintainer verification for rebased head 85f7f3a.

Reviewed current and shipped behavior: main and v2026.6.1 overwrite buffered Responses tool-call arguments in src/llm/providers/openai-responses-shared.ts when response.function_call_arguments.done omits or empties arguments, while LM Studio Responses API is a documented supported local-provider path in docs/gateway/local-models.md.

Local proof:

OPENCLAW_VITEST_MAX_WORKERS=1 node scripts/run-vitest.mjs src/llm/providers/openai-responses-shared.test.ts extensions/lmstudio/src/stream.test.ts
pnpm exec oxfmt --check --threads=1 src/llm/providers/openai-responses-shared.ts src/llm/providers/openai-responses-shared.test.ts
node scripts/run-oxlint.mjs src/llm/providers/openai-responses-shared.ts src/llm/providers/openai-responses-shared.test.ts
pnpm tsgo:core
git diff --check origin/main...HEAD
pnpm changed:lanes --json (core + coreTests only)
/Users/steipete/Projects/agent-scripts/skills/autoreview/scripts/autoreview --mode branch --base origin/main -> clean, no accepted/actionable findings

Real behavior proof: I also ran a local HTTP/SSE Responses fixture through streamSimpleOpenAIResponses with provider lmstudio; it made a real POST to /v1/responses, streamed a tool-call delta, then sent response.function_call_arguments.done without arguments. After the patch the final event stream ended with stopReason: "toolUse" and preserved the read tool call arguments as { "path": "docs/gateway/local-models.md" }.

CI: gh pr checks --watch --fail-fast is green for head 85f7f3a.

Not tested: a live LM Studio model run; no LM Studio server is reachable at 127.0.0.1:1234 in this environment.

clawsweeper Bot mentioned this pull request Jun 5, 2026

fix(llm): preserve streamed tool args when Responses done omits arguments #90705

Closed

steipete self-assigned this Jun 7, 2026

fix: preserve responses tool args on empty done

85f7f3a

steipete force-pushed the fix/90585-lmstudio-responses-tool-args branch from 5756531 to 85f7f3a Compare June 7, 2026 07:05

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label Jun 7, 2026

clawsweeper Bot added proof: sufficient ClawSweeper judged the real behavior proof convincing. P1 High-priority user-facing bug, regression, or broken workflow. labels Jun 7, 2026

steipete merged commit 0543126 into openclaw:main Jun 7, 2026
176 of 179 checks passed

clawsweeper Bot mentioned this pull request Jun 28, 2026

fix(/v1/responses): drop the extra phantom assistant turn on client-tool calls #71062

Open

28 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix: preserve LM Studio Responses tool arguments#90593

fix: preserve LM Studio Responses tool arguments#90593
steipete merged 1 commit into
openclaw:mainfrom
849261680:fix/90585-lmstudio-responses-tool-args

849261680 commented Jun 5, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 5, 2026 •

edited

Loading

Uh oh!

ceo-nada commented Jun 5, 2026

Uh oh!

849261680 commented Jun 5, 2026

Uh oh!

ceo-nada commented Jun 6, 2026

Uh oh!

849261680 commented Jun 6, 2026

Uh oh!

steipete commented Jun 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

849261680 commented Jun 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Verification

Real behavior proof

Uh oh!

clawsweeper Bot commented Jun 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ceo-nada commented Jun 5, 2026

Uh oh!

849261680 commented Jun 5, 2026

Uh oh!

ceo-nada commented Jun 6, 2026

Uh oh!

849261680 commented Jun 6, 2026

Uh oh!

steipete commented Jun 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

849261680 commented Jun 5, 2026 •

edited

Loading

clawsweeper Bot commented Jun 5, 2026 •

edited

Loading