Agents: enable Codex parallel tool calls by sasan1200 · Pull Request #53819 · openclaw/openclaw

sasan1200 · 2026-03-24T17:04:06Z

Summary

Describe the problem and fix in 2–5 bullets:

Problem: src/agents/pi-embedded-runner/extra-params.ts applied parallel_tool_calls to openai-completions and openai-responses, but not openai-codex-responses.
Why it matters: openai-codex/gpt-5.3-codex sessions could serialize tool work by default, leading to extra assistant/tool rounds and unnecessary prompt replay costs.
What changed: the runner now applies parallel_tool_calls to openai-codex-responses and defaults openai-codex to parallel_tool_calls: true unless explicitly overridden.
What did NOT change (scope boundary): no other providers were changed, no tool policies changed, and explicit parallel_tool_calls: false / null overrides still win.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes #
Related #

User-visible / Behavior Changes

openai-codex embedded runs now default to parallel tool calls on the responses API path unless explicitly disabled. This should reduce serialized tool loops and prompt replay cost for Codex workflows.

Security Impact (required)

New permissions/capabilities? (Yes/No) No
Secrets/tokens handling changed? (Yes/No) No
New/changed network calls? (Yes/No) No
Command/tool execution surface changed? (Yes/No) No
Data access scope changed? (Yes/No) No
If any Yes, explain risk + mitigation:

Repro + Verification

Environment

OS: macOS
Runtime/container: local Node/tsx run from clean worktrees
Model/provider: openai-codex/gpt-5.3-codex
Integration/channel (if any): none
Relevant config (redacted): isolated temp state dir with only openai-codex:default auth + model store; minimal config pinned to openai-codex/gpt-5.3-codex; plugins disabled

Steps

Run the same real embedded Codex workflow from a clean origin/main worktree with isolated state.
Run the same workflow from this branch with the same prompt, auth profile, and isolated state.
Compare settled transcript files for assistant turns, toolUse turns, and summed assistant token usage.

Expected

The patched branch should complete the same workflow with fewer serialized assistant/tool rounds and lower total token usage.

Actual

Settled transcript comparison:
- origin/main: 19 assistant turns, 17 toolUse turns, 1,414,601 assistant tokens
- this branch: 13 assistant turns, 12 toolUse turns, 961,260 assistant tokens
- delta: -31.6% assistant turns, -29.4% toolUse turns, -32.0% assistant tokens

Evidence

Attach at least one:

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios:
- Confirmed origin/main excluded openai-codex-responses from parallel_tool_calls handling in src/agents/pi-embedded-runner/extra-params.ts.
- Ran pnpm test -- src/agents/pi-embedded-runner-extraparams.test.ts after adding Codex regression coverage.
- Ran one real isolated openai-codex/gpt-5.3-codex workflow on origin/main and on this branch, then recomputed metrics from the settled transcript files.
Edge cases checked:
- Explicit parallel_tool_calls values for Codex responses.
- Default-on behavior for openai-codex.
- Explicit false still disables the default.
What you did not verify:
- More than one live end-to-end Codex scenario.
- Behavior of every ordering-sensitive tool sequence under parallel execution.

Review Conversations

I replied to or resolved every bot review conversation I addressed in this PR.
I left unresolved only the conversations that still need reviewer or maintainer judgment.

If a bot review conversation is addressed by this PR, resolve that conversation yourself. Do not leave bot review conversation cleanup for maintainers.

Compatibility / Migration

Backward compatible? (Yes/No) Yes
Config/env changes? (Yes/No) No
Migration needed? (Yes/No) No
If yes, exact upgrade steps:

Failure Recovery (if this breaks)

How to disable/revert this change quickly: set parallel_tool_calls: false for the affected Codex model/provider config, or revert this commit
Files/config to restore: src/agents/pi-embedded-runner/extra-params.ts
Known bad symptoms reviewers should watch for: ordering-sensitive Codex workflows behaving differently because tool calls are no longer serialized by default

Risks and Mitigations

List only real risks for this PR. Add/remove entries as needed. If none, write None.

Risk: some Codex workflows may implicitly rely on serialized tool execution order.
- Mitigation: explicit parallel_tool_calls: false still disables the new default, and the patch scope is limited to openai-codex responses.
Risk: the live improvement can vary by prompt/workflow.
- Mitigation: unit coverage verifies the config behavior directly, and this PR includes one real before/after Codex comparison with transcript-backed numbers.

greptile-apps · 2026-03-24T17:06:25Z

Greptile Summary

This PR enables parallel_tool_calls: true by default for the openai-codex provider when using the openai-codex-responses API, reducing serialized tool round-trips for Codex coding workloads.

Adds openai-codex-responses to the API guard in createParallelToolCallsWrapper so the injection path is reachable for Codex models.
Introduces resolveDefaultParallelToolCalls(provider, modelApi) which returns true only when both provider === "openai-codex" and modelApi === "openai-codex-responses", keeping all other providers/APIs unaffected.
Adds modelApi? as a new trailing optional parameter to applyExtraParamsToAgent to preserve backward compatibility with all existing call sites.
Wires params.model.api through in attempt.ts so the production path receives the correct API string.
Adds 4 targeted unit tests (default-on, camelCase override, snake_case override, non-Responses API no-op).
Existing null-override suppression and camelCase/snake_case aliasing are fully preserved.

Confidence Score: 5/5

This PR is safe to merge; the change is well-scoped, backward-compatible, and fully covered by new and existing tests.
The implementation is minimal and surgical: a new optional parameter appended to the end of applyExtraParamsToAgent, a pure function resolveDefaultParallelToolCalls, and a one-line expansion of the API guard. All existing callers omitting modelApi continue to behave identically. The only production call site (attempt.ts) is updated correctly. Four new tests cover the happy path and all meaningful override combinations. No regressions are introduced.
No files require special attention.

_{Reviews (1): Last reviewed commit: "fix(test): pass modelApi through to appl..." | Re-trigger Greptile}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3f7c1b45b8

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 894e6bb799

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

steipete · 2026-04-25T09:47:54Z

Closing this as implemented after Codex review.

Current main already covers this request. Codex Responses is treated as a supported parallel_tool_calls API, GPT-5/OpenAI and openai-codex defaults now set parallel_tool_calls: true, and current tests cover both configured and default Codex Responses behavior.

What I checked:

Current API support: supportsGptParallelToolCallsPayload() now includes openai-codex-responses, so the payload patch path reaches Codex Responses models on main. (src/agents/provider-api-families.ts:1, 8f64cd3e4d83)
Current default behavior: applyDefaultOpenAIGptRuntimeParams() applies to both openai and openai-codex GPT-5 model IDs and sets parallel_tool_calls = true when not overridden. (src/agents/pi-embedded-runner/extra-params.ts:221, 8f64cd3e4d83)
Current regression coverage: Tests on main explicitly cover configured Codex Responses injection and the default GPT-5 Codex Responses behavior. (src/agents/pi-embedded-runner-extraparams.test.ts:831, 8f64cd3e4d83)
Implementing commit on main: Main already contains commit 40be5ad58187fd20b8defe5fc30e649190c131a8 (fix: harden GPT-5 runtime paths), which added openai-codex-responses handling in the parallel-tool-calls wrapper and corresponding tests. (src/agents/pi-embedded-runner/extra-params.ts:374, 40be5ad58187)

So I’m closing this as already implemented rather than keeping a duplicate issue open.

Review notes: reviewed against 8f64cd3e4d83; fix evidence: commit 40be5ad58187.

openclaw-barnacle Bot added agents Agent runtime and tooling size: S labels Mar 24, 2026

sasan1200 changed the title ~~fix(agents): enable Codex parallel tool calls~~ Agents: enable Codex parallel tool calls Mar 24, 2026

chatgpt-codex-connector Bot reviewed Mar 24, 2026

View reviewed changes

Comment thread src/agents/pi-embedded-runner-extraparams.test.ts Outdated

sasan1200 added 2 commits April 1, 2026 13:46

Agents: enable Codex parallel tool calls

917ae27

Agents: scope Codex parallel-tool default to responses

894e6bb

sasan1200 force-pushed the fix-codex-parallel-tool-calls branch from 3f7c1b4 to 894e6bb Compare April 1, 2026 17:57

chatgpt-codex-connector Bot reviewed Apr 1, 2026

View reviewed changes

Comment thread src/agents/pi-embedded-runner-extraparams.test.ts

This was referenced Apr 24, 2026

[codex] Add Pi/Codex harness extension seams #70772

Merged

[codex] Harden GPT-5.4 runtime paths #70743

Merged

steipete closed this Apr 25, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Agents: enable Codex parallel tool calls#53819

Agents: enable Codex parallel tool calls#53819
sasan1200 wants to merge 2 commits intoopenclaw:mainfrom
sasan1200:fix-codex-parallel-tool-calls

sasan1200 commented Mar 24, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Mar 24, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

sasan1200 commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

greptile-apps Bot commented Mar 24, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sasan1200 commented Mar 24, 2026 •

edited

Loading