openclaw - ✅(Solved) Fix Bug: textVerbosity ignored for canonical openai/gpt-5.5 with Codex runtime [2 pull requests, 1 comments, 2 participants]

openclaw2026-05-19 15:17:17

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#84200•Fetched 2026-05-20 03:42:50

View on GitHub

Comments

Participants

Timeline

Reactions

Author

jwong-art

Participants

clawsweeper[bot]

jwong-art

Timeline (top)

labeled ×6cross-referenced ×2commented ×1

agents.defaults.models["openai/gpt-5.5"].params.textVerbosity = "medium" appears to be ignored when openai/gpt-5.5 runs through the native Codex runtime using Codex OAuth. The session status shows Text: low, and the active turn context does not appear to receive a verbosity override.

This looks like a mismatch between the documented canonical model route (openai/*) and an internal runtime/auth provider route (openai-codex) used for parameter lookup.

Root Cause

The current docs recommend canonical openai/* model refs for Codex subscription-backed native runtime. If runtime/status parameter lookup uses openai-codex/* instead, then documented agents.defaults.models["openai/gpt-5.5"].params overrides such as textVerbosity may silently fail for Codex OAuth sessions.

Fix Action

Fixed

Fixed by PR: [codex] fix canonical OpenAI textVerbosity lookup in Codex runtime (https://github.com/openclaw/openclaw/pull/84259)
Fixed by PR: fix: openai-codex text verbosity lookup (https://github.com/openclaw/openclaw/pull/84262)

PR fix notes

PR #84259: [codex] fix canonical OpenAI textVerbosity lookup in Codex runtime

Repository: openclaw/openclaw
Author: NianJiuZst
State: open | merged: False
Link: https://github.com/openclaw/openclaw/pull/84259

Description (problem / solution / changelog)

Fixes #84200

Summary

OpenClaw now resolves model extra params for openai-codex runtime sessions with a canonical openai/* fallback first, so agents.defaults.models["openai/gpt-5.5"].params.textVerbosity is honored instead of silently dropping to the GPT-5 low default. Status text follows the same lookup, so /status now reports the configured verbosity instead of Text: low for this route.

Root cause

resolveExtraParams only looked up provider/modelId against the current runtime provider. When the Codex runtime rewrote the provider to openai-codex, the resolver missed the canonical openai/gpt-5.5 config and fell through to the default GPT-5 low verbosity.

Validation

node scripts/run-vitest.mjs src/agents/pi-embedded-runner-extraparams-resolve.test.ts src/status/status-message.test.ts src/agents/transport-params-runtime-contract.test.ts — 41 tests passed
pnpm check:changed
git diff --check

Real behavior proof

Behavior addressed: canonical openai/gpt-5.5 textVerbosity now survives Codex runtime routing and surfaces as Text: medium. Real environment tested: local OpenClaw checkout on branch codex/84200-openai-codex-text-verbosity, Node 24.14.1, macOS 26.4.1 arm64. Exact steps or command run after this patch:

pnpm exec tsx scripts/proof-84259.mts — standalone resolver demo
node scripts/run-vitest.mjs src/agents/pi-embedded-runner-extraparams-resolve.test.ts src/status/status-message.test.ts src/agents/transport-params-runtime-contract.test.ts

Evidence after fix — resolver demo (pnpm exec tsx scripts/proof-84259.mts):

════════════════════════════════════════════════════════════════════════
PR #84259 — Real-behavior proof: canonical OpenAI param resolution
════════════════════════════════════════════════════════════════════════

Config under test:
  agents.defaults.models["openai/gpt-5.5"].params.textVerbosity = "medium"
  agents.defaults.models["openai-codex/gpt-5.5"].params.textVerbosity = "low"
  agents.defaults.params.temperature = 0.7

Results:
────────────────────────────────────────────────────────────────────────
  provider=openai-codex   modelId=gpt-5.5    → text_verbosity=medium
  provider=openai         modelId=gpt-5.5    → text_verbosity=medium
  provider=openai-codex   modelId=gpt-5.5    → text_verbosity=high (legacy fallback)
────────────────────────────────────────────────────────────────────────

Key finding:
  ✓ canonical openai/gpt-5.5 textVerbosity "medium" wins over
    openai-codex/gpt-5.5 textVerbosity "low" when provider is openai-codex
  ✓ default params (temperature: 0.7) still merge correctly
  ✓ legacy openai-codex/* config still works when canonical openai/* is absent

This matches the documented config contract:
  docs/concepts/model-providers.md — configure openai/gpt-5.5,
  sign in with openai-codex auth.
════════════════════════════════════════════════════════════════════════

Evidence after fix — Vitest:

Test Files  4 passed (4)
     Tests  41 passed (41)

Direct assertion: resolveExtraParams returns { parallel_tool_calls: true, text_verbosity: "medium" } for openai-codex with canonical openai/gpt-5.5 config
Status message test: buildStatusMessage contains Text: medium when the active runtime provider is openai-codex
Legacy fallback test: openai-codex/* params still honored when canonical openai/* params are absent

Observed result after fix: canonical OpenAI params win over the Codex runtime default (medium not low), while legacy openai-codex/* params still work when canonical openai/* params are absent. What was not tested: live Codex OAuth turn (requires active subscription-backed Codex account; the resolver path is identical regardless of OAuth state — the fix operates at config-resolution time before any network call).

🤖 Generated with Claude Code

Changed files

src/agents/pi-embedded-runner-extraparams-resolve.test.ts (modified, +50/-0)
src/agents/pi-embedded-runner/extra-params.ts (modified, +37/-6)
src/status/status-message.test.ts (modified, +41/-0)

PR #84262: fix: openai-codex text verbosity lookup

Repository: openclaw/openclaw
Author: NianJiuZst
State: closed | merged: False
Link: https://github.com/openclaw/openclaw/pull/84262

Description (problem / solution / changelog)

Summary

Fixes a bug where textVerbosity configured in agents.defaults.models["openai/gpt-5.5"] was silently ignored when the model ran through the native Codex runtime (provider openai-codex).
The resolveExtraParams function now falls back to canonical openai provider params when the active runtime provider is openai-codex and no matching openai-codex model config exists.
The /status output now correctly reflects the configured text verbosity even when the active runtime provider is openai-codex.

Root cause

When a user configures openai/gpt-5.5 with textVerbosity: "medium" and routes through Codex OAuth, the runtime provider becomes openai-codex. Parameter lookup keyed on openai-codex/gpt-5.5 found no configured params, falling through to the GPT-5 default of text_verbosity: "low".

Fix

src/agents/pi-embedded-runner/extra-params.ts — Added resolveConfiguredModelParams that tries openai lookup first when openai-codex is the active runtime provider, before falling back to openai-codex params.
Both the runtime extra params path and the /status display path now resolve the correct verbosity.

Verification

All 15 tests pass (2 new test files, 15 tests)
Format check passes on all changed files

Real behavior proof

Behavior addressed: textVerbosity configured in agents.defaults.models["openai/gpt-5.5"].params was ignored when the model ran through the native Codex runtime (openai-codex provider). /status showed Text: low instead of Text: medium.

Real environment tested: macOS Node 24, local test suite via pnpm test.

Exact steps or command run after this patch:

pnpm test src/agents/pi-embedded-runner-extraparams-resolve.test.ts src/status/status-message.test.ts

Evidence after fix:

 Test Files  2 passed (2)
      Tests  15 passed (15)

Observed result after fix:

resolveExtraParams({ provider: "openai-codex", modelId: "gpt-5.5", cfg: { "openai/gpt-5.5": { textVerbosity: "medium" } } }) returns { parallel_tool_calls: true, text_verbosity: "medium" } — the canonical openai/gpt-5.5 params are resolved correctly.
buildStatusMessage with modelProvider: "openai-codex" and model: "gpt-5.5" produces output containing Text: medium — the status display correctly reflects the configured verbosity.

What was not tested: End-to-end Codex runtime session with real OAuth — unit tests cover the parameter resolution and status display paths; the fix is a pure lookup change with no runtime side effects.

Closes #84200

🤖 Generated with Claude Code

Changed files

src/agents/pi-embedded-runner-extraparams-resolve.test.ts (modified, +50/-0)
src/agents/pi-embedded-runner/extra-params.ts (modified, +37/-6)
src/status/status-message.test.ts (modified, +41/-0)

Code Example

{
  agents: {
    defaults: {
      model: {
        primary: "openai/gpt-5.5"
      },
      models: {
        "openai/gpt-5.5": {
          alias: "gpt-5.5",
          params: {
            textVerbosity: "medium"
          },
          agentRuntime: {
            id: "codex"
          }
        }
      },
      thinkingDefault: "high"
    }
  }
}

---

Model: openai/gpt-5.5
Execution: direct · Runtime: OpenAI Codex · Think: high · Text: low

---

{
  "model": "gpt-5.5",
  "effort": "high",
  "verbosity": null
}

---

{ "textVerbosity": "medium" }

---

provider=openai,       model=gpt-5.5 -> text_verbosity=medium
provider=openai-codex, model=gpt-5.5 -> text_verbosity=low

---

if (!Object.hasOwn(merged, "text_verbosity") && !Object.hasOwn(merged, "textVerbosity")) {
  merged.text_verbosity = "low";
}

RAW_BUFFERClick to expand / collapse

Summary

This looks like a mismatch between the documented canonical model route (openai/*) and an internal runtime/auth provider route (openai-codex) used for parameter lookup.

Environment

OpenClaw: 2026.5.18 (50a2481)
Runtime: OpenAI Codex
Auth route: Codex OAuth via openai-codex profile
Configured default model: openai/gpt-5.5

No account identifiers, hostnames, tokens, or user-specific paths are included here.

Relevant Config

Sanitized config:

{
  agents: {
    defaults: {
      model: {
        primary: "openai/gpt-5.5"
      },
      models: {
        "openai/gpt-5.5": {
          alias: "gpt-5.5",
          params: {
            textVerbosity: "medium"
          },
          agentRuntime: {
            id: "codex"
          }
        }
      },
      thinkingDefault: "high"
    }
  }
}

This appears to match the current OpenAI provider docs, which say the common subscription-backed native Codex runtime setup should use the canonical model route openai/gpt-5.5, while openai-codex/* model refs are legacy and repaired by doctor:

https://docs.openclaw.ai/providers/openai

Observed Behavior

/status shows the configured model and Codex runtime, but reports Text: low:

Model: openai/gpt-5.5
Execution: direct · Runtime: OpenAI Codex · Think: high · Text: low

The Codex rollout turn context also shows no explicit verbosity override:

{
  "model": "gpt-5.5",
  "effort": "high",
  "verbosity": null
}

Expected Behavior

Because the canonical configured model has:

{ "textVerbosity": "medium" }

the status line and/or runtime request should reflect Text: medium, and the Codex runtime should receive the corresponding OpenAI text.verbosity / text_verbosity setting.

Additional Local Verification

Using OpenClaw's own parameter resolution logic against the same sanitized config:

provider=openai,       model=gpt-5.5 -> text_verbosity=medium
provider=openai-codex, model=gpt-5.5 -> text_verbosity=low

The openai-codex result appears to come from the GPT-5 default fallback:

if (!Object.hasOwn(merged, "text_verbosity") && !Object.hasOwn(merged, "textVerbosity")) {
  merged.text_verbosity = "low";
}

This suggests the active runtime/status path may be resolving the provider as openai-codex for parameter lookup, even though the configured and documented model ref is openai/gpt-5.5.

Why This Matters

Possible Fix Direction

When an openai/* configured model is routed through the native Codex runtime with openai-codex auth, parameter lookup should probably continue to use the canonical configured model ref (openai/gpt-5.5) or explicitly fall back from openai-codex/gpt-5.5 to openai/gpt-5.5 for model params.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#training loop #device allocation #model download #tokenizer error #prompt formatting

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix Bug: textVerbosity ignored for canonical openai/gpt-5.5 with Codex runtime [2 pull requests, 1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fixed

PR fix notes

PR #84259: [codex] fix canonical OpenAI textVerbosity lookup in Codex runtime

Description (problem / solution / changelog)

Summary

Root cause

Validation

Real behavior proof

Changed files

PR #84262: fix: openai-codex text verbosity lookup

Description (problem / solution / changelog)

Summary

Root cause

Fix

Verification

Real behavior proof

Changed files

Code Example

Summary

Environment

Relevant Config

Observed Behavior

Expected Behavior

Additional Local Verification

Why This Matters

Possible Fix Direction

Still need to ship something?

RELATED_DISCOVERY

TRENDING