Greptile Summary

Removes the three places in the Qwen plugin that hard-blocked qwen3.6-plus on Coding Plan endpoints (buildQwenModelCatalogForBaseUrl filter, normalizeConfig model strip, and suppressBuiltInModel error return), and updates docs and tests accordingly. The approach is sound — letting the upstream provider surface unsupported-model errors is preferable to OpenClaw pre-emptively blocking a model that some tenants can already use.

Confidence Score: 5/5

• Safe to merge; all remaining findings are minor cleanup suggestions with no impact on correctness.
• No P0 or P1 issues. The two P2 findings are a dead stub function (isQwen36PlusUnsupportedForConfig) with an unreachable branch in suppressBuiltInModel, and Coding Plan noteMessage arrays that don't yet mention qwen3.6-plus. Neither blocks the intended behavior.
• extensions/qwen/index.ts — dead stub and unreachable return; Coding Plan note messages

This is a comment left during a code review.
Path: extensions/qwen/index.ts
Line: 42-180

Comment:
**Dead stub and unreachable branch**

`isQwen36PlusUnsupportedForConfig` always returns `false`, so `!isQwen36PlusUnsupportedForConfig(...)` is always `true`, which means the `if` in `suppressBuiltInModel` is always entered and the second `return undefined` at line 180 is unreachable. The stub function and the whole conditional can be deleted — the hook can just be `suppressBuiltInModel: () => undefined`.

```suggestion
    suppressBuiltInModel: () => undefined,
```

`isQwen36PlusUnsupportedForConfig` (lines 42–47) should be removed as well since it has no callers after this cleanup.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: extensions/qwen/index.ts
Line: 118-122

Comment:
**Coding Plan note messages omit qwen3.6-plus**

Both Coding Plan auth option `noteMessage` arrays (China and Global/Intl, lines ~118–122 and ~138–142) list `"Models: qwen3.5-plus, glm-5, kimi-k2.5, MiniMax-M2.5, etc."` but don't mention `qwen3.6-plus`, even though this PR is specifically exposing it on those endpoints. Users reading the onboarding wizard will see an incomplete model list for Coding Plan.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "qwen: expose qwen3.6-plus on coding plan..." | Re-trigger Greptile}

@codex review

Codex Review: Didn't find any major issues. You're on a roll.

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve.

Keep this PR open. Current main still does not implement the broad behavior this PR proposes: qwen3.6-plus is still filtered from Qwen Coding Plan catalogs, stripped from Coding Plan provider config, suppressed through manifest metadata, and documented/tested as Standard-endpoint-only. The related maintainer PR #72664 is open and intentionally narrower, so it is not a merged or accepted replacement yet.

Best possible solution:

Keep this PR open for maintainer product and security review. If maintainers want the broader provider-truth behavior, rebase or land an equivalent patch that removes Coding Plan filtering from extensions/qwen/models.ts, extensions/qwen/index.ts, and the manifest suppression in extensions/qwen/openclaw.plugin.json, then update Qwen tests and docs. If maintainers choose #72664's narrower explicit-config policy as canonical, close this PR only after that policy lands or is explicitly accepted as the replacement.

What I checked:

• Current main still filters Coding Plan catalogs: isQwen36PlusSupportedBaseUrl() returns false for Coding Plan hosts, and buildQwenModelCatalogForBaseUrl() filters out QWEN_36_PLUS_MODEL_ID when that predicate is false. (extensions/qwen/models.ts:126, 0b82a7e718e3)
• Current main still strips configured Coding Plan models: The Qwen provider normalizeConfig path checks Coding Plan base URLs and removes qwen3.6-plus from configured provider models. (extensions/qwen/index.ts:163, 0b82a7e718e3)
• Current main still has manifest model suppression: The Qwen plugin manifest suppresses qwen3.6-plus for qwen and modelstudio when the base URL host is a Coding Plan host. (extensions/qwen/openclaw.plugin.json:35, 0b82a7e718e3)
• Runtime paths consume suppression before explicit inline resolution: Explicit model resolution calls shouldSuppressBuiltInModel() before findInlineModelMatch(), so manifest suppression can still block configured qwen3.6-plus before inline model lookup. (src/agents/pi-embedded-runner/model.ts:583, 0b82a7e718e3)
• Tests and docs still assert Standard-only behavior: The provider catalog test expects qwen3.6-plus to be absent on the default/Coding Plan provider and present on Standard endpoints; the public Qwen docs still tell users to prefer Standard endpoints for qwen3.6-plus. (extensions/qwen/provider-catalog.test.ts:18, 0b82a7e718e3)
• Latest PR scope and security pass: The latest PR head is a69fb71c5e4da240e363b8e6ed6c6f0fd876f971, open and unmerged, changing only docs/providers/qwen.md, extensions/qwen/index.ts, extensions/qwen/models.ts, and extensions/qwen/provider-catalog.test.ts. It does not touch workflows, dependency manifests, lockfiles, publish metadata, scripts, secret handling, downloaded artifacts, or other broad code-execution surfaces. (a69fb71c5e4d)

Remaining risk / open question:

• The latest PR file list does not include extensions/qwen/openclaw.plugin.json, while current main now carries Coding Plan qwen3.6-plus suppression there; a rebase or equivalent maintained patch needs to handle that manifest path for full runtime/list behavior.
• Upstream Coding Plan availability appears plan- or tenant-dependent, so advertising qwen3.6-plus by default remains a maintainer product-policy choice, especially while fix(qwen): allow explicit qwen3.6-plus on Coding Plan #72664 proposes a narrower explicit-config policy.
• fix(qwen): allow explicit qwen3.6-plus on Coding Plan #72664 says this PR was routed to central security handling and not used as an autonomous mutation target, so this PR should not be auto-merged without maintainer/security review.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 0b82a7e718e3.

Thanks for pushing this forward. The fix has now landed via the narrower replacement PR #72664, which preserves explicit qwen/qwen3.6-plus opt-in on Coding Plan while keeping the built-in catalog suppression behavior. Closing this as superseded, with the original report/work credited in the landed changelog path.