fix(chat): preserve refusal across JSON and streamed projections

docs-site/src/content/docs/guides/codex-app-models.md

@@ -64,6 +64,23 @@ or grant account entitlement. The separately billed
 `openai-apikey/daybreak-blue-latest` API row is a different route and its 1,050,000 / 922,000 limits
 are never copied into the Codex-login row.
 
+For custom Astra and Daybreak rows on that canonical `openai` Codex-forward destination,
+explicit `reasoningEfforts` are bounded by the model's pinned Codex capabilities. A custom
+`["none", "minimal", "low"]` becomes `["low"]` in the catalog; a nonempty list with no
+supported values also falls back to the native default as a single choice. An explicit `[]`
+stays empty and has no advertised default. A declared default is retained only if it belongs to
+the resulting list; otherwise the native default is used when present, then the first surviving
+choice. Stored custom configuration is unchanged, and repeated syncs do not add `max` back to a
+narrow custom list.
+
+This requires the exact provider, destination, and capability-backed model identity. An arbitrary
+gateway such as `YYLJ/gpt-6-astra` does not inherit native capabilities from its name. Its explicit
+custom ladder continues to override discovered provider metadata under the normal routed rules.
+Codex's native Astra `ultra` choice is retained: it is a client delegation mode converted to a
+supported wire effort, distinct from the [API model's effort list](https://developers.openai.com/api/docs/models/gpt-6-astra).
+Catalog normalization does not rewrite existing thread settings or establish support for a
+particular installed Desktop version.
+
 When the `codexAccountNamespaces` map is empty, account-qualified picker rows are off. If
 `codexAccountPickerEnabled` is omitted with a non-empty map, they are treated as enabled for
 backward compatibility. Set it to `false` to hide generated qualified rows and restore bare native

docs-site/src/content/docs/reference/configuration/providers.md

@@ -201,6 +201,14 @@ predictions. Explicit provider/model price overrides still take precedence.
 | `unsafeAllowNativeLocalExec?` | `boolean` | Cursor legacy boolean, equivalent to `nativeLocalExec: "on"` only when the newer field is unset. |
 | `nativeLocalExec?` | `"off" \| "codex-sandbox" \| "on"` | Cursor local-exec policy. `off` is default; `codex-sandbox` currently fails closed like `off`. |
 
+Custom-model `reasoningEfforts` normally override discovered provider metadata. The bounded
+exception is an explicit Astra or Daybreak custom row on the canonical `openai` Codex-forward
+destination: its advertised list is intersected with that model's pinned native capabilities.
+An explicit empty list remains empty with no default; a nonempty incompatible list falls back
+to the native default as a single choice. Defaults must belong to the final list. This changes
+the catalog projection, not stored configuration or arbitrary gateway models sharing a GPT name.
+See [custom native catalog examples](/guides/codex-app-models/).
+
 ### Discovered model display names
 
 Use `modelDisplayNames` when a provider returns machine friendly ids but the Codex model picker

docs-site/src/content/docs/reference/proxy-formats.md

@@ -260,6 +260,14 @@ It does not issue an additional inference request. An incomplete response caused
 token limit or content filtering retains `length` or `content_filter`, even if it includes tool
 output. Other incomplete boundaries return an upstream error instead of claiming a normal finish.
 
+Refusal text stays separate from answer text: JSON completions use nullable `message.refusal`,
+and streaming chunks use `delta.refusal`. Native Chat JSON-to-SSE and SSE-to-JSON conversions
+preserve that field; native streaming relay preserves the provider's refusal deltas. On translated
+Responses streams, refusal parts are buffered until the terminal event and emitted once in their
+original output/content order. Compatible repeated or sparse snapshots do not duplicate or erase
+text. Contradictory refusal snapshots and buffer overflow produce a typed error without a successful
+finish or `[DONE]`. This preserves the upstream refusal; it does not introduce a proxy policy decision.
+
 Because the internal execution path is Responses-based, a provider adapter can impose a narrower
 feature set. For example, a request feature that cannot be represented by the selected adapter is
 returned as an error instead of silently changing its meaning.

scripts/test-layout/layout.json

@@ -271,6 +271,7 @@
     "catalog-vision-sidecar-modalities.test.ts": "codex-integration",
     "chat-completions-endpoint.test.ts": "responses",
     "chat-json-sse-fallback.test.ts": "responses",
+    "chat-refusal.test.ts": "responses",
     "chatgpt-device-auth.test.ts": "oauth",
     "chatgpt-oauth.test.ts": "oauth",
     "chatgpt-token-expiry.test.ts": "oauth",