fix: show the host's purge refusal and say auto is modern

README.md

@@ -251,15 +251,15 @@ Nine controls in four sections — the decisions a user actually makes. Everythi
 | **Models & Free Tier** | Enrich model metadata from the OpenCode catalog | `on` | Fills catalog context limits, modalities and thinking levels into model info, and lets the gateway `/models` listing carry the full catalog. Does **not** add models to the picker — the picker shows declared models only; use the card's **Check catalog** row to declare more |
 | **Models & Free Tier** | Inject Core Tools | `on` | Adds the `read` + `bash` schemas free-tier `/responses` bodies require |
 | **Quota Meter** | Enable Composer Usage Meter | `on` | Mounts the live meter in the composer dock — the Go quota ring, or the Zen session spend |
-| **Account & Endpoint** | API Endpoint | `auto` | Which endpoint family requests go to: **Automatic** (follow the catalog) · **Legacy** (`zen/v1`) · **Modern** (`inference/…`) |
+| **Account & Endpoint** | API Endpoint | `auto` | Which endpoint family requests go to: **Automatic** (same as Modern) · **Legacy** (`zen/v1`, `zen/go/v1`) · **Modern** (per-protocol `inference/{openai,anthropic,google}/…`, `inference/go/…` on Go) |
 | **Account & Endpoint** | Credential Source | `auto` | Which key wins when several are known: **Automatic** · **Live request first** · **Declared key first** |
 | **Account & Endpoint** | Read the OpenCode installation's key | `off` | Lets the plugin read the credential OpenCode itself stored, as the LAST resort — off by default |
 
 ### Choosing the endpoint and the credential
 
 Two questions decide where a request goes, and the **Account & Endpoint** section is where you answer them.
 
-**Which endpoint family.** The gateway answers on two families — the older `/zen/v1` and the newer `inference` family, whose base is per protocol (`…/inference/openai/v1`, `…/inference/anthropic/v1`, `…/inference/google/v1beta`) — and they are entitled **separately**: an account that works on one is not guaranteed to work on the other. `auto` follows the gateway's own catalog, which today names the newer family, so a catalog move carries you along; `legacy` and `modern` pin one, which is what you want when your account only works on the older one, or when you are checking that the newer one does.
+**Which endpoint family.** The gateway answers on two families — the older `/zen/v1` (`/zen/go/v1` on Go), one base for every protocol, and the newer per-protocol `inference` family (`/inference/openai/v1`, `/inference/anthropic/v1`, `/inference/google/v1beta`, with `/inference/go/…` on Go) — and they are entitled **separately**: an account that works on one is not guaranteed to work on the other. `modern` sends each protocol to its own inference path; `auto`, the default, is an alias of `modern`; `legacy` pins the older family, which is what you want when your account only works there.
 
 **Which credential wins.** `keySource` orders the sources the plugin can see — `auto` prefers a key you declared, then one captured from a live request, then the declared reference; `request` promotes a captured key first; `configured` does the reverse. Below it, `readOpencodeStore` lets the plugin read the credential OpenCode itself stored, as the **last** resort only. Both are `off`/`auto` by default, so nothing changes until you say so.
 

README.zh-CN.md

@@ -249,15 +249,15 @@ opencode-responses:
 | **模型与免费额度** | 用 OpenCode 官方目录补全模型信息（默认开启） | `on` | 把目录的上下文上限、模态与思考档位补进模型信息，并让网关 `/models` 列表带上完整目录。**不会**往选择器里添加模型——选择器只显示已声明的；要添加请用卡片的**检查目录** |
 | **模型与免费额度** | 自动补全核心工具（默认开启） | `on` | 补上免费层 `/responses` 请求体所需的 `read` + `bash` schema |
 | **配额计量** | 开启输入区用量计量（默认开启） | `on` | 在输入框停靠栏挂载实时计量表——Go 额度环，或 Zen 会话消耗 |
-| **账户与端点** | API 端点 | `auto` | 请求发往哪个端点家族：**自动**（跟随目录） · **旧端点**（`zen/v1`） · **新端点**（`inference/…`） |
+| **账户与端点** | API 端点 | `auto` | 请求发往哪个端点家族：**自动**（同新端点） · **旧端点**（`zen/v1`、`zen/go/v1`） · **新端点**（按协议的 `inference/{openai,anthropic,google}/…`，Go 为 `inference/go/…`） |
 | **账户与端点** | 凭据来源 | `auto` | 多个密钥同时可用时以哪个为准：**自动** · **优先实时请求** · **优先已声明密钥** |
 | **账户与端点** | 读取 OpenCode 安装版的密钥（默认关闭） | `off` | 允许本插件读取 OpenCode 自己保存的凭据，且仅作最后手段——默认关闭 |
 
 ### 选择端点与凭据
 
 一个请求发往哪里由两个问题决定，答案就在**账户与端点**分区里。
 
-**发往哪个端点家族。** 网关在两套端点上都应答——较早的 `/zen/v1` 与较新的 `inference` 家族（其 base 按协议区分：`…/inference/openai/v1`、`…/inference/anthropic/v1`、`…/inference/google/v1beta`）——而两者的授权是**彼此独立**的：在一边可用的账户不保证在另一边可用。`auto` 会跟随网关自己的目录，目前它指向较新的一套，因此目录迁移会带着你一起走；`legacy` 与 `modern` 则固定其一，适用于账户只在较早一套上可用，或你想确认较新的一套是否可用的情况。
+**发往哪个端点家族。** 网关在两套端点上都应答——较早的 `/zen/v1`（Go 为 `/zen/go/v1`），所有协议共用一个基址；以及较新的按协议划分的 `inference` 家族（`/inference/openai/v1`、`/inference/anthropic/v1`、`/inference/google/v1beta`，Go 为 `/inference/go/…`）——而两者的授权是**彼此独立**的：在一边可用的账户不保证在另一边可用。`modern` 将每个协议发往各自的 inference 路径；`auto`（默认值）是 `modern` 的别名；`legacy` 固定使用较早的一套，适用于账户只在那里可用的情况。
 
 **以哪个凭据为准。** `keySource` 决定插件可见的多个凭据源的优先级——`auto` 优先使用你声明过的密钥，其次是从实时请求中捕获的密钥，最后是声明的凭据引用；`request` 把捕获到的密钥提到最前；`configured` 则相反。其下的 `readOpencodeStore` 允许插件读取 OpenCode 自己保存的凭据，且**仅作最后手段**。两者默认分别为 `auto` 与关闭，因此在你说要改之前一切照旧。
 

docs/improvement-plan.md

@@ -21,7 +21,7 @@ Phases land as **stacked pull requests**. Each one is based on the previous phas
 | Phase | PR | Title | State | Restack onto 1.8.0 |
 | :-- | :-- | :-- | :-- | :-- |
 | P0 | #105 | suite hardening, 10 defect fixes, this plan | in review | **merged `main` 1.8.0**; catalog refreshes and the pricing fix dropped (on `main`) |
-| P1a | #106 | `fix:` shipped breakage | in review | pending: merges P0; shrinks hard (title changes) |
+| P1a | #106 | `fix:` the purge row's refusal and the endpoint copy | in review | **merged P0**; shrunk to the purge row's `RemoteError.message`, the `auto` = `modern` copy and the `opencode-google` regression tests; the warning colour drops its token fallback; the catalog's refusal tests gain three cases `main` lacks; the rest is on `main`, and the debug gate follows D24 |
 | P1b | #107 | `chore:` hygiene | in review | pending: merges P1a; drops its tool-fallback variant (D25) |
 | P2 | #108 | `feat:` a smaller public API | in review | pending: merges P1b; re-retires the wrapper (D9) |
 | P-T1a | #109 | `build:` pin the host | in review | pending: merges P2; redone on the alpha cluster and pi 1.3.0 (D14', D22') |
@@ -112,11 +112,11 @@ Phases land as **stacked pull requests**. Each one is based on the previous phas
 | Settings shows pi's snapshot, not the catalog | `models-discovery.ts:142-198,532-539` |
 | Mode merging is done only for Go→Gemini. For the other modes the picker shows catalog levels while dispatch validates against a different profile: toggle models show Minimal…Max with no Off, and Go `minimax-m3` + Xhigh/Max is refused on the internal Anthropic route. | `models-discovery.ts:607-616,735-760,1063-1203`; `pi-provider.ts:351-386`; `responses-provider.ts:111-150` |
 | ~~`isZenProvider("opencode-google")` is false~~ Fixed on `main`: only `opencode-go` and `opencode-go-*` are excluded. P1a adds the regression test. | `utils/request-shape.ts:273-282` |
-| ~~`enrichModelsResponse` turns a 401/403/5xx into a 200 catalog. It must still fill the Go plane's 404.~~ Fixed on `main` (#133): a refusal other than 404 passes through. | `models-catalog.ts:731` |
+| ~~`enrichModelsResponse` turns a 401/403/5xx into a 200 catalog. It must still fill the Go plane's 404.~~ Fixed on `main` (#133): a refusal other than 404 passes through. P1a adds the Zen-plane 404 fill, the non-JSON 200 answer and the retry-after passthrough cases. | `models-catalog.ts:731` |
 | ~~The warning token doesn't exist, so warnings render as caption text.~~ Fixed on `main` (#133): both use `--dsw-alias-state-warn-label`. | `settings-card.tsx:159`, `settings-store-inspect.tsx:250` |
 | ~~The Settings spinner animates only if the meter is mounted~~ Fixed on `main` (#133): the card carries its own keyframes. | `settings-usage.tsx:45` |
 | ~~The stream hook writes `debugFile` with `debug` off~~ Not a defect under D24: naming `debugFile` writes both halves. | `stream-hook.ts:327-340` vs `lifecycle.ts:361` |
-| `settings-purge`: the `{ ok: false }` branch and the parser are on `main` (#133). Still open: a refused purge whose `error` is the host's `RemoteError` object, not a string, shows the generic message; P1a reads its `message`. | `settings-purge.ts:56-69` |
+| `settings-purge`: the `{ ok: false }` branch and the parser are on `main` (#133). ~~Still open: a refused purge whose `error` is the host's `RemoteError` object, not a string, shows the generic message.~~ Fixed by P1a: `settings-purge.ts` reads the `RemoteError`'s `message`. | `settings-purge.ts:56-69` |
 | The stream hook's `prepared` handling is dead. The host never reads `options.prepared`. | `stream-hook.ts:44,200-203,258` |
 | The partition test misses `debugFile` | `test/settings-page.test.tsx:1223` |
 | `cordis.patch.yml`: phantom `showUsagePrice`, wrong `freeModelMarker: ''` text, incomplete field lists | `cordis.patch.yml:23,36,61` |
@@ -337,15 +337,13 @@ What stays private is pinned by contract tests against the real host packages:
   - completions models move from `/zen(/go)/v1` to `/inference(/go)/openai/v1`;
   - `decoratePlaneAdapters`' baseUrl repair never reaches the wire.
 
-1. **P1a `fix:` shipped breakage.**
-   - The wrapper's broken `export { default }`.
-   - `isZenProvider("opencode-google")` answers false.
-   - `enrichModelsResponse` passes a 401/403/5xx through, and still fills the Go plane's 404.
-   - The warning colour uses `--dsw-alias-state-warn-label`.
-   - The spinner keyframes are in the settings styles.
-   - The debug gate follows D24.
-   - `settings-purge` uses `parsePurgeResult`.
-   - The endpoint copy (`config.ts`, `config-values.ts`, README) names the per-protocol inference family for `modern`, with `auto` as its alias.
+1. **P1a `fix:` the purge row's refusal and the endpoint copy.** The rest of what shipped broken in 1.4.1 was fixed on `main` (#133, #135) before this phase landed.
+   - `settings-purge` shows the `message` of a refusal whose `error` is the host's `RemoteError` object.
+   - The endpoint copy (`config.ts`, `config-values.ts`, the card, both READMEs) names the per-protocol inference family for `modern`, with `auto` as its alias (D19), pinned by `endpoint-copy.test.ts`.
+   - Regression tests for `isZenProvider("opencode-google")`: the predicate, its agreement with `catalogPlaneForRoute`, and the usage service's Zen fallback on that route.
+   - The warning colour is the bare `--dsw-alias-state-warn-label` (no token fallback), and every token the card names is checked against the host's vocabulary.
+   - The stream hook's debug test asserts D24 rather than the gate this phase first proposed.
+   - The catalog's refusal tests gain three cases `main` lacks: the Zen-plane 404 fill, the non-JSON 200 answer, and a `retry-after` header that survives the passthrough. The passthrough itself is on `main` (#133).
 2. **P1b `chore:` hygiene.**
    - Delete `adapter-replacement.ts` and its test, with a note in the engineering notes on why the approach is unsupported.
    - Delete the dead `prepared` handling and `.freebuff/`.
@@ -452,7 +450,7 @@ How each phase changes now that it sits on `main` 1.8.0 and the alpha host. Each
 | Phase | Adjustment |
 | :-- | :-- |
 | P0 | Shrinks: the three catalog refreshes and the per-plane pricing fix are on `main`. Keeps the suite hardening, the lifecycle, mount and credential fixes, D23's scoping with `--copies`, and this plan. |
-| P1a | Shrinks hard: the purge row shows the host's `RemoteError.message`, the endpoint copy says `auto` is `modern`, and `isZenProvider("opencode-google")` gains a regression test. The wrapper fix, refusal passthrough, warning token, spinner and `parsePurgeResult` are on `main` or superseded; the debug gate follows D24. |
+| P1a | Shrinks hard: the purge row shows the host's `RemoteError.message`, the endpoint copy says `auto` is `modern`, and `isZenProvider("opencode-google")` gains a regression test. The wrapper fix, refusal passthrough, warning token, spinner and `parsePurgeResult` are on `main` or superseded; P1a only drops the warning token's fallback, checks every token the card names against the host's vocabulary, and adds three refusal-passthrough test cases `main` lacks (the Zen-plane 404 fill, the non-JSON 200 answer, the `retry-after` passthrough). The debug gate follows D24. |
 | P1b | Takes `main`'s `coreToolDecision` (D25) and ports only the dead Anthropic branch's deletion; its catalogue fixture merges with `main`'s `test/fixtures/catalogs.ts` behind the one door. |
 | P2 | Re-retires the wrapper `main` revived (D9); the door list grows by every `src` name `main`'s tests import. |
 | P-T1a | Redone on D14' and D22': `main`'s pins, then the pi 1.3.0 override and patch; the 1.1.0 patch is deleted. |

docs/redirect-planes.md

@@ -15,7 +15,7 @@ A **plane** is one gateway base URL with its own catalog and its own credential.
 
 Counts as of 2026-11-03 (the shipped shim, regenerated in #132 and refreshed for `pulse-alpha-free`); the document's 2026-10-09 stamp is for its measurements.
 
-**There is no longer one base URL per plane to quote.** The gateway answers on two endpoint families and the `endpoint` setting picks one: `auto` follows the catalog, which today names `…/inference/openai/v1` and `…/inference/go/openai/v1`, and `legacy` pins the older `…/zen/v1` and `…/zen/go/v1`. Both are still served and they are entitled separately, which is why the choice is the user's rather than a migration. The dated measurement below was taken on the legacy family.
+**There is no longer one base URL per plane to quote.** The gateway answers on two endpoint families and the `endpoint` setting picks one: `auto`, the default, is an alias of `modern`, the per-protocol inference family (`…/inference/openai/v1` and `…/inference/go/openai/v1` for completions), and `legacy` pins the older `…/zen/v1` and `…/zen/go/v1`. Both are still served and they are entitled separately, which is why the choice is the user's rather than a migration. The dated measurement below was taken on the legacy family.
 
 A plane is a property of the **account and the base URL**, never of a model. The same model id can exist on both planes and be served on a different shape in each.
 

src/config-values.ts

@@ -76,14 +76,10 @@ export const isKeySourcePolicy = (value: unknown): value is KeySourcePolicy =>
  * The gateway answers on two families and the user, not the catalog, is the one
  * who knows which their account is entitled to:
  *
- * - `auto` follows the catalog — the `api` the vendor's own entry names, which
- *   today is the `inference` family.
- * - `legacy` forces the older pair, which really is one base per plane
- *   (`…/zen/v1`, `…/zen/go/v1`).
- * - `modern` forces the newer `inference` family, whose base is PER PROTOCOL
- *   (`…/inference/openai/v1`, `…/inference/anthropic/v1`,
- *   `…/inference/google/v1beta`, and the `go/` variants) — so "inference/v1" is
- *   not an address the plugin ever sends, and the copy says `inference/…`.
+ * - `auto` — an alias of `modern` (the default).
+ * - `legacy` — the older single base per plane, `/zen/v1` and `/zen/go/v1`.
+ * - `modern` — the per-protocol `inference` family,
+ *   `/inference/{openai,anthropic,google}/…` and `/inference/go/…` on Go.
  *
  * Declared here rather than in `config.ts` so the client card can read the same
  * list without importing schemastery.

src/config.ts

@@ -214,8 +214,8 @@ export interface ResolvedPluginConfig {
    * Which endpoint family the gateway's requests go to.
    *
    * The two families are independently entitled and both are still served, so
-   * the choice is the user's: `auto` follows the catalog, and `legacy` /
-   * `modern` pin one. See {@link ENDPOINT_FAMILIES}.
+   * the choice is the user's: `auto` is an alias of `modern` (the default), and
+   * `legacy` pins the older single base per plane. See {@link ENDPOINT_FAMILIES}.
    */
   endpoint: EndpointFamily;
   /** `*` when the core-tool fallback should run for every model. */
@@ -276,8 +276,8 @@ const resolvedKeySource = (raw: unknown): KeySourcePolicy => {
  * A row's endpoint family, or the default when it names something else.
  *
  * Lenient for the same reason {@link resolvedKeySource} is: a typo in a
- * hand-edited row degrades to `auto`, which follows the catalog and is the
- * answer that cannot be wrong about which family is current.
+ * hand-edited row degrades to `auto`, the default, so a typo cannot select a
+ * family the user never named.
  */
 const resolvedEndpoint = (raw: unknown): EndpointFamily => {
   const value = readString(raw);
@@ -563,10 +563,13 @@ export const Config = z.object({
     .string()
     .default(CONFIG_DEFAULTS.endpoint)
     .description(
-      "Which endpoint family the gateway's requests go to. auto: follow the " +
-        "catalog's own address (today the inference family). legacy: force " +
-        "https://opencode.ai/zen/v1. modern: force " +
-        "https://opencode.ai/inference/<protocol>. The two families are separately " +
+      "Which endpoint family the gateway's requests go to. modern: the " +
+        "per-protocol inference family, " +
+        "https://opencode.ai/inference/{openai,anthropic,google}/... on Zen " +
+        "and https://opencode.ai/inference/go/{openai,anthropic,google}/... " +
+        "on Go. auto: an alias of modern, and the default. legacy: the older " +
+        "single base per plane, https://opencode.ai/zen/v1 and " +
+        "https://opencode.ai/zen/go/v1. The two families are separately " +
         "entitled and both still answer, so this is a choice rather than a " +
         "migration; an unrecognised value falls back to auto."
     )