fix: use native OpenCode V2 provider runtimes

README.md

@@ -35,7 +35,7 @@ For published plugin versions, the six-hour CI schedule extracts the latest `com
 - **Extract + filter** — model entries (ids, names, reasoning, `inputModalities`, limits) are evaluated out of the minified CLI bundle (`src/catalog.ts`), then intersected with the callable IDs reported by the provider API.
 - **Metadata + costs merge** — vendor context length tightens only a fallback context limit, while `supported_endpoints` selects Messages, Chat Completions, or Responses per model in that preference order; missing endpoint metadata preserves the last known route data. models.dev contributes release date, family, input limit, status, modalities, and cost tiers when present. Tier rows are accepted only when their base prices match this Command Code catalog. Base costs use CLI bundle → official Command Code docs → free SKUs (`$0`) → [models.dev](https://models.dev) reference prices → unmatched placeholder. Anything still unmatched marks the catalog `degraded`. This runs at sync time only; runtime never fetches metadata or prices.
 - **Artifacts** — `models.json` (the catalog), `_version.txt` (upstream version), `manifest.json` (counts, per-source cost stats, `healthy`/`degraded`/`broken` status).
-- **Version-specific registration** — V1 uses the `plugin` config key, `server()` provider map, and V1 auth callback. V2 uses `plugins`, provider/model transforms, and a `commandcode` integration with key and `COMMANDCODE_API_KEY` environment methods. Both retain the Command Code wire model ID. V2 exposes tiered context pricing; V1 emits its supported `context_over_200k` field and keeps flat pricing for other tiers.
+- **Version-specific registration** — V1 uses the `plugin` config key, direct AI SDK packages, the `server()` provider map, and its auth callback. V2 uses `plugins`, OpenCode's native provider runtimes, provider/model transforms, and a `commandcode` integration with key and `COMMANDCODE_API_KEY` environment methods. Both retain the Command Code wire model ID. V2 exposes tiered context pricing; V1 emits its supported `context_over_200k` field and keeps flat pricing for other tiers.
 - **Degraded/cache fallbacks** — a `degraded`/`broken` manifest sets the degraded flag with a reason; an unreadable bundled `models.json` falls back to the last-good cache; auth/connect still registers even with an empty catalog.
 
 ## Quick Start
@@ -64,7 +64,7 @@ The bare package name is unpinned and resolves npm's `latest` release when OpenC
 
 ### 2. No provider block needed
 
-On OpenCode V2 the plugin registers the `commandcode` provider, its models, and its API base URL through the V2 provider API. On V1 the `server` hook fills `provider.commandcode` defaults — `npm: "@ai-sdk/openai-compatible"` plus the Provider API `baseURL`; the plugin package itself is never the SDK `npm` field. Only add a manual provider entry if you need non-default transport options.
+On OpenCode V2 the plugin registers the `commandcode` provider, its models, the native OpenCode runtime, and its API base URL through the V2 provider API. On V1 the `server` hook fills `provider.commandcode` defaults — `npm: "@ai-sdk/openai-compatible"` plus the Provider API `baseURL`; the plugin package itself is never the SDK `npm` field. Only add a manual provider entry if you need non-default transport options.
 
 ### 3. Connect
 

docs/specs/2026-09-28-v2-parity.md

@@ -47,9 +47,13 @@ remain for future models that advertise Responses without another route, but
 the current catalog has no such entry. Other models' Responses tool-call streams
 were not individually live-tested.
 V1 uses `@ai-sdk/openai` for Responses and the Anthropic SDK for Messages. V2
-uses `aisdk:@ai-sdk/openai` for Responses and `aisdk:@ai-sdk/anthropic` for
-Messages. OpenCode 2.0.18 accepts the AI SDK package route; its newer native
-`@opencode/ai` package is not present in that tested image. If an older catalog
+uses OpenCode's native `@opencode/ai/providers/openai-compatible`,
+`@opencode/ai/providers/openai/responses`, and
+`@opencode/ai/providers/anthropic` runtimes. The legacy `aisdk:` adapter did
+not reliably carry provider settings into model requests: image requests and
+an existing image-bearing session returned `Invalid input` until a manual
+provider base URL was supplied. The native runtime preserves the plugin's base
+URL and requires no manual provider block. If an older catalog
 lacks route metadata, Claude retains the Messages fallback and other models
 retain Chat Completions. The current 84-model catalog has 66 models advertising
 both Chat Completions and Responses, 10 Messages-only, and 8 Chat-only; it has
@@ -142,10 +146,11 @@ auth and metadata behavior.
   `/api/integration` exposed the `key` and `env` methods. Both images returned
   `OK` for a GOAT-eligible DeepSeek V4.1 Flash smoke prompt; the current V2
   local pin also returned `OK` for DeepSeek V4 Flash.
-- Responses routing passed both official Docker images against a temporary
-  local Responses stream: V1 sent `POST /provider/v1/responses` via
-  `@ai-sdk/openai`; V2 sent the same route via `aisdk:@ai-sdk/openai`.
-  No live Command Code inference was made for this check.
+- Earlier Responses routing checks passed both official Docker images against a
+  temporary local stream through the legacy adapters. The current mapping unit
+  test selects OpenCode's native Responses runtime for future Responses-only
+  models; the current catalog has no such model, so no live inference exercises
+  that route.
 - Claude plan-denial and disconnected-state results are described above; no
   higher-plan model request was made.
 
@@ -162,5 +167,13 @@ allowed only a `printf*` shell command. The same `printf TOOL_CALL_SMOKE_OK`
 tool request succeeded on the installed local V2.0.18 binary and in the
 official V1.18.30 and V2.0.18 Docker images. These runs verify the formerly
 failing model/variant/tool-call path through the new Chat route; they do not
-validate every model's Responses stream. The user's active config, session, and
-auth files were not changed.
+validate every model's Responses stream.
+
+The native V2 runtime fix was then verified on local OpenCode 2.0.19 with a
+synthetic PNG and an exported clone of the original 121-message failing session;
+both completed with the image still present. Official Docker V2.0.18 returned
+`OK` through the same local plugin, official Docker V1.18.30 still returned
+`OK` through its unchanged V1 runtime, and the Claude Messages route reached
+the provider's expected `MODEL_NOT_IN_PLAN` response. The active config was
+temporarily pinned to the local checkout for these checks; its backup matched
+before the swap, and no real session or auth record was modified.

plugin.ts

@@ -16,10 +16,10 @@ import { ManifestSchema, ModelEntrySchema, PluginFileConfigSchema } from "./src/
 
 const { models: MODELS_PATH, manifest: MANIFEST_PATH, version: VERSION_PATH } = catalogPaths();
 
-// Transport decision (docs/2026-08-28-ci-catalog/spec.md:20): OpenAI-compatible AI SDK +
-// Provider API; this package is the plugin, never the SDK `npm` field. V1 takes the bare
-// npm name; the `aisdk:` prefix is V2-only.
-const PROVIDER_SDK_NPM = "@ai-sdk/openai-compatible";
+// V1 loads the AI SDK package directly. V2 uses OpenCode's native runtime so provider
+// settings such as baseURL survive registry materialization.
+const V1_PROVIDER_SDK_NPM = "@ai-sdk/openai-compatible";
+const V2_PROVIDER_PACKAGE = "@opencode/ai/providers/openai-compatible";
 const PROVIDER_API_BASE_URL = "https://api.commandcode.ai/provider/v1";
 
 type PluginFileConfig = z.infer<typeof PluginFileConfigSchema>;
@@ -239,13 +239,13 @@ export async function server() {
       const pluginCfg = loadPluginConfig();
       const debug = pluginCfg.debugStartupLogs === true;
 
-      if (!cc.npm) cc.npm = PROVIDER_SDK_NPM;
+      if (!cc.npm) cc.npm = V1_PROVIDER_SDK_NPM;
       if (!cc.name) cc.name = "Command Code";
       if (!cc.env) cc.env = ["COMMANDCODE_API_KEY"];
       // V1 resolves `npm` through BunProc.install and calls the first `create*` export, so
       // the SDK package must be the OpenAI-compatible one; give it the Provider API URL
       // unless the user already configured another baseURL (custom npm stays untouched).
-      if (cc.npm === PROVIDER_SDK_NPM) {
+      if (cc.npm === V1_PROVIDER_SDK_NPM) {
         if (cc.options === undefined) cc.options = {};
         if (typeof cc.options === "object" && cc.options !== null) {
           const options = cc.options as Record<string, unknown>;
@@ -307,7 +307,7 @@ async function setup(ctx: any): Promise<void> {
           activation: "auto",
           integrationID: "commandcode",
           env: ["COMMANDCODE_API_KEY"],
-          package: `aisdk:${PROVIDER_SDK_NPM}`,
+          package: V2_PROVIDER_PACKAGE,
           settings: {
             baseURL: PROVIDER_API_BASE_URL,
           },

src/v2models.ts

@@ -92,8 +92,8 @@ export function toV2Model(entry: ModelEntry): V2Model {
     },
   };
   const api = modelApi(entry);
-  if (api === "responses") model.package = "aisdk:@ai-sdk/openai";
-  else if (api === "messages") model.package = "aisdk:@ai-sdk/anthropic";
+  if (api === "responses") model.package = "@opencode/ai/providers/openai/responses";
+  else if (api === "messages") model.package = "@opencode/ai/providers/anthropic";
   return model;
 }
 

tests/unit/plugin-v2.test.ts

@@ -65,7 +65,7 @@ test("V2 setup adds provider with models when missing", async () => {
   expect(calls.map((call) => call.name)).toEqual(["provider.transform", "integration.transform"]);
   expect(added.info.id).toBe("commandcode");
   expect(added.info.activation).toBe("auto");
-  expect(added.info.package).toBe("aisdk:@ai-sdk/openai-compatible");
+  expect(added.info.package).toBe("@opencode/ai/providers/openai-compatible");
   expect(added.info.integrationID).toBe("commandcode");
   expect(added.info.env).toEqual(["COMMANDCODE_API_KEY"]);
   expect(added.info.settings.apiKey).toBeUndefined();

tests/unit/v2models.test.ts

@@ -53,7 +53,9 @@ test("Claude catalog models use the Anthropic Messages package", () => {
     "claude-haiku-4-5-20251001",
   ];
   const models = generateV2Models(ids.map((id) => ({ ...base, id })));
-  expect(models.map((model) => model.package)).toEqual(ids.map(() => "aisdk:@ai-sdk/anthropic"));
+  expect(models.map((model) => model.package)).toEqual(
+    ids.map(() => "@opencode/ai/providers/anthropic"),
+  );
 });
 
 test("V2 maps advertised Responses and Messages packages", () => {
@@ -63,7 +65,7 @@ test("V2 maps advertised Responses and Messages packages", () => {
       id: "deepseek/deepseek-v4-flash",
       supported_endpoints: ["/provider/v1/responses"],
     }).package,
-  ).toBe("aisdk:@ai-sdk/openai");
+  ).toBe("@opencode/ai/providers/openai/responses");
   expect(
     toV2Model({
       ...base,
@@ -77,7 +79,7 @@ test("V2 maps advertised Responses and Messages packages", () => {
       id: "vendor/messages-model",
       supported_endpoints: ["/v1/messages"],
     }).package,
-  ).toBe("aisdk:@ai-sdk/anthropic");
+  ).toBe("@opencode/ai/providers/anthropic");
 });
 
 test("V2 maps models.dev release and catalog metadata", () => {