fix(opencode-plugin): track active release branch in CI + align combo-id fixture

The opencode-plugin CI workflow trigger was pinned to release/v3.8.2 since
its creation, so it stopped firing once the active release line moved on —
the workflow never ran mid-cycle. Switch push/pull_request branch filters
to release/** so it tracks whatever release branch is active.

Also align the provider.test.ts fixture expectations with the combo-id
contract from #10345/#10821: mapRawModelToModelV2 leaves bare combo ids
(owned_by: "combo") unprefixed so OpenCode's `-m <plugin>/<combo>` lookup
resolves them directly. The two assertions still expected the old
provider-prefixed form.

Closes #11291
Closes #11292

.github/workflows/opencode-plugin-ci.yml

@@ -2,11 +2,11 @@ name: opencode-plugin CI
 
 on:
   push:
-    branches: [main, release/v3.8.2]
+    branches: [main, "release/**"]
     paths:
       - "@omniroute/opencode-plugin/**"
   pull_request:
-    branches: [main, release/v3.8.2]
+    branches: [main, "release/**"]
     paths:
       - "@omniroute/opencode-plugin/**"
     types: [opened, synchronize, reopened, ready_for_review]

@omniroute/opencode-plugin/tests/provider.test.ts

@@ -104,7 +104,10 @@ test("models: extracts apiKey from ctx.auth (type=api) and calls fetcher with it
   // #6859: dynamic-hook catalog keys use the unprefixed omnirouteProviderId
   // ("omniroute"), not the OC-gate-prefixed hook.id ("opencode-omniroute") —
   // that prefix must never leak into anything OmniRoute's server parses.
-  assert.ok(out["omniroute/claude-primary"]);
+  // #10345/#10821: bare combo ids (owned_by: "combo") stay unprefixed —
+  // OpenCode looks up `-m <plugin>/<combo>` as model id `<combo>` under the
+  // plugin provider, so `claude-primary` here carries no provider prefix.
+  assert.ok(out["claude-primary"]);
 });
 
 test("models: returns {} when ctx.auth is null/undefined/wrong-type/empty-key", async () => {
@@ -159,11 +162,15 @@ test("models: maps a sample /v1/models entry to ModelV2 (sanity)", async () => {
   // omnirouteProviderId ("omniroute") — the OC-gate prefix ("opencode-")
   // must stay OC-internal (hook.id / AuthHook.provider) and never leak into
   // anything OmniRoute's own server parses for credential lookup.
-  const claude = out["omniroute/claude-primary"];
+  // #10345/#10821: bare **combo** ids (owned_by: "combo", e.g.
+  // "claude-primary") must also stay unprefixed — OpenCode looks up
+  // `-m <plugin>/<combo>` as model id `<combo>` under the plugin provider.
+  const claude = out["claude-primary"];
   assert.ok(claude, "claude-primary present");
-  // `mapRawModelToModelV2` stamps the provider prefix on the id so OC's
-  // static-catalog reader resolves `(providerID, modelID)` from the key.
-  assert.equal(claude.id, "omniroute/claude-primary");
+  // `mapRawModelToModelV2` leaves bare combo ids unprefixed (see
+  // src/index.ts mapRawModelToModelV2) so OC's `-m <plugin>/<combo>` lookup
+  // resolves the combo id directly.
+  assert.equal(claude.id, "claude-primary");
   assert.equal(claude.name, "claude-primary");
   assert.equal(claude.providerID, "omniroute");
   assert.equal(claude.api.id, "openai-compatible");

open-sse/handlers/chatCore.ts

@@ -20,7 +20,6 @@ import { checkSemanticCache } from "./chatCore/semanticCache.ts";
 import { checkLifecycle, resolveLifecycle } from "./chatCore/modelLifecyclePolicy.ts";
 import {
   shouldDefaultAllowClassifier,
-  detectClassifierFormat,
   buildDefaultAllowClaudeMessage,
 } from "./chatCore/claudeClassifierCompat.ts";
 import { applyClientUsageBuffer } from "./chatCore/clientUsageBuffer.ts";
@@ -779,12 +778,11 @@ export async function handleChatCore({
         classifierSettings.claudeClassifierCompat as string | undefined
       )
     ) {
-      const classifierFormat = detectClassifierFormat(body as Record<string, unknown>);
       log?.warn?.(
         "CHAT",
-        `classifier compat=${classifierSettings.claudeClassifierCompat} format=${classifierFormat} | short-circuit default-allow`
+        `classifier compat=${classifierSettings.claudeClassifierCompat} | short-circuit default-allow`
       );
-      return buildDefaultAllowClaudeMessage(requestedModel, classifierFormat);
+      return buildDefaultAllowClaudeMessage(requestedModel);
     }
   }
 

open-sse/handlers/chatCore/claudeClassifierCompat.ts

@@ -24,19 +24,14 @@ const SECURITY_MONITOR_MARKER = "You are a security monitor for autonomous AI co
 
 export type ClaudeClassifierCompatMode = "off" | "auto" | "always";
 
-/** The two synthetic-response shapes Claude Code's classifier can expect. */
-export type ClaudeClassifierFormat = "block" | "severity";
-
 function extractSystemTexts(body: Record<string, unknown> | null | undefined): string[] {
   const system = body?.system;
   if (typeof system === "string") return [system];
   if (Array.isArray(system)) {
     return system
-      .map((part) =>
-        part && typeof (part as { text?: unknown }).text === "string"
-          ? (part as { text: string }).text
-          : ""
-      )
+      .map((part) => (part && typeof (part as { text?: unknown }).text === "string"
+        ? ((part as { text: string }).text)
+        : ""))
       .filter(Boolean);
   }
   return [];
@@ -65,29 +60,6 @@ export function shouldDefaultAllowClassifier(
   return extractSystemTexts(body).some((text) => text.includes(SECURITY_MONITOR_MARKER));
 }
 
-/**
- * Detect which synthetic-response shape the classifier request expects.
- *
- * Newer Claude Code builds send a "severity classifier" variant of the same internal
- * request: it carries `stop_sequences: [..., "</severity>", ...]` and parses a
- * `<severity>N</severity>` reply instead of `<block>no</block>`/`<block>yes</block>`.
- * Feeding it the legacy `<block>no</block>` shape is unparseable, so it retries both
- * stages and then fails closed — the same "blocking it for safety" failure this compat
- * shim exists to avoid. Only `stop_sequences` distinguishes the two shapes; callers
- * should only consult this after `shouldDefaultAllowClassifier` has already confirmed
- * the request is the classifier (via the system-prompt marker), so an unrelated app
- * that merely happens to use `</severity>` as a stop token is never affected (#8189).
- */
-export function detectClassifierFormat(
-  body: Record<string, unknown> | null | undefined
-): ClaudeClassifierFormat {
-  const stopSequences = body?.stop_sequences;
-  if (Array.isArray(stopSequences) && stopSequences.includes("</severity>")) {
-    return "severity";
-  }
-  return "block";
-}
-
 /**
  * Build the synthetic Claude `message` ALLOW response. Always returns a plain JSON
  * body (matching the upstream reference implementation) — Claude Code's classifier
@@ -95,10 +67,7 @@ export function detectClassifierFormat(
  * satisfies both streaming and non-streaming callers without needing to plumb a
  * synthetic SSE encoding through the streaming/sseToJson/non-streaming handlers.
  */
-export function buildDefaultAllowClaudeMessage(
-  model?: string | null,
-  format: ClaudeClassifierFormat = "block"
-): {
+export function buildDefaultAllowClaudeMessage(model?: string | null): {
   success: true;
   response: Response;
 } {
@@ -107,12 +76,7 @@ export function buildDefaultAllowClaudeMessage(
     type: "message",
     role: "assistant",
     model: model || "claude-3-5-sonnet-20241022",
-    content: [
-      {
-        type: "text",
-        text: format === "severity" ? "<severity>0</severity>" : "<block>no</block>",
-      },
-    ],
+    content: [{ type: "text", text: "<block>no</block>" }],
     stop_reason: "end_turn",
     stop_sequence: null,
     usage: { input_tokens: 1, output_tokens: 1 },

tests/unit/claude-classifier-compat.test.ts

@@ -25,8 +25,9 @@ process.env.DATA_DIR = TEST_DATA_DIR;
 const core = await import("../../src/lib/db/core.ts");
 const { updateSettings } = await import("../../src/lib/db/settings.ts");
 const { handleChatCore } = await import("../../open-sse/handlers/chatCore.ts");
-const { shouldDefaultAllowClassifier, detectClassifierFormat, buildDefaultAllowClaudeMessage } =
-  await import("../../open-sse/handlers/chatCore/claudeClassifierCompat.ts");
+const { shouldDefaultAllowClassifier, buildDefaultAllowClaudeMessage } = await import(
+  "../../open-sse/handlers/chatCore/claudeClassifierCompat.ts"
+);
 const { FORMATS } = await import("../../open-sse/translator/formats.ts");
 
 const originalFetch = globalThis.fetch;
@@ -57,14 +58,6 @@ const CLASSIFIER_BODY = {
   max_tokens: 8,
 };
 
-// Newer Claude Code builds send a "severity classifier" variant of the same internal
-// request: same security-monitor marker, but `stop_sequences` carries `</severity>`
-// instead of `</block>`, and it expects a `<severity>N</severity>` reply (#11289).
-const SEVERITY_CLASSIFIER_BODY = {
-  ...CLASSIFIER_BODY,
-  stop_sequences: ["</severity>"],
-};
-
 test.after(() => {
   globalThis.fetch = originalFetch;
   core.resetDbInstance();
@@ -130,12 +123,7 @@ test("detector: always does NOT fire for normal chat without classifier marker (
 
 test("detector: always fires when classifier marker is present", () => {
   const classifier = {
-    system: [
-      {
-        type: "text",
-        text: "You are a security monitor for autonomous AI coding agents. Evaluate the following action.",
-      },
-    ],
+    system: [{ type: "text", text: "You are a security monitor for autonomous AI coding agents. Evaluate the following action." }],
     stop_sequences: ["</block>"],
   };
   assert.equal(
@@ -145,21 +133,6 @@ test("detector: always fires when classifier marker is present", () => {
   );
 });
 
-// ─── Pure detector: detectClassifierFormat (#11289) ──────────────────────────
-
-test("format detector: defaults to 'block' for the legacy </block> classifier shape", () => {
-  assert.equal(detectClassifierFormat(CLASSIFIER_BODY), "block");
-});
-
-test("format detector: returns 'severity' when stop_sequences carries </severity>", () => {
-  assert.equal(detectClassifierFormat(SEVERITY_CLASSIFIER_BODY), "severity");
-});
-
-test("format detector: defaults to 'block' when stop_sequences is missing/empty", () => {
-  assert.equal(detectClassifierFormat({}), "block");
-  assert.equal(detectClassifierFormat({ stop_sequences: [] }), "block");
-});
-
 // ─── Pure builder: buildDefaultAllowClaudeMessage ────────────────────────────
 
 test("builder: synthetic message text STARTS WITH <block>no</block>", async () => {
@@ -182,16 +155,6 @@ test("builder: synthetic message text STARTS WITH <block>no</block>", async () =
   assert.ok(!text.includes("<block>yes"), "must not signal BLOCK");
 });
 
-test("builder: format='severity' returns <severity>0</severity> (#11289)", async () => {
-  const built = buildDefaultAllowClaudeMessage("claude-3-5-haiku-20241022", "severity");
-  assert.equal(built.success, true);
-  const payload = (await built.response.json()) as {
-    content: Array<{ type: string; text?: string }>;
-  };
-  const text = payload.content.find((b) => b.type === "text")?.text ?? "";
-  assert.equal(text, "<severity>0</severity>");
-});
-
 // ─── Handler-level: end-to-end short-circuit through handleChatCore ──────────
 
 test("handler: claudeClassifierCompat=auto short-circuits WITHOUT calling upstream, text starts with <block>no</block>", async () => {
@@ -233,44 +196,3 @@ test("handler: claudeClassifierCompat=auto short-circuits WITHOUT calling upstre
     globalThis.fetch = originalFetch;
   }
 });
-
-test("handler: claudeClassifierCompat=auto emits <severity>0</severity> for the severity-classifier shape (#11289)", async () => {
-  await updateSettings({ claudeClassifierCompat: "auto" });
-
-  let fetchCalls = 0;
-  globalThis.fetch = (async () => {
-    fetchCalls++;
-    throw new Error("upstream fetch should NOT be called when the classifier short-circuits");
-  }) as typeof fetch;
-
-  try {
-    const result = await handleChatCore({
-      body: structuredClone(SEVERITY_CLASSIFIER_BODY),
-      modelInfo: { provider: "openai", model: "gpt-4o-mini", extendedContext: false },
-      credentials: { apiKey: "sk-test", providerSpecificData: {} },
-      log: noopLog(),
-      clientRawRequest: {
-        endpoint: "/v1/messages",
-        body: structuredClone(SEVERITY_CLASSIFIER_BODY),
-        headers: new Headers({ accept: "application/json" }),
-      },
-      userAgent: "unit-test",
-    });
-
-    assert.equal(fetchCalls, 0, "upstream fetch must NOT be called");
-    assert.equal(result.success, true, "handleChatCore must report success");
-    const payload = (await (result as { response: Response }).response.json()) as {
-      type: string;
-      content: Array<{ type: string; text?: string }>;
-    };
-    assert.equal(payload.type, "message");
-    const text = payload.content.find((b) => b.type === "text")?.text ?? "";
-    assert.equal(
-      text,
-      "<severity>0</severity>",
-      `expected severity-classifier response to be <severity>0</severity>, got: ${text}`
-    );
-  } finally {
-    globalThis.fetch = originalFetch;
-  }
-});