fix(coding-agent): evicted resolved completions session URLs

The completions provider stores session state under the request-time resolved base URL, which can differ from the catalog baseUrl for Moonshot, Alibaba Coding Plan, Azure deployments, and similar provider overrides. The model-switch cleanup now evicts the previous provider prefix whenever the switch leaves that completions backend, so those resolved-url keys cannot survive the switch.

packages/coding-agent/CHANGELOG.md

@@ -4,7 +4,7 @@
 
 ### Fixed
 
-- Fixed `openai-completions` provider session state surviving `/model` switches across different providers or base URLs. `AgentSession.#closeProviderSessionsForModelSwitch` only evicted `openai-codex-responses` and `openai-responses:<provider>` keys; entries keyed `openai-completions:<provider>:<baseUrl>:<modelId>` (cached strict-tools disable scopes and reasoning-effort fallbacks for the old transport) lingered indefinitely. Moving away from an `openai-completions` backend now evicts every cached entry sharing the previous `(provider, baseUrl)` pair, while same-backend model toggles keep their cached state ([#3260](https://github.com/can1357/oh-my-pi/issues/3260))
+- Fixed `openai-completions` provider session state surviving `/model` switches across different providers or base URLs. `AgentSession.#closeProviderSessionsForModelSwitch` only evicted `openai-codex-responses` and `openai-responses:<provider>` keys; entries keyed `openai-completions:<provider>:<resolvedBaseUrl>:<modelId>` (cached strict-tools disable scopes and reasoning-effort fallbacks for the old transport) lingered indefinitely. Moving away from an `openai-completions` backend now evicts every cached entry for the previous provider, including entries whose base URL was resolved at request time rather than read from the catalog, while same-backend model toggles keep their cached state ([#3260](https://github.com/can1357/oh-my-pi/issues/3260))
 
 ## [16.1.14] - 2026-06-22
 

packages/coding-agent/src/session/agent-session.ts

@@ -9026,18 +9026,18 @@ export class AgentSession {
 			providerKeys.add(`openai-responses:${nextModel.provider}`);
 		}
 
-		// `openai-completions` sessions are keyed `openai-completions:<provider>:<baseUrl>:<modelId>`
+		// `openai-completions` sessions are keyed `openai-completions:<provider>:<resolvedBaseUrl>:<modelId>`
 		// and cache backend-specific decisions (strict-tools disable scopes, reasoning-effort
-		// fallbacks). When the user moves away from that backend — different `api`, `provider`,
-		// or `baseUrl` — the cached decisions no longer track the active transport. Evict the
-		// full `(provider, baseUrl)` prefix; same-backend model toggles keep their state.
+		// fallbacks). The resolved request base URL can differ from the catalog `model.baseUrl`
+		// (Moonshot env override, Alibaba Coding Plan enterprise URL, Azure deployment URL),
+		// so evict by provider prefix when the user moves away from that completions backend.
 		let completionsPrefixToEvict: string | undefined;
 		if (currentModel.api === "openai-completions") {
 			const currentScope = `${currentModel.provider}:${currentModel.baseUrl ?? ""}`;
 			const nextScope =
 				nextModel.api === "openai-completions" ? `${nextModel.provider}:${nextModel.baseUrl ?? ""}` : undefined;
 			if (currentScope !== nextScope) {
-				completionsPrefixToEvict = `openai-completions:${currentScope}:`;
+				completionsPrefixToEvict = `openai-completions:${currentModel.provider}:`;
 			}
 		}
 

packages/coding-agent/test/agent-session-openai-completions-model-switch.test.ts

@@ -92,7 +92,7 @@ describe("AgentSession openai-completions provider session eviction", () => {
 		expect(session.providerSessionState.has(completionsSessionKey(deepseek))).toBe(false);
 	});
 
-	it("evicts every cached entry under the old (provider, baseUrl) prefix", async () => {
+	it("evicts every cached entry under the old provider prefix", async () => {
 		const deepseekPro = completionsModel("deepseek", "deepseek-v4-pro");
 		const deepseekFlash = completionsModel("deepseek", "deepseek-v4-flash");
 		const cerebras = completionsModel("cerebras", "llama3.1-8b");
@@ -119,6 +119,25 @@ describe("AgentSession openai-completions provider session eviction", () => {
 		expect(session.providerSessionState.has(completionsSessionKey(deepseekFlash))).toBe(false);
 	});
 
+	it("evicts entries whose base URL was resolved at request time", async () => {
+		const moonshot = completionsModel("moonshot", "kimi-k2-thinking");
+		const cerebras = completionsModel("cerebras", "llama3.1-8b");
+		authStorage.setRuntimeApiKey(cerebras.provider, "cerebras-test-key");
+
+		session = buildSession(moonshot);
+
+		const closeSpy = vi.fn();
+		const resolvedBaseUrlKey = `openai-completions:${moonshot.provider}:https://api.moonshot.cn/v1:${moonshot.id}`;
+		session.providerSessionState.set(resolvedBaseUrlKey, {
+			close: closeSpy,
+		} satisfies ProviderSessionState);
+
+		await session.setModel(cerebras);
+
+		expect(closeSpy).toHaveBeenCalledTimes(1);
+		expect(session.providerSessionState.has(resolvedBaseUrlKey)).toBe(false);
+	});
+
 	it("leaves unrelated provider session state untouched", async () => {
 		const deepseek = completionsModel("deepseek", "deepseek-v4-pro");
 		const cerebras = completionsModel("cerebras", "llama3.1-8b");