fix(coding-agent): classify user-invoked /skill turns under auto thinking

A user-invoked /skill:<name> reaches the session as a user-attributed skill custom message (role custom, attribution user) whose expanded SKILL.md body is the task prompt. The auto-thinking gate in #promptWithMessage only accepted role === "user", so these turns skipped classifyDifficulty/applyAutoThinkingLevel and the effort stayed stuck on pending auto.

Broaden the gate to also accept user-invoked skill prompts via the now-exported isUserInvokedSkillPrompt helper; agent-originated and autoload skill injections stay excluded.

Fixes #8554

packages/coding-agent/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed `defaultThinkingLevel: auto` skipping classification for user-invoked `/skill:<name>` turns, which left the effort stuck on pending `auto`; user-attributed skill prompts now classify like any user turn while agent/autoload injections stay excluded ([#8554](https://github.com/can1357/oh-my-pi/issues/8554)).
+
 ## [17.3.4] - 2026-08-14
 
 ### Changed

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

@@ -297,6 +297,7 @@ import {
 	type InterruptedThinkingDetails,
 	isEmptyErrorTurn,
 	isUserInterruptAbort,
+	isUserInvokedSkillPrompt,
 	logProviderTurnError,
 	normalizeCustomMessagePayload,
 	type PythonExecutionMessage,
@@ -5568,10 +5569,14 @@ export class AgentSession {
 			}
 
 			// Auto thinking: classify this real user turn and set the effective level
-			// before the model request. Synthetic/tool-continuation turns (developer/
-			// custom roles) and non-auto sessions are skipped. Never blocks the turn —
-			// failures fall back to a concrete level inside the helper.
-			if (this.isAutoThinking && message.role === "user") {
+			// before the model request. A user-invoked `/skill:<name>` arrives as a
+			// user-attributed skill custom message whose expanded body is the task
+			// prompt, so it counts as a user turn. Synthetic/tool-continuation turns
+			// (developer roles), agent-originated or autoloaded skill injections, and
+			// non-auto sessions are skipped. Never blocks the turn — failures fall
+			// back to a concrete level inside the helper.
+			const isUserTurn = message.role === "user" || (message.role === "custom" && isUserInvokedSkillPrompt(message));
+			if (this.isAutoThinking && isUserTurn) {
 				await this.#models.applyAutoThinkingLevel(expandedText, generation);
 				if (this.#promptGeneration !== generation) {
 					return;

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

@@ -1085,7 +1085,8 @@ function customMessageContentToLlmContent(content: CustomMessage["content"]): (T
 	return typeof content === "string" ? [{ type: "text", text: content }] : content;
 }
 
-function isUserInvokedSkillPrompt(message: CustomMessage): boolean {
+/** True for a `/skill:<name>` prompt the user invoked directly (attribution `user`), as opposed to an agent/autoload injection. */
+export function isUserInvokedSkillPrompt(message: CustomMessage): boolean {
 	return message.customType === SKILL_PROMPT_MESSAGE_TYPE && message.attribution === "user";
 }
 

packages/coding-agent/test/agent-session-role-thinking.test.ts

@@ -8,6 +8,7 @@ import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
 import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
+import { SKILL_PROMPT_MESSAGE_TYPE } from "@oh-my-pi/pi-coding-agent/session/messages";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
 import {
 	AUTO_THINKING,
@@ -374,6 +375,63 @@ describe("AgentSession role model thinking behavior", () => {
 		expect(session.agent.state.thinkingLevel).toBe(Effort.Medium);
 	});
 
+	it("classifies a user-invoked /skill turn under auto (resolves concrete effort)", async () => {
+		const model = getAnthropicModelOrThrow("claude-sonnet-4-5");
+		await createSession({
+			initialModelId: model.id,
+			initialThinkingLevel: Effort.High,
+			modelRoles: { default: `${model.provider}/${model.id}` },
+		});
+		const promptSpy = vi.spyOn(session.agent, "prompt").mockResolvedValue(undefined);
+		const classifierSpy = vi.spyOn(autoThinkingClassifier, "classifyDifficulty").mockResolvedValue(Effort.Medium);
+
+		session.setThinkingLevel(AUTO_THINKING);
+		expect(session.autoResolvedThinkingLevel()).toBeUndefined();
+
+		// A /skill:<name> invocation reaches the session as a user-attributed
+		// custom message, not a `user` role. It is still a real user turn.
+		await session.promptCustomMessage({
+			customType: SKILL_PROMPT_MESSAGE_TYPE,
+			content: "Expanded SKILL.md body: implement the focused parser fix",
+			display: true,
+			details: { name: "implement", path: "/skills/implement/SKILL.md", args: "the parser" },
+			attribution: "user",
+		});
+
+		expect(classifierSpy).toHaveBeenCalledTimes(1);
+		expect(classifierSpy.mock.calls[0]?.[0]).toContain("implement the focused parser fix");
+		expect(promptSpy).toHaveBeenCalledTimes(1);
+		expect(session.configuredThinkingLevel()).toBe(AUTO_THINKING);
+		expect(session.thinkingLevel).toBe(Effort.Medium);
+		expect(session.autoResolvedThinkingLevel()).toBe(Effort.Medium);
+		expect(session.agent.state.thinkingLevel).toBe(Effort.Medium);
+	});
+
+	it("does not classify an agent-originated skill custom message under auto", async () => {
+		const model = getAnthropicModelOrThrow("claude-sonnet-4-5");
+		await createSession({
+			initialModelId: model.id,
+			initialThinkingLevel: Effort.High,
+			modelRoles: { default: `${model.provider}/${model.id}` },
+		});
+		vi.spyOn(session.agent, "prompt").mockResolvedValue(undefined);
+		const classifierSpy = vi.spyOn(autoThinkingClassifier, "classifyDifficulty").mockResolvedValue(Effort.Medium);
+
+		session.setThinkingLevel(AUTO_THINKING);
+
+		// Autoloaded / agent-originated skill injections must stay excluded.
+		await session.promptCustomMessage({
+			customType: SKILL_PROMPT_MESSAGE_TYPE,
+			content: "Autoloaded skill body",
+			display: false,
+			details: { name: "autoload", path: "/skills/autoload/SKILL.md" },
+			attribution: "agent",
+		});
+
+		expect(classifierSpy).not.toHaveBeenCalled();
+		expect(session.autoResolvedThinkingLevel()).toBeUndefined();
+	});
+
 	it("keeps auto active on resume (pending until the next turn reclassifies)", async () => {
 		const model = getAnthropicModelOrThrow("claude-sonnet-4-5");
 		const agent = new Agent({