test: replace exec directive e2e with pure coverage

src/auto-reply/reply.directive.exec-agent-defaults.test.ts

@@ -1,154 +0,0 @@
-import fs from "node:fs/promises";
-import "./reply.directive.directive-behavior.e2e-mocks.js";
-import { beforeAll, describe, expect, it } from "vitest";
-import {
-  MAIN_SESSION_KEY,
-  installDirectiveBehaviorE2EHooks,
-  makeEmbeddedTextResult,
-  sessionStorePath,
-  withTempHome,
-} from "./reply.directive.directive-behavior.e2e-harness.js";
-import { runEmbeddedPiAgentMock } from "./reply.directive.directive-behavior.e2e-mocks.js";
-import { withFullRuntimeReplyConfig } from "./reply/get-reply-fast-path.js";
-
-let getReplyFromConfig: typeof import("./reply/get-reply.js").getReplyFromConfig;
-
-type ExpectedExecOverrides = {
-  host: "node" | "auto" | "gateway";
-  security: "allowlist" | "deny" | "full";
-  ask: "always" | "off";
-  node: string;
-};
-
-const AGENT_EXEC_DEFAULTS = {
-  host: "node",
-  security: "allowlist",
-  ask: "always",
-  node: "worker-alpha",
-} as const satisfies ExpectedExecOverrides;
-
-const WHATSAPP_EXEC_PROMPT_REQUEST = {
-  Body: "run a command",
-  From: "+1004",
-  To: "+2000",
-  Provider: "whatsapp",
-  SenderE164: "+1004",
-} as const;
-
-const AUTHORIZED_EXEC_DIRECTIVE_REQUEST = {
-  From: "+1004",
-  To: "+2000",
-  CommandAuthorized: true,
-} as const;
-
-function makeAgentExecConfig(home: string) {
-  return withFullRuntimeReplyConfig({
-    agents: {
-      defaults: {
-        model: "anthropic/claude-opus-4-6",
-        humanDelay: { mode: "off" },
-        workspace: `${home}/openclaw`,
-      },
-      list: [
-        {
-          id: "main",
-          tools: {
-            exec: AGENT_EXEC_DEFAULTS,
-          },
-        },
-      ],
-    },
-    channels: { whatsapp: { allowFrom: ["*"] } },
-    session: { store: sessionStorePath(home) },
-  });
-}
-
-async function runExecPrompt(home: string) {
-  await getReplyFromConfig(WHATSAPP_EXEC_PROMPT_REQUEST, {}, makeAgentExecConfig(home));
-}
-
-async function runExecDirective(home: string, body: string) {
-  await getReplyFromConfig(
-    { ...AUTHORIZED_EXEC_DIRECTIVE_REQUEST, Body: body },
-    {},
-    makeAgentExecConfig(home),
-  );
-}
-
-function expectLastExecOverrides(overrides: Partial<ExpectedExecOverrides> = {}) {
-  expect(runEmbeddedPiAgentMock).toHaveBeenCalledOnce();
-  const call = runEmbeddedPiAgentMock.mock.calls[0]?.[0];
-  expect(call?.execOverrides).toEqual({
-    ...AGENT_EXEC_DEFAULTS,
-    ...overrides,
-  });
-}
-
-describe("directive behavior exec agent defaults", () => {
-  installDirectiveBehaviorE2EHooks();
-
-  beforeAll(async () => {
-    ({ getReplyFromConfig } = await import("./reply/get-reply.js"));
-  });
-
-  it("threads per-agent tools.exec defaults into live runs without a persisted session override", async () => {
-    await withTempHome(async (home) => {
-      runEmbeddedPiAgentMock.mockResolvedValue(makeEmbeddedTextResult("done"));
-
-      await runExecPrompt(home);
-
-      expectLastExecOverrides();
-    });
-  });
-
-  it("prefers standalone inline exec directives over per-agent exec defaults on the next live run", async () => {
-    await withTempHome(async (home) => {
-      runEmbeddedPiAgentMock.mockResolvedValue(makeEmbeddedTextResult("done"));
-
-      await runExecDirective(home, "/exec host=auto");
-
-      runEmbeddedPiAgentMock.mockClear();
-
-      await runExecPrompt(home);
-
-      expectLastExecOverrides({ host: "auto" });
-    });
-  });
-
-  it("prefers persisted session exec overrides over per-agent exec defaults", async () => {
-    await withTempHome(async (home) => {
-      runEmbeddedPiAgentMock.mockResolvedValue(makeEmbeddedTextResult("done"));
-      await fs.writeFile(
-        sessionStorePath(home),
-        JSON.stringify({
-          [MAIN_SESSION_KEY]: {
-            sessionId: "main",
-            updatedAt: Date.now(),
-            execHost: "auto",
-          },
-        }),
-        "utf-8",
-      );
-
-      await runExecPrompt(home);
-
-      expectLastExecOverrides({ host: "auto" });
-    });
-  });
-
-  it("replaces a prior deny override with newer exec settings on later turns", async () => {
-    await withTempHome(async (home) => {
-      runEmbeddedPiAgentMock.mockResolvedValue(makeEmbeddedTextResult("done"));
-
-      await runExecDirective(home, "/exec host=gateway security=deny ask=off");
-
-      await runExecDirective(home, "/exec host=gateway security=full ask=always");
-
-      runEmbeddedPiAgentMock.mockClear();
-
-      await runExecPrompt(home);
-
-      expectLastExecOverrides({ host: "gateway", security: "full" });
-    });
-  });
-});

src/auto-reply/reply/get-reply-directives.ts

@@ -1,5 +1,4 @@
 import { listAgentEntries } from "../../agents/agent-scope.js";
-import type { ExecToolDefaults } from "../../agents/bash-tools.js";
 import { DEFAULT_CONTEXT_TOKENS } from "../../agents/defaults.js";
 import { resolveFastModeState } from "../../agents/fast-mode.js";
 import type { ModelAliasIndex } from "../../agents/model-selection.js";
@@ -25,6 +24,7 @@ import {
 } from "./get-reply-directive-aliases.js";
 import { applyInlineDirectiveOverrides } from "./get-reply-directives-apply.js";
 import { clearExecInlineDirectives, clearInlineDirectives } from "./get-reply-directives-utils.js";
+import { type ReplyExecOverrides, resolveReplyExecOverrides } from "./get-reply-exec-overrides.js";
 import { shouldUseReplyFastTestRuntime } from "./get-reply-fast-path.js";
 import { defaultGroupActivation, resolveGroupRequireMention } from "./groups.js";
 import { CURRENT_MESSAGE_MARKER, stripMentions, stripStructuralPrefixes } from "./mentions.js";
@@ -38,8 +38,6 @@ import { stripInlineStatus } from "./reply-inline.js";
 import type { TypingController } from "./typing.js";
 
 type AgentDefaults = NonNullable<OpenClawConfig["agents"]>["defaults"];
-type ExecOverrides = Pick<ExecToolDefaults, "host" | "security" | "ask" | "node">;
-type AgentEntry = NonNullable<NonNullable<OpenClawConfig["agents"]>["list"]>[number];
 
 let commandsRegistryPromise: Promise<typeof import("../commands-registry.runtime.js")> | null =
   null;
@@ -96,7 +94,7 @@ export type ReplyDirectiveContinuation = {
   resolvedVerboseLevel: VerboseLevel | undefined;
   resolvedReasoningLevel: ReasoningLevel;
   resolvedElevatedLevel: ElevatedLevel;
-  execOverrides?: ExecOverrides;
+  execOverrides?: ReplyExecOverrides;
   blockStreamingEnabled: boolean;
   blockReplyChunking?: {
     minChars: number;
@@ -119,33 +117,6 @@ export type ReplyDirectiveContinuation = {
   };
 };
 
-function resolveExecOverrides(params: {
-  directives: InlineDirectives;
-  sessionEntry?: SessionEntry;
-  agentEntry?: AgentEntry;
-}): ExecOverrides | undefined {
-  const host =
-    params.directives.execHost ??
-    (params.sessionEntry?.execHost as ExecOverrides["host"]) ??
-    (params.agentEntry?.tools?.exec?.host as ExecOverrides["host"]);
-  const security =
-    params.directives.execSecurity ??
-    (params.sessionEntry?.execSecurity as ExecOverrides["security"]) ??
-    (params.agentEntry?.tools?.exec?.security as ExecOverrides["security"]);
-  const ask =
-    params.directives.execAsk ??
-    (params.sessionEntry?.execAsk as ExecOverrides["ask"]) ??
-    (params.agentEntry?.tools?.exec?.ask as ExecOverrides["ask"]);
-  const node =
-    params.directives.execNode ??
-    params.sessionEntry?.execNode ??
-    params.agentEntry?.tools?.exec?.node;
-  if (!host && !security && !ask && !node) {
-    return undefined;
-  }
-  return { host, security, ask, node };
-}
-
 export type ReplyDirectiveResult =
   | { kind: "reply"; reply: ReplyPayload | ReplyPayload[] | undefined }
   | { kind: "continue"; result: ReplyDirectiveContinuation };
@@ -566,7 +537,11 @@ export async function resolveReplyDirectives(params: {
   model = applyResult.model;
   contextTokens = applyResult.contextTokens;
   const { directiveAck, perMessageQueueMode, perMessageQueueOptions } = applyResult;
-  const execOverrides = resolveExecOverrides({ directives, sessionEntry, agentEntry });
+  const execOverrides = resolveReplyExecOverrides({
+    directives,
+    sessionEntry,
+    agentExecDefaults: agentEntry?.tools?.exec,
+  });
 
   return {
     kind: "continue",

src/auto-reply/reply/get-reply-exec-overrides.test.ts

@@ -0,0 +1,121 @@
+import { describe, expect, it } from "vitest";
+import type { ModelAliasIndex } from "../../agents/model-selection.js";
+import type { OpenClawConfig } from "../../config/config.js";
+import type { SessionEntry } from "../../config/sessions.js";
+import { parseInlineDirectives } from "./directive-handling.parse.js";
+import { persistInlineDirectives } from "./directive-handling.persist.js";
+import { type ReplyExecOverrides, resolveReplyExecOverrides } from "./get-reply-exec-overrides.js";
+
+const AGENT_EXEC_DEFAULTS = {
+  host: "node",
+  security: "allowlist",
+  ask: "always",
+  node: "worker-alpha",
+} as const satisfies ReplyExecOverrides;
+
+function createSessionEntry(overrides?: Partial<SessionEntry>): SessionEntry {
+  return {
+    sessionId: "main",
+    updatedAt: Date.now(),
+    ...overrides,
+  };
+}
+
+async function persistExecDirective(params: {
+  sessionEntry: SessionEntry;
+  sessionStore: Record<string, SessionEntry>;
+  body: string;
+}) {
+  await persistInlineDirectives({
+    directives: parseInlineDirectives(params.body),
+    cfg: { commands: { text: true } } as OpenClawConfig,
+    agentDir: "/tmp/agent",
+    sessionEntry: params.sessionEntry,
+    sessionStore: params.sessionStore,
+    sessionKey: "agent:main:main",
+    elevatedEnabled: false,
+    elevatedAllowed: false,
+    defaultProvider: "anthropic",
+    defaultModel: "claude-opus-4-6",
+    aliasIndex: { byAlias: new Map(), byKey: new Map() } satisfies ModelAliasIndex,
+    allowedModelKeys: new Set(),
+    provider: "anthropic",
+    model: "claude-opus-4-6",
+    initialModelLabel: "anthropic/claude-opus-4-6",
+    formatModelSwitchEvent: (label) => label,
+    agentCfg: undefined,
+    surface: "whatsapp",
+  });
+}
+
+describe("reply exec overrides", () => {
+  it("uses per-agent exec defaults when session and message are unset", () => {
+    expect(
+      resolveReplyExecOverrides({
+        directives: parseInlineDirectives("run a command"),
+        sessionEntry: createSessionEntry(),
+        agentExecDefaults: AGENT_EXEC_DEFAULTS,
+      }),
+    ).toEqual(AGENT_EXEC_DEFAULTS);
+  });
+
+  it("prefers inline exec directives, then persisted session overrides, then agent defaults", () => {
+    const sessionEntry = createSessionEntry({
+      execHost: "gateway",
+      execSecurity: "deny",
+    });
+
+    expect(
+      resolveReplyExecOverrides({
+        directives: parseInlineDirectives("/exec host=auto security=full"),
+        sessionEntry,
+        agentExecDefaults: AGENT_EXEC_DEFAULTS,
+      }),
+    ).toEqual({
+      ...AGENT_EXEC_DEFAULTS,
+      host: "auto",
+      security: "full",
+    });
+
+    expect(
+      resolveReplyExecOverrides({
+        directives: parseInlineDirectives("run a command"),
+        sessionEntry,
+        agentExecDefaults: AGENT_EXEC_DEFAULTS,
+      }),
+    ).toEqual({
+      ...AGENT_EXEC_DEFAULTS,
+      host: "gateway",
+      security: "deny",
+    });
+  });
+
+  it("resolves the latest persisted exec directive for later turns", async () => {
+    const sessionEntry = createSessionEntry();
+    const sessionStore = { "agent:main:main": sessionEntry };
+
+    await persistExecDirective({
+      sessionEntry,
+      sessionStore,
+      body: "/exec host=gateway security=deny ask=off",
+    });
+    await persistExecDirective({
+      sessionEntry,
+      sessionStore,
+      body: "/exec host=gateway security=full ask=always",
+    });
+
+    expect(
+      resolveReplyExecOverrides({
+        directives: parseInlineDirectives("run a command"),
+        sessionEntry,
+        agentExecDefaults: AGENT_EXEC_DEFAULTS,
+      }),
+    ).toEqual({
+      ...AGENT_EXEC_DEFAULTS,
+      host: "gateway",
+      security: "full",
+      ask: "always",
+    });
+  });
+});

src/auto-reply/reply/get-reply-exec-overrides.ts

@@ -0,0 +1,30 @@
+import type { ExecToolDefaults } from "../../agents/bash-tools.js";
+import type { SessionEntry } from "../../config/sessions.js";
+import type { InlineDirectives } from "./directive-handling.parse.js";
+
+export type ReplyExecOverrides = Pick<ExecToolDefaults, "host" | "security" | "ask" | "node">;
+
+export function resolveReplyExecOverrides(params: {
+  directives: InlineDirectives;
+  sessionEntry?: SessionEntry;
+  agentExecDefaults?: ReplyExecOverrides;
+}): ReplyExecOverrides | undefined {
+  const host =
+    params.directives.execHost ??
+    (params.sessionEntry?.execHost as ReplyExecOverrides["host"]) ??
+    params.agentExecDefaults?.host;
+  const security =
+    params.directives.execSecurity ??
+    (params.sessionEntry?.execSecurity as ReplyExecOverrides["security"]) ??
+    params.agentExecDefaults?.security;
+  const ask =
+    params.directives.execAsk ??
+    (params.sessionEntry?.execAsk as ReplyExecOverrides["ask"]) ??
+    params.agentExecDefaults?.ask;
+  const node =
+    params.directives.execNode ?? params.sessionEntry?.execNode ?? params.agentExecDefaults?.node;
+  if (!host && !security && !ask && !node) {
+    return undefined;
+  }
+  return { host, security, ask, node };
+}