Checkpoint Continuation Issue

apps/ade-cli/src/tuiClient/aggregate.ts

@@ -389,6 +389,7 @@ const SILENCED_EVENT_TYPES = new Set<AgentChatEvent["type"]>([
   "error",
   "tokens",
   "codex_token_usage",
+  "codex_turn_stalled",
   "codex_goal_updated",
   "codex_goal_cleared",
   "pending_input_resolved",

apps/desktop/src/main/services/ai/tools/systemPrompt.test.ts

@@ -295,6 +295,8 @@ describe("buildCodingAgentSystemPrompt", () => {
   it("always includes operating loop, editing rules, and verification rules", () => {
     const result = buildCodingAgentSystemPrompt({ cwd: "/x" });
     expect(result).toContain("## Operating Loop");
+    expect(result).toContain("status checks, interruptions, and tool/subagent timeouts as checkpoints");
+    expect(result).toContain("unless the user explicitly says stop, pause, or only report status");
     expect(result).toContain("## ADE");
     expect(result).toContain("read the matching `ade-*` skill");
     expect(result).toContain("Your ADE capabilities ship as Agent Skills");

apps/desktop/src/main/services/ai/tools/systemPrompt.ts

@@ -237,6 +237,7 @@ export function buildCodingAgentSystemPrompt(args: {
     "3. When you mutate code, keep edits narrow, preserve surrounding conventions, and avoid speculative rewrites.",
     "4. Verify every meaningful change with diffs, tests, type checks, or targeted inspection.",
     "5. Only finish once the task is complete or you are truly blocked.",
+    "6. Treat status checks, interruptions, and tool/subagent timeouts as checkpoints. Give the requested status, then continue the active directive unless the user explicitly says stop, pause, or only report status.",
     "",
     "## User-Facing Progress",
     "Before the first meaningful tool burst, send one short preamble sentence describing what you are about to do.",