Skip to content

Open-source SEO, GEO, and marketing skills for AI agents.

fix(chat): show "Thinking" between tool calls, not "Starting"

Yuting Zhongb8e0d2b0da9a
diff --git a/src/components/live-transcript.tsx b/src/components/live-transcript.tsx
index a135f63c..e15ede22 100644
--- a/src/components/live-transcript.tsx
+++ b/src/components/live-transcript.tsx
@@ -1191,6 +1191,23 @@ function deriveWorkingView(input: {
mood: "writing",
};
}
+ // Between-tools window: SSE has seen at least one tool finish this
+ // turn but the model hasn't started the next thing yet (no in-flight
+ // pending tool, no streaming text, no fresh JSONL event because Codex
+ // buffers the file until end-of-turn). Without this branch we'd fall
+ // through to "Starting", which is wrong — the agent IS thinking, just
+ // about its next move. Surface the last tool so the user has context.
+ if (pendingTools.length > 0) {
+ const lastDone = [...pendingTools].reverse().find((t) => t.done);
+ return {
+ headline: "Thinking",
+ subtitle: lastDone
+ ? `${formatToolName(lastDone.name)} ${lastDone.ok ? "✓" : "failed"} — picking next step`
+ : null,
+ phases,
+ mood: "waiting",
+ };
+ }
if (!lastEvent || lastEvent.kind === "user_message" || lastEvent.kind === "unknown") {
const lifecycleSummary = lifecyclePhase
? humanLifecyclePhase(lifecyclePhase)