fix: track opencode session switches - #2455

src/integration/assets/opencode/herdr-tui-session.test.ts (1)

112-125: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider covering the exhausted-retry branch.

SELECTION_RETRY_DELAYS_MS has three entries. After the last retry, herdr-tui-session.js sets nextReportAt to Number.POSITIVE_INFINITY and stops reporting the same selection. No test asserts that the reporting stops. A test that waits past the final delay and asserts a stable request count would lock in that boundary.

src/terminal/state.rs (2)

1374-1375: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider recording the stale session when a selection clears suppression.

selection_can_reconcile removes the suppression entry for the source without inspecting suppressed.reason. Every other removal site distinguishes HookClear from ProcessExit. clear_full_lifecycle_hook_suppression_for_detected_agent records the suppressed session_ref as a stale session before it drops a HookClear entry (lines 1098-1106). This path drops it silently, so a later report that carries the previous session ref is not filtered by full_lifecycle_hook_report_matches_stale_session.

The opencode_cross_talk guard at lines 863-870 rejects that report while the process is present and the anchor differs, so the reachable exposure is small. Recording the stale session here would keep the two paths consistent.

♻️ Suggested change

         if selection_can_reconcile {
-            self.suppressed_full_lifecycle_hook_reports.remove(&source);
+            if let Some(suppressed) = self.suppressed_full_lifecycle_hook_reports.remove(&source) {
+                if let Some(suppressed_ref) = suppressed
+                    .session_ref
+                    .filter(|suppressed_ref| suppressed_ref != &session_ref)
+                {
+                    self.remember_stale_full_lifecycle_hook_session(
+                        source.clone(),
+                        suppressed.agent_label,
+                        suppressed_ref,
+                    );
+                }
+            }
         } else if full_lifecycle_source && unsequenced_selection {

---

4620-4656: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider adding a boundary test for cross-talk while the process is absent.

opencode_cross_talk at lines 863-870 requires process_present. When the process is absent, a report carrying a different session ref still reaches the buffering path at lines 906-953. That asymmetry is intentional, but no test pins it. A test that clears the detected agent and then sends a mismatched sequenced report would document the boundary.