fix(tail): resolve before destroy on timeout; move breaking change to Changed

- Swap resolve()/req.destroy() order in the timeout handler so the
  promise settles as ETIMEDOUT before destroy() is called; update the
  comment to accurately describe this ordering rather than the reverse
- Move the --clear breaking change entry into ### Changed per
  Keep-a-Changelog convention (was a floating ### Breaking Changes block)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: nawi-25

CHANGELOG.md

@@ -19,12 +19,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 - **Shadow Git Snapshots (Phase 2):** (Coming Soon) Automatic lightweight git commits before AI edits, allowing `node9 undo`.
 - **`flightRecorder` setting:** New `settings.flightRecorder` flag (default `true`) controls whether the daemon records tool call activity to the flight recorder ring buffer. Can be set to `false` to disable activity recording when the browser dashboard is not in use.
 
-### Breaking Changes
-
-- **`node9 tail --clear` no longer streams after clearing:** Previously `--clear` wiped the ring buffer and then continued tailing live events — visually identical to plain `node9 tail`. It now clears the buffer and exits immediately. To start fresh and watch, chain the commands: `node9 tail --clear && node9 tail --history`. Scripts relying on the old streaming-after-clear behaviour must be updated.
-
 ### Changed
 
+- **`node9 tail --clear` no longer streams after clearing** ⚠️ **Breaking:** Previously `--clear` wiped the ring buffer and then continued tailing live events — visually identical to plain `node9 tail`. It now clears the buffer and exits immediately. To start fresh and watch, chain the commands: `node9 tail --clear && node9 tail --history`. Scripts relying on the old streaming-after-clear behaviour must be updated.
 - **Default mode is now `audit`:** Fresh installs now default to `mode: "audit"` instead of `mode: "standard"`. In audit mode every tool call is approved and logged, with a desktop notification for anything that _would_ have been blocked. This lets teams observe agent behaviour before committing to a blocking policy. Switch to `mode: "standard"` or `mode: "strict"` when you are ready to enforce.
 - **Approval timeout default is now 30 seconds:** `approvalTimeoutMs` defaults to `30000` (was `0` / wait forever). Pending approval prompts now auto-deny after 30 seconds if no human responds, preventing agents from stalling indefinitely.
 - **Cloud approver disabled by default:** `approvers.cloud` defaults to `false`. Cloud (Slack/SaaS) approval must be explicitly opted in via `settings.approvers.cloud: true` after running `node9 login`.

src/tui/tail.ts

@@ -149,10 +149,11 @@ export async function startTail(options: TailOptions = {}): Promise<void> {
         }
       );
       req.setTimeout(2000, () => {
-        // destroy() may fire the error listener once more (ECONNRESET) after this
-        // resolves — req.once ensures the listener is already gone by then.
-        req.destroy();
+        // resolve() before destroy() so the promise settles as ETIMEDOUT first.
+        // destroy() may then emit an error event, but req.once ensures the
+        // listener has already been consumed and won't call resolve() again.
         resolve({ ok: false, code: 'ETIMEDOUT' });
+        req.destroy();
       });
       req.once('error', (err: NodeJS.ErrnoException) => resolve({ ok: false, code: err.code }));
       req.end();