v0.3.13

• Fully silent wakes need one Hermes setting. A wake that decides to stay silent keeps its final reply out of the chat, but Hermes posts interim text (what the model writes alongside a tool call) by default. The README now says so, with the fix: display.interim_assistant_messages: false. The wake text now tells the model why it must not write alongside tool calls.
• Cleaner check-ins and desk. A check-in's wake header (which the model reads) now says check-in instead of watcher and drops the empty title slot, so it no longer shows a doubled separator. The desk says 1 wake and 1 watcher instead of 1 wakes (#151).
• /pulse why remembers why something was held. After a held occurrence is released and fires, its row still shows why it was held, for example was held: waiting for your reply (#151).
• Shorter /pulse list rows. Long watcher filter values (a task id, message id or emoji list) are shortened in the list; /pulse why still shows them in full (#151).
• /pulse test matches what would be sent. The preview includes the same "(held by …)" note a dispatched turn carries, and a wake you made for owner silence no longer says "Would run now: yes" before you've been quiet long enough (#151).
• Gateway tests on current Hermes. The end-to-end suite now passes against Hermes 0.23.x in local runs (a 0.23.x CI pin comes next) and adds two scenarios: a role taking its own turn, and arming plus a silent verdict when Hermes's default tool search reaches pulse through its tool_call bridge (#155). One 0.23.x difference found along the way is tracked in #157.