claude-threads 1.35.0 → 1.36.0

package/CHANGELOG.md

@@ -7,6 +7,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.36.0] - 2026-09-10
+
+### Added
+- **`!usage` reports the subscription windows for the seat a thread runs on** (#523, thanks @kaza). The session window and the weekly window, as a bar with the percentage used and the time each one resets; `!usage all` does the same for every account in the `claudeAccounts` pool, which is what tells you whether the pool is actually balanced or whether one seat is carrying everything. It reuses the existing `/usage` probe rather than reading the OAuth endpoint directly, so it costs nothing and needs no extra credentials. API-key accounts report no windows and say so. Login emails stay hidden unless you set `usage.showEmails`.
+- **`bugReports: false` removes the `!bug` path entirely** (#574). A bug report leaves the operator's infrastructure — attached screenshots go to a public anonymous file host, and the body (session context plus recent daemon log lines) becomes an issue on this project's public repository, behind best-effort redaction. `claudeCanExecute` means the agent can trigger it unprompted. Regulated deployments need one switch that closes all of it: the typed command, Claude invoking it itself, the 🐛 error reaction, the approval that files the issue, and the `!help` listing. Defaults to `true`, so nothing changes unless you set it, and it **fails closed** — a malformed value, or a bare `bugReports:`, disables the feature rather than defaulting to on.
+
+### Fixed
+- **A quiet Mattermost channel no longer reconnects in a loop** (#498, thanks @Jadefalkner). A channel with no traffic produces no WebSocket frames, and the heartbeat could not tell that apart from a dead socket: after 90 s of silence it declared the connection dead and reconnected, every 90 to 150 s, all day, each teardown a window in which a message can be missed. The heartbeat now sends a `ping` once the socket has been silent for half the interval, and the server's `pong` counts as activity, so a quiet connection stays up while a genuinely dead one is still torn down on the next tick. Verified against Mattermost 11.7.8, and a socket that accepts sends but never replies still reconnects exactly once. This is the Mattermost half; Node 22+ exposes no frame-level ping visibility to the Slack client, so #498 stays open for that.
+- **Exhausted reconnection attempts now exit instead of leaving the process deaf** (#500, thanks @kaza). When the socket was gone for good the bot stayed alive with nothing listening, and systemd still reported the unit `active`: every health check said healthy while the bot heard nothing. It now exits, so `Restart=always` revives it, which is what every deployment already expects. A fail-loud change by design, and configurable through the reconnect policy.
+
+## [1.35.1] - 2026-09-08
+
+### Fixed
+- **A finished session no longer leaks its event listeners** (#569, thanks @R09722akaBennett). `TypedEventEmitter.removeAllListeners()` forwarded its optional argument as an explicit `undefined`, which Node's `EventEmitter` reads as "the event named undefined", so `MessageManager.dispose()` removed nothing and every session's closures stayed alive. Invisible in CI because Bun 1.3.3 behaves differently from Node there; Bun 1.4.2 agrees with Node and fails the existing tests on the old code.
+- **The resume notice no longer promises that interrupted work continued** (#571, thanks @kaza; item 1 of #533). After a daemon restart or a platform re-enable, a thread whose turn was in flight was told "You can continue where you left off" while nothing continued: `isProcessing` starts false and nothing is sent to the CLI. The notice now says why the resume happened and, for the two daemon-initiated cases, that anything still running did not survive; only a resume a person asked for keeps the invitation. Attribution to the session owner is gone from the boot path, where nobody was there.
+
 ## [1.35.0] - 2026-09-07
 
 ### Changed

package/README.md

@@ -203,6 +203,19 @@ A few more things set it apart. The full Claude Code permission model lives in t
 
 This table is current as of August 2026. All of these products move quickly — if it has gone stale, [open an issue](https://github.com/anneschuth/claude-threads/issues).
 
+## Staying connected
+
+If the connection to Slack or Mattermost drops, the bot reconnects with
+exponential backoff. When those attempts run out it does **not** sit there
+alive with a dead socket — that "active but deaf" state looks healthy to a
+supervisor and looks broken to everyone in the channel. By default it waits a
+minute, resets, and keeps trying, which recovers on its own.
+
+Running under systemd or another supervisor? Set `reconnectPolicy: exit` on
+the platform and the bot shuts down cleanly and exits non-zero instead, so
+`Restart=always` brings it back with a fresh socket. See the
+[Configuration Reference](https://github.com/anneschuth/claude-threads/blob/main/docs/CONFIGURATION.md).
+
 ## Documentation
 
 - **[Setup Guide](https://github.com/anneschuth/claude-threads/blob/main/SETUP_GUIDE.md)** - Creating the bot account on Mattermost or Slack, step by step