claude-threads 1.38.0 → 1.39.0

package/CHANGELOG.md

@@ -5,6 +5,24 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.39.0] - 2026-09-25
+
+### Added
+- **Per-platform `toolActivity`: `summary` or `hidden` instead of the tool stream** (#534, thanks @kaza). `summary` replaces the inline `Bash …` / `↳ ✓` lines with one live line at the top of the turn's post — `🔧 12 tools · 40 s · Bash`, naming the tool most recently started so the line stays a liveness signal; `hidden` says nothing about tools at all. `toolDetails: thread` keeps the full rendering in a thread under the reply, `none` drops it. Defaults are unchanged (`full` / `none`). Permission prompts, plan approvals, questions, task lists and errors are untouched in every mode. Part of #505. Hardened before release in #616: the turn marker keeps the summary line, details in a thread session arrive after the whole reply, and long or rate-limited turns lose no tool lines.
+
+### Fixed
+- **Long replies are no longer cut off at the platform limit** (#616). A reply that reached the chat in one large burst, with no post of its own yet, was truncated past the limit (about 12K characters on Slack, 16K on Mattermost), with "(truncated)" at the end. It is now split over several posts; a code block cut in two is closed and reopened with its language. Normal streaming replies are unchanged.
+- **A refused post no longer loses part of a long reply** (#616). When the platform refused one post of a reply split over several (a rate limit, for instance), the text of that post was dropped once a later one went through, and anything Claude streamed meanwhile could be wiped. The refused part is now kept and posted on the next flush.
+
+### Changed
+- `hono` 4.13.7 → 4.13.8 (#608).
+
+## [1.38.1] - 2026-09-24
+
+### Fixed
+- **Sessions start on Windows again** (#600). On Windows the bot spawned the Claude CLI with `shell: true`, which makes Node join the arguments into one cmd.exe command line without escaping them. The `|` in the session context passed as `--append-system-prompt` was read as a pipe, so every session exited with code 255 before Claude answered. Any other shell character in an argument had the same effect, including text from a user's message in the one-shot haiku calls, which made this a command-injection path too. Spawning now goes through the `cross-spawn` package: it runs a native `claude.exe` directly and escapes arguments for cmd.exe when the target is an npm `.cmd` shim. The auto-update installer (which spawned `npm.cmd` without a shell, rejected by Node since 20.12) and the self-respawn after an update use the same path. A new `windows-latest` CI job round-trips shell characters through a real `.cmd` shim, so this path is tested on Windows from now on.
+- **Auto-restart on Windows no longer picks the WSL `bash.exe`** (#600). `bash` on PATH is often the WSL launcher, which runs the daemon script inside Linux, where the Windows paths it gets do not exist. The bot now looks for Git for Windows' bash and starts without auto-restart, with a notice, when there is none. The daemon script also stopped using `eval` to start the bot, which dropped the backslashes from a Windows path and split any path containing a space, on macOS and Linux too.
+
 ## [1.38.0] - 2026-09-24
 
 ### Added

package/README.md

@@ -59,7 +59,7 @@
 - **Files both ways** - Drop any file into the chat for Claude to read, with full multimodal for images and PDFs; Claude posts screenshots, plots, or PDFs back with `send_file` (100 MB cap)
 - **Voice notes** - With `transcription:` configured (ElevenLabs Scribe), an audio clip is transcribed before Claude sees it and the transcript is echoed into the thread; see [Configuration](docs/CONFIGURATION.md#transcription-transcription)
 - **Quiet mode and verbosity dials** - `!mentions on` makes a session respond only when mentioned; session headers and the channel sticky each have `full`/`minimal`/`hidden` modes
-- **Runs on macOS, Linux, and Windows** - Windows via Git Bash or WSL
+- **Runs on macOS, Linux, and Windows** - natively or inside WSL; on native Windows, auto-restart needs Git for Windows' bash
 - **Auto-update** - The bot watches npm for new versions; `!update now` applies one from chat
 
 ## What Claude can do in your chat

package/bin/claude-threads-daemon

@@ -52,12 +52,12 @@ if [ -n "$CLAUDE_THREADS_BIN" ]; then
             echo "[daemon] Error: node not found in PATH" >&2
             exit 1
         fi
-        CLAUDE_THREADS_CMD="node $CLAUDE_THREADS_BIN"
+        CLAUDE_THREADS_CMD=(node "$CLAUDE_THREADS_BIN")
     else
-        CLAUDE_THREADS_CMD="$CLAUDE_THREADS_BIN"
+        CLAUDE_THREADS_CMD=("$CLAUDE_THREADS_BIN")
     fi
 else
-    CLAUDE_THREADS_CMD="claude-threads"
+    CLAUDE_THREADS_CMD=(claude-threads)
 fi
 
 # Track restart count
@@ -115,7 +115,7 @@ log() {
 }
 
 log "Starting claude-threads daemon"
-log "Binary: $CLAUDE_THREADS_CMD"
+log "Binary: ${CLAUDE_THREADS_CMD[*]}"
 log "Max restarts: ${MAX_RESTARTS:-unlimited}"
 log "Restart delay: ${RESTART_DELAY}s"
 log "Restart on error: $RESTART_ON_ERROR"
@@ -140,9 +140,10 @@ while true; do
     log "Starting claude-threads (restart #$restart_count)..."
 
     # Run claude-threads with any remaining arguments
-    # Note: CLAUDE_THREADS_CMD may be "bun /path/to/file.js" so we use eval
+    # An array, not eval: eval re-parses the path, dropping the backslashes
+    # of a Windows path and splitting one that contains a space (#600).
     set +e
-    eval $CLAUDE_THREADS_CMD '"$@"' &
+    "${CLAUDE_THREADS_CMD[@]}" "$@" &
     child_pid=$!
     wait $child_pid
     exit_code=$?