π CodeBuddy Code v2.168.2 Release β
Note: v2.168.1 was not successfully released. This version re-releases all of its content.
π¦ Version Information β
| Component | Version |
|---|---|
| CodeBuddy Code CLI | v2.168.2 |
| Agent SDK JS | v0.3.282 |
| Agent SDK Python | v0.3.281 |
π§ Improvements β
New TUI Sessions Default to PTC Mode β
When a new blank session is created in the interactive terminal without an explicitly specified Agent, a configured default mode, or a saved previous selection, it now defaults to PTC mode.
- After switching back to standard mode with
/agent-mode, the choice is remembered, and the next launch will not automatically switch back to PTC. - Use
--agent clito explicitly start in standard mode. - SDK, non-interactive invocations, ACP, background sessions, and resumed sessions keep their existing behavior; standard mode is still used when the main Agent master switch is turned off.
π Bug Fixes β
More Reliable File Paths in the Web UI β
Relative paths in conversations are now resolved against the current session's working directory, with continued lookup in parent directories and project files; a candidate menu is provided for files with the same name.
- Supports file names with line numbers, line and column numbers, multiple line numbers, and file links in Markdown.
- After the assistant finishes its output, file existence is verified, and only real files are shown as clickable links, avoiding load failures that only surface after clicking.
Multi-Agent Collaboration Restored β
- Consistent model selection: Default
litesub-Agents without a separately configured model follow the model selected in the current session, including custom models. - Channel quote replies: Quote replies received later by busy members keep the message number and the quoted original text, avoiding misaligned replies.
- Team lead permissions: After creating sub-Agents, the lead can still speak in the channel and update the task board normally.
- Hooks plugin compatibility: Fixed tool calls by expert team members and the leader in subsequent turns being incorrectly intercepted.
Session Resume and Terminal Experience β
- Large image results no longer freeze the terminal: Inline images such as screenshots are shown as brief summaries, and very long text no longer leaves the UI unresponsive for extended periods.
- Task list restored correctly: After resuming an interrupted session, the list shows the current progress again; Chinese and other non-English titles no longer have
ingincorrectly appended. - More accurate context usage: Updated usage is shown after compaction and stays consistent across configuration switches and session resumes, while cumulative billing remains unchanged.
- Retrying failed requests: Harness can retry the current failed request without creating duplicate user input, and supports resuming from saved history.
- Native runtime error recovery: Fixed tool exceptions failing to self-correct, misjudged step limits, and inaccurate classification of cancellations and permission denials.
Connections, Search, and Mods β
- More stable WeCom bot connections: Auto-connect from a newly opened command-line process no longer preempts a bot connection that is currently serving; it can reconnect automatically after a manual takeover ends.
- Mod Hook logs: Mods that load normally no longer falsely report "invalid hook configuration format".
- Search sources preserved: Search results whose URLs are hidden for compliance still keep their summaries and source information.
Sandbox and Safe Deletion β
- Batch safe deletion: Fixed approved batch deletions still failing in standalone binaries; approval and rejection results are now saved correctly.
- Session authorization reuse and isolation: Reduced repeated approvals caused by security center restarts; switching or releasing sessions no longer mixes up rules or clears other sessions' restricted rules, and expired protected entries still require re-approval.
- Command completion and cancellation: Results are returned only after file operations and cleanup finish; cancellation timeouts no longer falsely report a trusted exit, and cleanup failures clearly indicate that files may have been modified.
- Sandbox resource compatibility: Fixed validation conflicts between the Desktop local sandbox package and the deletion tool, with clear errors when resources are incomplete; the user approval flow for batch deletion is preserved even when the security center is temporarily unavailable.
- Windows terminal selection: Gateway correctly selects an available PowerShell from PATH, preventing terminal detection from blocking requests.
- macOS tool output: Safe deletion and related command output in the signed IDE no longer include Electron environment variable warnings.