AI Agent News/stories/opencode-v2-2-5-adds-percentage-compaction-buffer-and-worktree-shortcuts.md

Frameworks

opencode v2.2.5 adds percentage compaction buffer and worktree shortcuts

opencode v2.2.5 adds a percentage-based compaction buffer and keyboard shortcuts for choosing worktrees and toggling shell mode. The release also fixes MCP server disconnects on large stdio messages and reasoning errors when switching models.

Source: github.com/anomalyco/opencode/releases/tag/v2.2.5

The opencode project has released v2.2.5 with a compaction buffer that can be set as a percentage of each model's context window. Setting the buffer to 20% in opencode.json keeps that share free, and token counts still work. Calls to a tool named bash now run as a shell instead of failing, using the same permissions, though a plugin tool with that name still takes precedence.

Desktop users get shortcuts to choose a worktree and its base branch, plus a key to toggle shell mode in the composer. Holding Cmd or Ctrl in the quote popup sends one quote and its comment without touching the rest of the composer. Error messages for missing read folders and for edit calls that cannot find a match now give more useful detail.

Several fixes cover MCP servers and model handling. Stdio messages larger than 10 MiB no longer disconnect MCP servers, and idle locations no longer restart them right after shutdown. Reasoning is no longer converted to plain text when switching models or resuming an interrupted response, which had caused request errors on some providers.

Why it matters if you build agents

Agent builders running opencode can size context reserves with one percentage instead of tuning token counts for each model. Anything that depends on a tool named bash should be rechecked, since those calls now execute rather than fail. Sessions that send large MCP messages or switch models mid-task should be less likely to break.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe