v0.4.0

Minor Changes

• ce3adcf Thanks @liruifengv! - Add automatic conversation compaction for long-running sessions.

  This release persists per-session compaction snapshots, rebuilds active LLM context from a compact summary plus recent history, automatically compacts when context usage approaches the model limit, and recovers from context overflow errors by compacting and retrying. The UI now also shows a real compaction notice when this happens.

Patch Changes

• 5214b12 Thanks @liruifengv! - Fix workspace file mentions and path validation.

  File mention suggestions now load from the active session workspace instead of an empty IPC placeholder response. Mentioned @path entries are also validated against the real workspace before the UI highlights them as existing files.

• 7274ebc Thanks @liruifengv! - Enhance skill uninstall to support multiple configured skill directories.

  The uninstall feature now uses the resolved directory path instead of skill name, allowing skills to be removed from workspace .amon/skills, user ~/.amon/skills, and any configured extraDirs. Added safety validation to prevent deletion outside allowed skill roots.

• 0c5a413 Thanks @liruifengv! - Refine conversation compaction behavior and configuration.

  Compaction summaries now require a validated XML structure, which makes the compacted handoff more predictable for later turns. Automatic compaction also uses a safer boundary that keeps a recent user-led tail instead of trimming into an incomplete turn. In settings, the redundant compaction.enabled flag is folded into compaction.autoCompact while preserving backward compatibility for existing configs.