✨ New Features

Built-in skills selector — A new plugin accessible from the chat input + menu lets you browse and activate available skills directly in the current context or project, without leaving the conversation flow.

⚡ Improvements

Skills can be active from the start — Default configuration now allows skills to be available immediately, saving setup time and reducing token overhead before productive work begins.
FAISS index integrity verification — Memory saves now write an index.faiss.sha256 sidecar file so FAISS index integrity can be verified on load.
Hardened memory_load filter evaluation — Filter execution is restricted with an allowlist, a length cap, and tighter simple_eval boundaries to reduce risk from unsafe expressions.
More accurate memory consolidation scoring — Consolidation now uses real relevance scores and keeps the best match per memory ID, improving deduplication quality.
Safer utility-model context size — Input history is truncated before being sent to the utility model, preventing avoidable context overflows during memory operations.

Empty tool_args no longer break tool validation — validate_tool_request now checks for key presence instead of treating an empty dict as missing, fixing crashes for valid no-argument tool calls such as scheduler listings and health checks.