v0.6.2

March 17, 2026

Netclaw 0.6.2

Netclaw v0.6.2 — Reminder history, onboarding chat, and turn recovery hardening

Reminders

Added per-reminder execution history: every reminder execution now appends a record (fired time, success/failure, duration, session ID, error message) to ~/.netclaw/reminders/{id}.history.jsonl. History is capped at 500 records (configurable) with atomic overflow trimming. Accessible via netclaw reminder history <id> [--last N], GET /api/reminders/{id}/history, and the get_reminder_history agent tool (scheduling grant required). History file is deleted when a reminder is cancelled. (#259)

Init Wizard

Added an interactive onboarding chat at the end of the init wizard — after health checks pass, the agent introduces itself and asks what the operator wants to use it for, updating SOUL.md with what it learns. This replaces the static "primary use" text field with a richer conversational approach. (#250)
Fixed the chat TUI getting stuck on "Generating..." after an idle SignalR reconnect — IsGenerating is now reset on disconnect. (#250)

Providers

Fixed OpenAI-compatible provider fallback path using /api/v1/ instead of the standard /v1/ — llama.cpp, vLLM, and other servers that follow the OpenAI path convention without the /api/ prefix now work correctly. (#253)

Skills Platform

System skills are now sourced directly from the feed directory (feeds/skills/.system/files/) at build time, eliminating the dual-copy maintenance problem where the embedded copy and the feed copy had silently diverged. The feed directory is the single source of truth for both on-disk seeding and published feed artifacts. (#258)

Stability

Fixed agent turn recovery for empty LLM responses, forced no-tools turns, and hang scenarios — the session actor now reliably recovers and continues rather than stalling or silently dropping turns in these edge cases. (#260, #269, #271)