Pulled from GitHub Releases

Changelog

Every release of netclaw-dev/netclaw. Pick a version from the sidebar — this page rebuilds whenever a new release is published.

v0.11.0

Netclaw 0.11.0

0.11.0 2026-04-11

Netclaw v0.11.0 — Slack thread backfill, approval flow fixes, memory recall improvements, and internal cleanup

Features

  • Added thread history backfill on mid-thread @-mention — when Netclaw is mentioned mid-thread (rather than at the start), it now fetches and hydrates the full prior thread history so the LLM has complete context for the conversation. (#576)

  • Added inbound webhook observability and stats surface — the webhooks ingress layer now exposes per-route call counters, last-call timestamps, and error rates, giving operators a live view of webhook activity without digging through logs. (#550)

Bug Fixes

  • Fixed approval wait handling and Slack prompt resolution after selection — approval flows could hang or leave stale interactive prompts in Slack after a user selected an option. Both issues are resolved. (#587)

  • Fixed missing gap image in Slack thread hydration — DataContent for gap images was dropped during thread reconstruction, causing broken display when Netclaw replayed thread history. (#583)

  • Fixed reminder notifications using a hardcoded audience instead of the deployment posture audience — reminders now correctly derive the notification audience from the active deployment posture rather than always targeting Team. (#570)

  • Fixed memory recall under-returning results — recall overfetch and formation over-production were reduced; the composite-score ranker now applies a score floor so weak matches are excluded rather than surfaced. (#567, #582, #585)

  • Fixed per-channel memory domains collapsing to a shared namespace — memory was written to project:default instead of the per-channel domain, causing cross-channel knowledge bleed. Channels now have properly isolated memory audiences. (#557)

  • Fixed netclaw init wizard issues from 0.10.1 — resolved several init wizard regressions including config write ordering and step skip logic. (#537, #538, #540, #559)

  • Fixed LLM sessions not draining on SIGTERM — active sessions were abandoned rather than drained during shutdown, causing stale active session counts to persist after restart. Sessions are now given a grace period to complete their current turn before the process exits. (#564)

  • Fixed CLI endpoint state being entangled with daemon config — CLI connection state is now tracked separately from the daemon's runtime configuration, preventing config mutations from affecting active CLI sessions. (#556)

  • Hardened skill scanning — duplicate frontmatter names from the same source are now rejected in non-strict mode, and several edge cases in skill directory scanning are closed out. (#552, #555)

Internal

  • Removed dead memory migration cruft, sidecar infrastructure, HardScope plumbing, and the Domain concept in favor of proper audience isolation. These are internal refactors with no user-visible behavior changes. (#586, #588, #589)

Dependencies

  • Bumped Anthropic SDK from 12.11.0 to 12.13.0. (#571)