Skip to content

HybridClaw v0.28.1

Latest
Latest

Choose a tag to compare

View all tags
@furukama furukama released this 10 Jul 17:28
v0.28.1
c882d0b

Release Date: July 10, 2026

The LINE self-chat and channel-reliability patch — an explicitly warned unofficial personal LINE integration, restored Anthropic thinking traces, stricter delivery chunking and event deduplication, more resilient onboarding, preserved auxiliary token caps, and leaner CLI self-updates.

Highlights

  • LINE personal-account self-chat — Operators can link a personal LINE account by QR/PIN and use only Keep Memo self-chat for agent turns. Persistent auth, E2EE state, sender/recipient checks, reply-loop prevention, process locking, self-only outbound enforcement, CLI setup/reset commands, and admin-console pairing status keep the channel narrowly scoped. LINE does not officially support personal-account automation; use a dedicated account only after accepting the restriction risk.
  • Anthropic thinking continuity — Direct Anthropic and HybridAI-routed Claude streams preserve structured thinking across tool-call continuations and expose complete reasoning activity to the web chat trace.
  • Safer channel delivery — Discord and Slack delivery paths enforce hard character and line limits while preserving Unicode and fenced code, and duplicate Discord updates, Slack event twins, and WhatsApp LID aliases no longer start duplicate or misidentified turns.

Changed

  • CLI self-update footprint and output — npm self-updates install production dependencies only and suppress non-actionable deprecation, funding, and audit noise while continuing to surface real installation errors.
  • LINE dependency governance — LINEJS and its JSR runtime packages are version-pinned with integrity hashes, the JSR registry allowance is narrowly scoped, and Apache Thrift is overridden outside its vulnerable range.

Fixed

  • Onboarding autostart visibility and retry — Assistant-only first-run sessions remain visible while bootstrap autostart is active or awaiting a reply, and temporary provider failures can retry without consuming a hatching turn.
  • Delivery chunk boundaries — Long unbroken text, Unicode surrogate pairs, line limits, and closing/reopened code fences stay inside configured transport limits.
  • Duplicate channel dispatch — Discord ignores metadata-only message updates, Slack coalesces message and app_mention twins while retaining attachments, and WhatsApp resolves alternate phone JIDs for LID senders.
  • Auxiliary output caps — Explicit task-level maxTokens values remain effective even when model discovery does not report a provider maximum.

Contributors

Core

  • @furukama — LINE channel, onboarding resilience, delivery chunking, channel deduplication, auxiliary token caps, Anthropic thinking continuity, self-update footprint, docs, and release prep
  • @maxnoller — quieter npm self-update output

All Contributors

@furukama @maxnoller

Full Changelog: v0.28.0...v0.28.1

Contributors

furukama and maxnoller
Assets 2
Loading