Skip to content

Latest commit

 

History

History
57 lines (33 loc) · 2.04 KB

File metadata and controls

57 lines (33 loc) · 2.04 KB

Changelog

[1.1.0] - 2026-01-28

Added

  • PreToolUse hook for mid-workflow context injection - New lightweight hook that checks for Letta agent updates before each tool use. Addresses "workflow drift" in long workflows by injecting new messages and memory block diffs mid-stream. Silent no-op if nothing changed.

  • Letta Code GitHub Action - @letta-code can now respond to issues and PRs in this repository.

  • LETTA_BASE_URL support - Self-hosted Letta servers can now be configured via environment variable.

  • Windows compatibility - Fixed npx spawn ENOENT error on Windows.

  • Linux tmpfs workaround - Documented workaround for EXDEV error when /tmp is on a different filesystem.

Changed

  • Session start sync - CLAUDE.md now syncs at session start for fresh agent/conversation IDs.

  • Default model - Changed default agent model to GLM 4.7 (free tier on Letta Cloud).

  • Automatic model detection - Plugin now queries available models and auto-selects if configured model is unavailable.

Fixed

  • Plugin install syntax - Updated README with correct marketplace install commands.

  • Conversation message ordering - Fixed message fetch to correctly show newest messages first.

  • Conversation URL - Links now point to agent view with conversation query param.

Security

  • Sanitized default agent - Removed user-specific data from bundled Subconscious.af file.

[1.0.0] - 2026-01-16

Initial release.

Features

  • Bidirectional sync between Claude Code and Letta agents
  • Memory blocks sync to .claude/CLAUDE.md
  • Session transcripts sent to Letta agent asynchronously
  • Conversation isolation per Claude Code session
  • Auto-import default Subconscious agent if no agent configured
  • Memory block diffs shown on changes
  • New messages from Letta agent injected into context

Hooks

  • SessionStart - Notify agent of new session
  • UserPromptSubmit - Sync memory before each prompt
  • Stop - Send transcript after each response