Skip to content

CHANGELOG.md

Latest commit

 

History

History
31 lines (28 loc) · 1.7 KB

CHANGELOG.md

File metadata and controls

31 lines (28 loc) · 1.7 KB
Raw

Changelog

All notable changes to this plugin will be documented here.

Unreleased

  • Default memory file moved to a user-scoped path (~/.cursor/projects/<workspace-slug>/AGENTS.local.md) so the plugin is safe to enable inside team repositories. The repository's AGENTS.md / CLAUDE.md / GEMINI.md is no longer written by default.
  • Cadence and incremental-index state moved out of .cursor/hooks/state/ into ~/.cursor/projects/<workspace-slug>/continual-learning/. Legacy state files are migrated to the new location and removed from the repository on first run (a backup copy is kept under the user-scoped state directory).
  • New opt-in CONTINUAL_LEARNING_WORKSPACE_FILE env var lets users point the plugin at a workspace-scoped memory file when they explicitly want team-shareable workspace facts written somewhere inside the repo.
  • The hook now runs git check-ignore against any configured workspace memory file and refuses to write to it when the file is inside a git repo and not gitignored, unless CONTINUAL_LEARNING_ALLOW_SHARED=1.
  • Added CONTINUAL_LEARNING_USER_FILE, CONTINUAL_LEARNING_STATE_DIR, and CONTINUAL_LEARNING_ALLOW_SHARED env vars. All existing cadence env vars are unchanged.
  • The hook now embeds the already-resolved target paths and any safety-check block reason directly into the followup_message, so the agents-memory-updater subagent does not need to recompute them.
  • Section routing made explicit: ## Learned User Preferences always lives in the user file; ## Learned Workspace Facts lives in the workspace file when configured and not blocked; a ## Learned Workspace Facts (local) fallback heading is used in the user file otherwise.