Reverie 0.2.0

First release since 0.1.0. Reverie is early: see the README for known limits.

Breaking changes

• • Renamed from ec to reverie. The package, the CLI (reverie install, reverie review, and so on), the MCP server (reverie-mcp), the data folder (~/.reverie) and the environment variables (REVERIE_HOME, and so on) all use the new names. The 0.1.0 release used the old ones.
• Sessions are automatic per branch. There is no start or stop command.

Added

• Setup for four agents: Claude Code, Cursor, Codex CLI and OpenCode. reverie install configures each one it finds.
• Turn Reverie on or off per repo: reverie enable, reverie disable, reverie repos.
• One review queue across all branches and repos: reverie review. Accept, reject or skip each candidate.
• Agents can list and record review decisions, but only decisions you state.
• Export and import of your memory: reverie export, reverie import.
• reverie uninstall removes exactly what install added, and keeps your memory unless you pass --delete-data.
• LLM providers: OpenCode Zen (default), Anthropic, OpenAI, Google Gemini, OpenRouter and Ollama.
• reverie retry recovers failed extractions. reverie repair checks the database.

Changed

• The rules agents follow are now one shared policy. It's installed for each agent and sent in the server's instructions.
• Re-running reverie install refreshes outdated instruction blocks.
• Reverie's read tools are allowed in each agent's permission settings, so they don't prompt. Write tools still prompt.
• Claude Code loads Reverie's tools upfront. This needs Claude Code 2.1.121 or later.
• Codex CLI uses skills ($reverie-review, $reverie-status) instead of prompts.

Removed

• Gemini CLI support. It was never verified.
• The ec-start and ec-stop commands.

Fixed

• Claude 5.5-family models reject sampling settings. Reverie no longer sends them.
• Failed extractions are kept and retried, not lost.
• A failed import leaves your existing memory unchanged.