v3.53.0 - console 0.34.0: intelligence views, workflows board, control plane, mission autopilot

ruflo 3.53.0

The npm packages (@claude-flow/cli, claude-flow, ruflo) are a version bump plus a catalog entry. The big changes live in the console, ruflo-protector and mods plugins, which reach you through the git marketplace, not npm.

What is new: ruflo-console 0.34.0

• Intelligence views: a memory map (a hash layout, labelled as such, not a similarity map), a learning pipeline and router picture, explain-a-recall with a pattern lifecycle, and memory health.
• Workflows board: phases, agents, model, tokens and time per run, with drill-down, search and logs, replay, compare, export, cost and a budget guard, triage, alerts, a worktree manager, templates, and Anatole per agent.
• Control plane and multi-model bridge: Stop, Message and Redirect through TaskStop and SendMessage, proven for Agent-tool background agents and unverified for a workflow run's own agents. A Conversation pane covers Claude, the hive, federation, Codex, and OpenRouter or any OpenAI-compatible endpoint.
• Mission autopilot: one approved scope envelope, a journal, a kill switch, parking, and governed self-tuning that cannot widen its own envelope. It has not yet run in a live interactive session, and long runs need the session to stay alive.
• Three adversarial audits were run on the above.

How to get it

npm: npx ruflo@latest. For the console, ruflo-protector and mods:

claude plugin marketplace update ruflo
claude plugin update ruflo-console@ruflo

then restart Claude Code.

Known limits

• About 190 ms extra plugin load per start (the engine transpiles the whole module graph). All 216 kit tests pass file by file, but 3 to 8 time out when all 25 files run at once under heavy load.
• npm Audit checks are red on main from newly published advisories in dependencies (proxy-addr, tinypool, and one more transitive). Being fixed separately.
• SendMessage delivery and agent control inside a workflow run are unverified.