v3.53.0

## 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.