ruflo-ruvllm

Reviewed at commit b9f3245

Caution: Its skills pre-approve Bash and tell Claude to install the unpinned npm packages @ruvector/ruvllm-wasm and @ruvector/ruvllm@^4.1.0, so the installs run without asking.

RuVLLM local inference with chat formatting (Claude/GPT/Gemini/Ollama/Cohere), model configuration, MicroLoRA fine-tuning, and SONA real-time adaptation. As a…

Made by
ruvnet
What we checked

Watches calls to the plugin's own write tools (HNSW pattern add, MicroLoRA and SONA adapt) and refuses any call that holds something shaped like a key, token or password. Adds a /ruvllm-mod command (status, scan, tools) and writes a small counter file in your project. It makes no network calls and runs no programs itself.

Review details

Static review of the source at the pinned commit; not run. Delta review of the changes since 87d8687; the rest carries over from the review at that commit. The hook code is unchanged; the change is a new skill, ruvllm-runtime, plus manifest and doc updates (version 0.4.0, repository field, two keywords). The skill pre-approves Bash, Read and ruvllm_status, explains the three RuVLLM npm packages, and tells Claude to run npm ls and node -e require('@ruvector/ruvllm') to check the installed version, to npm i @ruvector/ruvllm-wasm (unpinned) and npm i @ruvector/ruvllm@^4.1.0 (range) when needed, and offers npx -y @ruvector/[email protected] route/embed (pinned) and gpu loop/run commands with a user-given config path. The module registers session.start (reads the project root, registers /ruvllm-mod, writes the status file), tool.call and command.run. tool.call only looks at calls to ruvllm_* tools; for the three writer tools (ruvllm_hnsw_add, ruvllm_microlora_adapt, ruvllm_sona_adapt) it flattens the input, runs regex and entropy secret checks and returns a deny, otherwise it passes the call unchanged (tighten-only; guard option default on). Findings are names only; input too large to screen is refused. Runtime calls: $.session.root, $.command.register, $.clock.now, $.fs.write (only .claude-flow/ruvllm-mod/status.json in the project, at most every 5 s, failures swallowed), $.tool.list. No network, process or environment access in the hooks. The chat-format and llm-config skills also pre-approve Bash and tell Claude to run npm i -g @ruvector/ruvllm-wasm when the WASM add-on is missing; the llm-specialist agent runs npx @claude-flow/cli@latest memory and hooks commands without pre-approving tools. The ruflo MCP server, @ruvector/ruvllm, @ruvector/ruvllm-wasm and @ruvector/ruvllm-gpu-loop were not reviewed. scripts/smoke.sh is a test script.

What it uses
  • Reads your prompts: no
  • Sees tool calls: yes
  • Blocks or changes tool calls: yes
  • Reads or writes files: yes
  • Runs programs on your machine: no
  • Calls the model (uses tokens): no
  • Sends data over the network: no
  • Stores data: yes
  • Shows something on screen: yes
Hooked events

RuVLLM local inference with chat formatting (Claude/GPT/Gemini/Ollama/Cohere), model configuration, MicroLoRA fine-tuning, and SONA real-time adaptation. As a mod (ADR-445): a guard that keeps secrets out of adapters and the pattern router, /ruvllm-mod, and a status file the console shows

Made this mod? Add a badge to your README

modsforClaude badge for ruflo-ruvllm

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/ruflo-ruvllm.svg)](https://modsforclaude.com/mods/ruflo-ruvllm/)

← All mods