ruflo-agent
Reviewed
at commit 13af5f1
New version, not yet reviewed
Agent runtimes for ruflo — local WASM-sandboxed agents (rvagent: 10 wasm_agent_*/wasm_gallery_* MCP tools, built on @ruvector/rvagent-wasm +…
0 upvotes
Log in to upvote or save- Made by
- ruvnet
- Pinned commit
-
13af5f1208960c98832d6d9096a8d7ec45bcdc3e - What we checked
-
Refuses calls to the Ruflo agent tools (local WASM agents and Anthropic cloud Managed Agents) when their text holds a key, token or password, and can cap the budget of cloud workflow creation. It adds a /agent-mod command and writes a small status file in your project. Its skills drive cloud sessions that bill your Anthropic account.
Review details
Static review of the source at the pinned commit; not run. Delta review of the changes since 1c0d63c; the rest carries over from the review at that commit. The hooks module registers session.start (reads the project root, registers the /agent-mod command, writes .claude-flow/agent-mod/status.json), tool.call (inspects only wasm_agent_prompt, wasm_agent_tool, wasm_agent_create, wasm_gallery_create and the managed_agent_* tools; it can only deny, via a secret-pattern screen over all strings in the input and, for managed_agent_create with dynamicWorkflows, checks of the managedWorkflows switch (off by default), budget cap (default 2500 cents), limited networking and predefined agents) and command.run for /agent-mod (status, scan, guarded, capabilities; capabilities calls $.tool.list only). No network, process, model or http calls in the hooks; findings are names only and never echo the matched value. Input scanning is bounded and fails closed when too large. The tools the plugin guards and its skills use come from the separate ruflo-core MCP server (not reviewed). Skills and agents tell Claude to use those tools; managed-agent and managed-workflow create Anthropic Claude Managed Agent sessions in Anthropic's cloud (billed tokens and container time, need ANTHROPIC_API_KEY in the environment, optional user-chosen MCP server URLs and allowed hosts). Four skills (managed-agent, nested-subagents, wasm-agent, wasm-gallery) pre-approve Bash, but none of them tells Claude to run an unpinned package; two agents (nested-coordinator, wasm-specialist) show 'npx @claude-flow/cli@latest memory ...' commands for pattern memory, which are not pre-approved. The two leaf agent templates (nested-leaf, nested-queen-leaf) set 'effort: low' and 'omitClaudeMd: true'. Changed in this version: only plugin.json gained a repository URL field (github.com/ruvnet/ruflo); no code, skill or prompt changed. npm packages and the ruflo-core server were not reviewed.
- 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
Agent runtimes for ruflo — local WASM-sandboxed agents (rvagent: 10 wasm_agent_*/wasm_gallery_* MCP tools, built on @ruvector/rvagent-wasm + @ruvector/ruvllm-wasm per ADR-070) plus Anthropic Claude Managed Agents as a cloud backend (managed_agent_* MCP tools per ADR-115). One interface, local-vs-cloud runtimes.
Made this mod? Add a badge to your README
Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.
[](https://modsforclaude.com/mods/ruflo-agent/)