Do you all feel that ruflo is a valuable upgrade to stock claude

[jbraseth]

I was using version 2 I believe pretty heavily in January, but since then I've found tremendous value just using claude.

[jbraseth]

Maybe it's obvious to you! Could you help walk me through where you see the most value out of ruflo, though?

For me, I loved having the agent personas right there—whenever I ran a adversarial tournament review, claude picked the appropriate ones and it blew my mind the first time I saw it in action.

[BrokedTV]

following as i just discovered ruflo

[jbraseth]

What are people's favorite aspects of the tool?

[roman-rr]

Short answer: stock Claude Code is better for almost everything.

We audited ruflo v3.5.51 end-to-end. Here's what ruflo adds vs what stock Claude already does better:

Capability	Ruflo	Stock Claude Code
Multi-agent orchestration	agent_spawn writes JSON, never executes	Native Agent tool spawns real subprocesses
Task management	task_assign fires event nobody listens to	Built-in task tracking that actually works
Code search	Terminal execute (redundant)	Grep, Glob, Read — purpose-built, faster
File editing	Terminal execute (redundant)	Edit tool with diff support
Memory/knowledge	HNSW vector search — genuinely useful	MEMORY.md files (simpler but no semantic search)

The only thing ruflo adds that stock Claude doesn't have is the HNSW semantic memory (~10 working tools out of 300+). Everything else is either a stub or redundant.

The cost of keeping ruflo:

• 300+ tool definitions bloating context (~5,000-10,000 tokens/session)
• Intelligence hooks injecting noise every message (~300 tokens/msg)
• 2 Node.js processes running permanently

You can get the same semantic memory capability with a standalone Qdrant MCP server or ~200 lines of custom code with hnswlib-node.

Full audit: Independent Audit