It uses sub-sub-agents in claude code

[anatoly-pashin-axon21]

Describe the bug

Claude code currently allows sub-sub-agents. This is what I see in the sub-agent logs:

Now I'll delegate the actual write operations (remember/link) to a sub-agent, per the mnemon skill's guardrail against running writes directly. I've already done the dedup recall myself and confirmed the target ID and store.

So it effectively prepares data, then launches a new agent, and the sub-sub-agents reads the skill again and learns how to use mnemon tool

I believe this consumes too many tokens and is slow and should be done using 1 agent only. Maybe the instructions should be split to two files or something else.

This complication doesn't block anything, the process works fine. But I believe it should be optimized.

Environment

• OS: Ubuntu 24
• mnemon version: 0.1.11
• LLM CLI: Claude Code 2.1.198

[Grivn]

Thanks for the report. Fixed in #81 and merged to master.

Claude memory writes now use only one delegated memory sub-agent: the sub-agent runs mnemon remember/link directly instead of spawning another sub-agent, which should reduce the extra token use and latency you saw.