#Jupyter-live-kernel (no dependencies flagged during setup), but fails to run

1 messages · Page 1 of 1 (latest)

daring edge
#

Debug report uploaded:
Report https://paste.rs/JAgGy
agent.log https://paste.rs/BoRzY
gateway.log https://paste.rs/SqKsi

[subagent-0] ⚠️ Unknown tool 'jupyter-live-kernel' — sending error to model for self-correction (1/3)
┊ 💬 The initial attempt to use the jupyter-live-kernel failed because the tool was not recognized by the execution environment, despite being listed in my available skills. I must switch strategy: since the goal is complex data processing and visualization, I will use a sequence of shell commands via terminal combined with Python scripting logic written to files, leveraging pandas for its robustness, which mimics the stateful nature needed.


No self-correction happens, this is installed bare-metal and I had YOLO enabled too. I'd expect tools with dependencies to say [dependencies] in Red or something indicating the tool won't work during setup or for the agent running on bare-metal with YOLO to grab any dependencies it needs.

daring edge
#

I just ran a brew install jupyter to get the dependencies. I explicitly did a gateway reset, resource ~/.zshrc, hermes chat, re-prompt (with explicit instructions to use jupyter):

dim ginkgo
#

This looks like a real UX mismatch, not just a random failure.

jupyter-live-kernel is not a Hermes tool. In the current repo it is a skill, and its own skill file explicitly says it uses terminal commands and adds no new tools. So when the model tries to call jupyter-live-kernel as if it were a tool, Unknown tool 'jupyter-live-kernel' is expected.

The other important part is that the skill has external prerequisites Hermes is not auto-installing for you just because the skill is present. The documented setup for that skill is more than brew install jupyter. It expects:

  • uv
  • uv tool install jupyterlab
  • a running Jupyter server
  • the hamelnb repo cloned under ~/.agent-skills/hamelnb

So the current issue is basically two layers:

  1. the model is treating a skill like a tool
  2. the external environment the skill expects may still not match the documented setup

So your complaint about setup/dependency visibility is fair. Right now Hermes does not make this clear enough during setup.

Please check these exact prerequisites and paste the output:

which uv
uv tool list
jupyter-lab --version
ls ~/.agent-skills/hamelnb

and also tell us whether the agent is actually loading the skill first, or just being told in plain English to “use jupyter.” In the current implementation, that difference matters. Skills are called with the corresponding / command.

daring edge
#

Solved - brew install jupyter

First run:
mkdir ~/workspace
python3.11 -m venv ~/workspace/.venv

Every run:
source ~/workspace/.venv/bin/activate
hermes gateway restart
hermes gateway chat