Lab: Trace one turn

#

Goal

#

Run a chat that triggers a tool call, then map the runtime path onto concrete symbols in the repo.

Prereqs

#

• Working hermes chat (Install and first chat)
• Optional but recommended: a local clone of hermes-agent for reading source
• Tools that can run commands/files enabled (default terminal/file toolsets)

Steps

#

1. Start Hermes and ask something that needs a tool, for example:
   • “Show my current working directory using your tools.”
   • or “List files in the current directory.”
2. Watch the UI: you should see a tool invocation (name + args) before the final answer.
3. In the repo (read-only), open in order:
   1. agent/conversation_loop.py — find the loop that calls the provider with tools=
   2. model_tools.py — find handle_function_call
   3. The tool’s module under tools/ (for example terminal/file tools) — find registry.register and the handler
4. On paper, write the message roles you expect after one tool round:
   • user → assistant (with tool_calls) → tool → assistant (final)

Expected outcome

#

• At least one tool call visible in the session
• A short notes file (mental or markdown) listing: loop file → dispatch function → handler module
• Role sequence matches strict alternation rules

Verify

#

1. Tool name you saw in the UI appears in toolsets.py or an enabled toolset
2. You can point to handle_function_call as the bridge from model → handler
3. You did not need to edit any core file

Common pitfalls

#

• Asking a pure trivia question that never calls tools — nudge the prompt toward filesystem/shell
• Reading only run_agent.py and missing that the loop lives in conversation_loop.py
• Confusing “tool appeared in UI” with “tool is in _HERMES_CORE_TOOLS” — check the active platform toolset

Stretch

#

• Repeat with gateway messaging once CLI works (hermes gateway setup + Telegram/Discord). Compare entry points using Data flows.
• Official Telegram path: website/docs/user-guide/messaging/telegram.md

Next lab

#

Write a skill