Problem

When Hermes requests approval for a dangerous command, the prompt currently shows only a static pattern label as the reason (e.g. "dangerous command", "service restart", "filesystem deletion").

Example of what the Telegram gateway shows:

⚠️ Command Approval Required

`hermes update`

Reason: dangerous command

The user has to infer from context WHY the agent wants to run this command — which is often unclear, especially when multiple things are happening in a session.

Desired Behavior (inspired by Codex)

Before presenting the approval prompt, the agent should include a brief contextual explanation of why the command is needed in the current task context. For example:

⚠️ Command Approval Required

I just installed a Hermes update that requires a gateway restart to take effect.

`systemctl --user restart hermes-gateway`

Reason: service restart

This makes approval decisions much easier — the user knows why the command is necessary before choosing Allow/Deny.

Suggested Implementation

• Allow the agent / LLM to prepend a contextual explanation to the approval prompt
• This could be:
  • An additional text field in prompt_dangerous_approval() (e.g. contextual_reason)
  • Or the agent writes the explanation as a normal message before the tool call, and the approval system preserves it
• The static description from detect_dangerous_command() would remain as supplementary info

Related

• [Feature]: Telegram approval messages should retain command context after approval/denial #23965 (Telegram approval messages should retain command context after approval/denial) — related but different: that issue is about the post-approval audit trail, this one is about the pre-approval rationale.