Summary

The dangerous command approval prompt (the TUI dialog shown when a potentially dangerous terminal command is detected) has a hardcoded 60-second timeout. When it expires, the command is automatically denied. For users running complex agent workflows — especially with delegation subtasks that trigger approval — 60 seconds is often too short, leading to unnecessary timeouts and denied commands.

Proposal

Read the timeout from config.yaml under approvals.timeout, falling back to 60 seconds when unset:

approvals:
  mode: smart
  timeout: 120   # seconds for dangerous command approval prompt

Affected locations

1. tools/approval.py — prompt_dangerous_approval() takes timeout_seconds=60 as a default. Should read from config via a helper like _get_approval_timeout().
2. hermes_cli/callbacks.py — approval_callback() hardcodes timeout = 60. Should read from CLI_CONFIG.get("approvals", {}).get("timeout", 60).

Why

• Long-running agent tasks (delegation, MCP tool chains) can trigger multiple approval prompts. If the user is AFK or slow to respond, the 60s timeout kills the entire workflow.
• The clarify callback already reads its timeout from config (CLI_CONFIG.get("clarify", {}).get("timeout", 120)), so this is consistent with existing patterns.
• No behavior change for users who don't set the key — default stays at 60s.

Minimal diff

The change is small (~10 lines): add a _get_approval_timeout() helper in approval.py that reads approvals.timeout from config, and replace the hardcoded 60 in both locations.