dangerousPatterns config field is not wired

[laynepenney]

Problem

ResolvedConfig has a dangerous_patterns field (from .codi.json dangerousPatterns) but there is no corresponding field in AgentConfig and no logic to check commands against these patterns before execution.

Expected Behavior

When a user configures dangerousPatterns (e.g., ["rm -rf", "DROP TABLE"]), the agent should:

1. Check tool call arguments against these patterns before execution
2. Require explicit user confirmation if a dangerous pattern is matched, even if the tool is in the auto-approve list

Proposed Approach

1. Add dangerous_patterns: Vec<String> to AgentConfig
2. In the tool execution path (before calling execute_tool), check if any argument matches a dangerous pattern
3. If matched, bypass auto-approve and prompt for user confirmation
4. Wire the field from ResolvedConfig in build_agent_config()

Related

PR #261 — Wire config options, context compaction, and auto-index on startup