Add autonomous agent with SSH execution and Claude token savings (93% reduction)

[lambertmt]

Summary

This PR adds an autonomous agent capability that executes tools internally, keeping raw data away from Claude's context window. Real-world testing shows 40-95% Claude token savings on analysis tasks.

Key Features

• Autonomous agent loop - local LLM decides what tools to call, executes them, analyzes results
• Built-in SSH execution - agent can run commands on configured hosts
• GPG-encrypted credentials - secure storage for SSH passwords
• Strict output formatting - clean JSON tool calls, plain text answers

Real-World Results

Task	Claude Direct	Claude w/ Agent	Local LLM (free)	Savings
Debugging workflow (7 calls)	~56,000	~4,100	~35,000	93%
Security audit	~11,800	~800	~11,000	93%
Docker logs analysis	~10,500	~500	~10,000	95%
System health check	~5,500	~1,500	~4,000	73%
Log analysis (journalctl)	~4,000	~800	~3,200	80%
Code gen (w/ exploration)	~2,700	~1,700	~1,000	37%
Disk analysis	~1,500	~500	~1,000	65%
Code gen (small input)	~1,550	~1,600	~1,500	0%
Simple query	~500	~300	~200	40%

Note: The 0% case matters - when raw data is small, there's no benefit. This shines on data-heavy tasks.

How It Works

Claude sends task → Agent (local LLM) executes SSH internally → 
Agent analyzes 40K chars locally → Returns 800-token summary to Claude

Claude never sees the raw output. Tokens shift from paid (Claude) to free (local LLM).

Test Plan

• Health check tool working
• SSH execution with GPG-encrypted credentials
• Agent loop with auto_execute=true
• Multi-iteration tool calls
• Token measurements validated

🤖 Generated with Claude Code