feat(security): Add security hardening for MCP, URL scheme, and per-terminal controls

[eyelock]

Summary

• Add comprehensive security hardening based on security audit findings
• Add per-terminal security settings that override global preferences
• Add Security documentation page to Help

Security Features Added

Input Validation

• String length limits (1,000 chars general, 50,000 chars LLM context, 4,096 chars paths)
• Path traversal prevention (blocks ../ sequences)
• UUID format validation

URL Scheme Protection

• User confirmation dialog for external LLM context modifications via termq:// URL scheme
• Prevents malicious scripts from injecting commands into AI workflows

Clipboard Access Control

• OSC 52 clipboard access preference (global and per-terminal)
• Control whether terminal programs can write to system clipboard

File Coordination

• NSFileCoordinator for safe concurrent board.json access
• Prevents data corruption when app, CLI, and MCP server access simultaneously

Per-Terminal Security Settings

• Allow OSC 52 Clipboard (overrides global)
• Confirm External Modifications (overrides global)
• Clear "Disabled globally" indicators when global setting takes precedence

Documentation

• New Security page in Help (Advanced Topics section)
• Covers all security features, best practices, and settings reference
• Links to related docs (configuration, environment variables, MCP server)

Includes

• Localized security strings for all 40 supported languages
• 25 new unit tests for security features
• Security audit report in .claude/security-audit-2026-01-16.md

Test plan

• make check passes (build, lint, 516 tests, localization validation)
• Verify OSC 52 clipboard toggle works (global and per-terminal)
• Verify external modification confirmation dialog appears for termq://update with LLM fields
• Verify input validation rejects oversized strings and path traversal attempts
• Verify Security help page renders correctly

🤖 Generated with Claude Code