Increase default context size from 20k to 100k tokens

[alvinunreal]

• Update DefaultConfig() in config/config.go: 20000 -> 100000
• Update config.example.yaml: 20000 -> 100000
• Update README.md example: 16500/20000 -> 82500/100000

Fixes #63 - 20k context size was too small and caused frequent squashing that disrupted conversation flow. 100k provides 5x more capacity while staying within modern AI model limits.

🤖 Generated with Claude Code

[claude]

Code Review

Overall Assessment

This PR addresses issue #63 by increasing the default context size from 20k to 100k tokens. The changes are minimal, consistent, and well-justified.

✅ Code Quality & Best Practices

• Clean and consistent: All 3 files (config.go, config.example.yaml, README.md) are updated consistently
• Follows existing patterns: The change maintains the existing configuration structure
• Good documentation: README.md example is updated to reflect the new values appropriately

✅ No Bugs or Issues Detected

• The change is straightforward numeric value updates
• No logic changes that could introduce bugs
• Configuration loading and validation remain unchanged
• Existing functionality preserved

✅ Performance Considerations

• Memory impact: 5x increase in context size will use ~5x more memory per conversation
• Squashing threshold: The 80% squashing threshold (config/squash.go:21) scales appropriately to 80k tokens
• Processing overhead: Token estimation and context management will handle larger contexts but with proportional computational cost
• Reasonable limits: 100k tokens is within modern AI model context windows

✅ Security Concerns

• No security implications identified
• Configuration value changes dont introduce attack vectors
• No credential or sensitive data exposure risks

✅ Test Coverage

• Existing tests still valid: The test in process_message_test.go:64 uses a small value (100 tokens) specifically to trigger squashing behavior, which remains appropriate
• No new tests needed: This is a configuration default change that doesnt require additional test coverage
• Existing logic tested: Context squashing logic and thresholds are already covered by tests

💡 Considerations

1. Resource usage: Users should be aware that larger context sizes will consume more memory and potentially slower processing
2. Backwards compatibility: Users with existing configs will retain their current settings (good)
3. Model compatibility: 100k is reasonable for modern models but consider documenting recommended ranges

Recommendation

✅ APPROVE - This is a well-implemented change that addresses the user experience issue of frequent conversation squashing while staying within reasonable resource limits.

The change is minimal, safe, and addresses a legitimate usability concern without introducing any risks.