[FEATURE]: implement smart pruning

[processtrader]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Currently prune is a basic remove after X amount of tokens have been reached, with a very basic filter of the latest operations. This is not enough when interleave thinking models are debugging complex/deep problems on their big context windows. Removing outputs/results from important steps is making them feel dumb or force them to hallucinate. By implementing a summary and smarter mechanism of that, keeping the feature of reducing unnecessary tokens, we can improve the interaction with the LLM

[github-actions]

This issue might be a duplicate of or closely related to existing issues. Please check:

• [FEATURE] Sliding window context management for long-running sessions #4659: [FEATURE] Sliding window context management for long-running sessions - proposes a more sophisticated context management system with inception messages and chess-clock time-based pruning
• Feature: generic /compact command, auto-compaction, and fork-aware conversations #4317: Feature: generic /compact command, auto-compaction, and fork-aware conversations - discusses provider-agnostic compaction and auto-compaction support
• Soft compaction / AI global workspace metabolism #3032: Soft compaction / AI global workspace metabolism - suggests a softer approach to pruning that preserves context better by allowing the model to intelligently select what to remove
• Excluding reasoning context to reduce token usage #4895: Excluding reasoning context to reduce token usage - asks about excluding AI reasoning content from prompts and session compaction
• Research subagent exceeded the context window and silently failed #5970: Research subagent exceeded the context window and silently failed - documents issues when context windows are exceeded
• DeepSeek V3.2 : reasoning_content not cleared from message history on new turns, causing excess token usage and violating API spec #5577: DeepSeek V3.2 : reasoning_content not cleared from message history on new turns - specifically addresses token waste with interleaved thinking models

These issues discuss various approaches to improving context management, pruning strategies, and handling of thinking/reasoning content in long-running sessions. Feel free to ignore if your specific use case differs from these.

[thdxr]

To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.