context_management

You prevent AI agent context overflow by monitoring token usage and triggering compression or task delegation. This skill tracks context windows and uses subagents or forks to maintain high-quality reasoning within constrained memory limits.

You should choose between a subagent and a fork based on a delegation decision matrix evaluating your context needs. Use a subagent for isolated code reviews, and fork the current session when you need to maintain complex state across parallel tasks.

Performance degradation in AI agents is caused by bloated conversation histories exhausting token capacity. As context windows fill, reasoning quality drops, making token usage analysis and output size control necessary to maintain efficient execution.

You monitor token usage by analyzing current context windows to estimate available capacity and identify when to trigger compression. This allows you to track consumption and proactively manage memory constraints before overflow occurs.

Yes, you can maintain complex state in a large codebase without hitting token limits by forking the current session or spawning background processes. This orchestration isolates tasks and prevents bloated histories from degrading agent performance.