The /context command displays percentage based on a 200K token baseline, but our model supports 1M tokens input.
Ignore the percentage - focus on the actual token count.
To calculate real usage: actual_tokens ÷ 1,000,000 × 100 = real_percentage
Example: If /context shows "140k tokens (90%)", that's actually only 14% of 1M capacity.
Run /context to monitor context usage under these conditions:
- After completing a feature or module
- After fixing a bug
- After extensive code review sessions
- Before switching to an unrelated task
Execute /compact when actual token count exceeds 900,000 (90% of 1M capacity).
Do NOT rely on the displayed percentage - it is misleading for our setup.
When running /compact, always include instructions to preserve:
- Project architecture decisions and rationale
- Current task objectives and progress
- Key code patterns and conventions established
- Outstanding TODOs and known issues
- Recent debugging context if still relevant
Example: /compact preserve architecture decisions, current task progress, and established coding patterns