context-budget

To manage context window bloat, track token consumption and enforce preloading constraints for agent sessions, ensuring total skill content lines adhere to per-agent tier limits to prevent context degradation and maintain agent performance.

To enforce skill loading limits, track preloaded skills per agent, sum the lines of skill content, and apply tier-specific maximums, such as limiting a 'Sonnet' tier agent to 5 preloaded skills and 400 total lines.

Signals of context overload in prompt engineering include ignored instructions and repetitive queries, indicating the agent's context window is bloated and requires token consumption monitoring and skill preloading optimization.

Optimize token consumption by checking the preloaded skill line count for the 'Opus' agent, monitoring token usage, and ensuring the total lines of code from preloaded skills do not exceed configured per-agent tier limits.

Yes, token consumption monitoring tracks token usage per agent to prevent exceeding limits, enforcing skill preloading constraints and alerting on potential context window bloat to ensure efficient agent performance.

Your agent ignores instructions due to context overload from exceeding preloaded skill limits, requiring context bloat detection and token consumption tracking to enforce per-agent tier constraints.