Use case · long refactor
Large refactors can build substantial context. Check session usage alongside the transcript so you can decide whether to continue, split the task, or review the current diff.
Token counts are signals about model usage, not a quality score. Beeplane can show supported input, output, and cache counts; provider billing and model limits remain provider-specific.
Name the modules in scope, compatibility requirements, and tests. Ask the agent to stop and report before broadening the change.
Review available input, output, and cache metrics for the run. Compare the numbers with the task progress rather than treating a large total as failure.
Look for repeated context, new assumptions, or changes beyond the target modules. Send a narrowing instruction if the task has expanded.
Inspect related changes in groups and run focused tests. If context is unwieldy, summarize the current state and start a smaller follow-up task.
A refactor with clear modules in scope and baseline tests you can compare.
Using token counts alone as a quality, cost, or completion score.
No. It indicates usage, not correctness or quality. Review the scope, transcript, and diff.
No exact invoice is promised. Built-in estimates can differ from current provider billing.
Not necessarily. Split when the scope, review burden, or context makes the work harder to reason about.
Free to download. Works with Claude Code, Codex, OpenCode, and other agents on macOS or Linux.