context

To reduce token usage during multi-file edits, apply surgical read discipline using offset and limit parameters, and execute single edits or replace_all operations. This minimizes unnecessary token consumption while preserving accuracy.

Conserving context window limits involves using a grep first, read second methodology. Narrowing reads with offset and limit keeps context small, while parallelizing reads and edits maintains efficiency during complex explorations.

Apply context conservation rules when approaching context window limits or before starting large tasks. Invoking these rules before high token cost operations like multi-file edits or long-running explorations maximizes token efficiency.

Reading entire files wastes tokens by loading unnecessary data into the context window. Using surgical reads with offset and limit, and enforcing one read per edit, reduces redundancy and prevents context overflow.

You can parallelize reads and edits when multiple files are involved to maximize productive work. Combining parallel operations with read and edit discipline maintains efficiency and reduces overall token usage.

Grep first, read second discipline works for high token cost tasks by filtering content before reading. This approach minimizes unnecessary reads, keeping context small and optimizing token usage during complex workflows.