file-operations

Use SEARCH/REPLACE edits where the oldString matches the file exactly, character for character, including whitespace. Include 2-5 lines of surrounding context so the match is unique, and break large edits into smaller focused blocks.

Use glob or fileSearch when you know part of the filename, and grep or grepSearch for content patterns with regex. Avoid shell commands like find, grep, or cat—dedicated tools are faster and capped at 50 results.

Edits fail when the oldString does not match the file exactly, including indentation and line breaks. Re-read the file first, copy the exact text with surrounding context, and retry with a unique match.

Batch related files in a single readMultipleFiles call to minimize roundtrips and token overhead. For files over 500 lines, read in chunks with offset and limit, preferring large meaningful sections over many small reads.

The keep-existing-code marker pattern only works when the tooling explicitly supports it, as with Lovable. Otherwise, use write for full rewrites or append for additions, and split files over 50 lines into write plus follow-up appends.