File Ops

File Ops

Touching a file on someone's disk is not like answering a question — a wrong guess doesn't just look bad, it overwrites something they cannot get back. Slow down in proportion to how hard the change is to undo.

Procedure

1. Find it before you touch it. Don't guess a path from memory or the user's description. Use find_files(query) for "the report from last week" style requests, or search_file_content(pattern) when you know a string that must be inside the file but not which file it's in. Confirm the exact path with get_file_info before acting on it — a near-miss filename is easy to write to by accident.
2. Read before you write or edit. read_file(file_path) first, every time — even for a file you edited two turns ago. It may have changed since, and edit_file only succeeds against the file's current exact content.
3. Prefer edit_file over write_file for existing files. edit_file replaces one exact string (old_content → new_content); it only changes what you targeted. write_file replaces the entire file — reach for it only when creating a new file or when you genuinely mean to replace everything.
4. When edit_file fails with "content not found," re-read the file. Do not retry with a guessed variation of the string. The failure means your copy of the file is stale or the whitespace doesn't match byte-for-byte — rewriting from memory to force it through is how you silently discard whatever changed since you read it.
5. Confirm before anything destructive. Overwriting a file whose existing content is unrelated to the new content, replacing most of a file in one edit_file call, or writing to a path that already holds something important — stop and ask with request_user_input first. A backup gets created automatically on overwrite/edit, but "recoverable with effort" is not the same as "the user expected this."
6. Report what actually changed. State the path, and for edits, name what changed rather than pasting the whole diff back at the user unless they ask for it.

Guardrails are not bugs

write_file and edit_file refuse system directories, credential-bearing dotfiles such as environment files and SSH key directories, and anything above a 10 MB size limit — these are the platform's sandbox, not something to route around. If a write is denied, that is the answer; report it and ask the user where they'd actually like the file, don't look for another tool that skips the check.

What this skill does not cover

read_file is for text — Python, Markdown, config, plain text. It refuses binary document formats (PDF, DOCX, XLSX, and similar) outright, because reading them raw produces garbage the model then reasons over as if it were real content. Those need index_document + query_documents instead — see the document-brief skill.

Fork this

Pin step 1 to a project's actual file layout (config always lives at X, generated output always goes to Y) and the search step collapses to a direct path, turning this into a fast in-place editor for one codebase.