file-operations
Read, write, search, and edit files across the workspace. Enforces the READ-BEFORE-WRITE rule, batch file reads, atomic SEARCH/REPLACE blocks, and file discovery via grep and glob rather than terminal
SKILL.mdprompt-orchestrator/.claude/skills/file-operations/SKILL.md
GitHub<!-- Generated from skills/file-operations.md by tools/generate_integrations.py. Edit the source file, not this one. -->
Skill: File Operations
Purpose
Read, write, search, and edit files in the filesystem with maximum efficiency and minimal token waste.
Tools Required
- readFile / readMultipleFiles
- write / fsWrite
- append / fsAppend
- edit / strReplace / search_replace
- grep / grepSearch
- glob / fileSearch / glob_file_search
- delete / deleteFile
- list_dir / listDirectory
General Principles
- Prefer batch reads (readMultipleFiles) over sequential single-file reads
- Read entire files when practical: partial reads force extra roundtrips
- Search first (grep/glob) before reading when you don't know exact file location
- Never print file contents to user: use edit/write tools instead
- Never generate binary, hashes, or non-textual content
Reading Files
Single File
- Use
readFilewith known absolute path - For large files (>500 lines), read in chunks with offset/limit
- Prefer reading a large meaningful section over many small sequential reads
Multiple Files
# Preferred: batch related files in one call
readMultipleFiles(paths=[...])
File Discovery
- Use
glob/fileSearchwhen you know part of the filename - Use
grep/grepSearchwhen searching for content patterns - Use
listDirectoryfor understanding structure - NEVER use shell
find,grep,catfor file operations: use dedicated tools
Writing Files
Creating New Files
- Use
write/fsWritefor new files or complete rewrites - For files >50 lines, prefer write + follow-up appends
- Always create with complete, immediately runnable content
- Include all imports, dependencies, and types
Appending to Existing Files
- Use
append/fsAppendwhen adding to the end of a file - File must already exist
Editing Existing Files (SEARCH/REPLACE)
- Use
edit/strReplace/search_replacefor targeted edits - CRITICAL:
oldString/SEARCHblock must match EXACTLY, character for character, including whitespace - Include 2-5 lines of surrounding context to ensure uniqueness
- Break large edits into a series of smaller, targeted SEARCH/REPLACE blocks
- Each block should change a focused section: don't edit half a file at once
# GOOD: precise with context
edit(
filePath="src/app.py",
oldString="def old_function():\n return x + 1\n\ndef another():\n pass",
newString="def new_function():\n return x * 2\n\ndef another():\n pass"
)
# BAD: too little context (may match multiple places)
edit(
filePath="src/app.py",
oldString="return x + 1",
newString="return x * 2"
)
Partial Write for Large Files
- For large files where only small sections change, use
// keep existing codemarkers - The unchanged code stays as a comment placeholder
- Only applies when the tooling supports this pattern
Deleting Files
- Use
deleteFile/deletewith explanation - Handles non-existent files gracefully
Searching
Content Search (grep)
- Use
grep/grepSearchfor regex pattern matching across files - Rust regex syntax. Escape special characters:
(,),[,],{,},+,*,?,^,$,|,.,\ - Include patterns to filter file types when possible
- Results capped at 50: refine query if results fill up
File Search (glob)
- Use
glob/fileSearchwhen you know part of the filename - Glob patterns like
**/*.ts,src/**/*.py
Directory Listing
- Use
listDirectory/list_dirwith optional depth parameter - Use for understanding project structure before diving in
Batch Editing Rule
- When making multiple edits to the same file, combine ALL changes into a SINGLE edit call
- This minimizes roundtrips and token overhead
- Only split into multiple calls when edits are in completely unrelated sections
Post-Edit Verification
- After editing, check for linter errors by running lint tools
- If errors introduced, fix them (max 3 fix cycles per file)
- Verify imports are complete and correct