file-manager

name: file-manager description: >- Manage, organize, sort, rename, dedupe, or archive files on disk. Use when reshaping a directory tree, batch-renaming, deduplicating, sorting media by date or location, mounting or extracting disk images, or cleaning up a folder. Skip for git operations and code edits.

file-manager

Filesystem management and reorganization with the tools on this machine.

Scale safety (read first)

These tools will happily walk a TB-scale tree or millions of files and appear to hang for many minutes. Four rules:

1. Probe before recursing. Run a depth-1 walk first to learn the shape of the tree:

   df -h <dir>                                    # filesystem capacity
   fd --max-depth 1 -t f . <dir> | wc -l          # top-level fanout
   dust -d 1 -n 10 <dir>                          # byte distribution
   

   Decide the strategy from these numbers. Never run a recursive command on a directory whose size you have not confirmed.

2. Cap every recursive command with the tool's depth, count, or size flag. Defaults by phase:

   • probe = --max-depth 1
   • inspect = --max-depth 2, -d 2, --level=2
   • work = depth 3+ only after a probe shows it is safe

   Never run dust ., fclones group ., eza -lT ., or czkawka_cli image -d . bare on an unknown tree.

3. Bound anything still unbounded with the Bash tool's timeout (default 120000 ms, max 600000 ms) or run_in_background: true for jobs that may exceed 10 min. If a bounded run trips its limit, stop and re-plan; do not bump the limit reflexively. See large-fs.md for per-job sizing and the non-Claude-Code timeout(1) fallback.

4. NUL-delimited pipes at scale. Million-file trees contain filenames with spaces and newlines. Use fd -0 ... | xargs -0 ... (or fd ... -X cmd, which delimits internally). Newline-delimited pipelines are a correctness bug at this scale.

Per-tool flag tables and segmentation patterns: large-fs.md.

Naming conventions (read second)

Output names and layouts must follow these rules. They override any tool's built-in defaults; post-process with rnr regex (or fd ... -x mv) when a tool's pattern doesn't fit.

1. Charset: [a-z0-9_-]. Lowercase. - between words, _ between logical fields (date/time, subject/version).
2. Periods only for semantic compound extensions -- ones a tool dispatches on (.tar.gz, .d.ts, .schema.json). Version, draft, and locale markers belong inside the stem.
3. Layout is domain-driven, not type-driven. Group by what the files are about, not their format.
4. Dates: yyyy-MM-dd (2026-04-28); add _HH-mm-ss only when seconds matter. Wall-clock / civil time, no zone marker -- zone info lives in sidecar metadata.

Rationale, edge cases, normalization recipes: naming.md.

Quick Reference

• Photos / videos with EXIF -> photo-cli copy. Always try it before scripting EXIF parsing yourself. See photo-cli.md.
• Find files / inspect a tree -> fd (not find), eza -lT (not ls). See inspect.md.
• What's eating disk space -> dust (tree summary). See inspect.md.
• Find exact duplicates -> fclones.
• Find similar duplicates (rotated photos, near-duplicate audio) -> czkawka_cli. See dedupe.md.
• Bulk rename -> rnr regex for pattern-based renames, or fd ... -x mv {} {.}.<new> for one-offs in a pipeline. See rename.md.
• Bulk-rewrite text inside many non-source files (notes, configs, CSV/TSV, generated data) -> sd (not sed). For single source files prefer the Edit tool -- sd rewrites in place with no undo. See rename.md.
• Pull fields from documents -> jq (JSON -- see inspect.md), yq (YAML).
• Tweak EXIF beyond photo-cli (e.g. shift timestamps, rename single files by tag) -> exiftool. See images.md.
• Optimize images -> jpegoptim (JPEG), oxipng (PNG). Verify with jpeginfo -c / pngcheck. See images.md.
• Convert / resize images -> magick (ImageMagick). See images.md.
• Extract zip / rar / 7z / tar / tar.gz -> 7zz. See archives.md.
• Mount or extract a disk image (.img, VHD, VMDK, QCOW2) -> 7zz or extraction, apfs-fuse / ntfs-3g / mount.exfat-fuse for FUSE mount. See fuse.md.
• Archive a sorted tree -> tar --zstd. See archives.md.
• Safer delete -> gomi instead of rm. See delete.md.
• Monitor progress of a long op -> pv (pipe throughput), progress (peek at running cp/mv/dd/tar), viddy (watch the destination grow), ts (timestamp stderr). See large-fs.md.

Workflow

The pattern for any non-trivial reorganization:

1. Probe with bounded walks: eza -lT --level=2 src/, dust -d 2 src/, or fd --max-depth 2 ... | head. Don't skip this even on small trees -- it's cheap insurance against running dust . on the wrong directory. Scale safety above explains why the caps are mandatory.
2. Plan the layout before any move. Domain-driven, not type-driven (see naming.md). Sketch the target tree, map source files into it, then reach for mv.
3. Dry-run destructive ops. rnr regex and fclones group both default to a preview; fd ... -x mv does not, so prefix the action with echo (fd ... -x echo mv ...) and read the printed commands before re-running without echo.
4. Apply with the bounded forms (depth flag, --max-results, NUL-delimited pipes at scale).
5. Verify afterwards: post-move tree (eza -lT --level=2 dst/), file count parity (fd -t f . src/ | wc -l vs fd -t f . dst/ | wc -l), and -- for destructive ops -- confirm the trash holds what you expect before anything is emptied.

For media, try photo-cli copy before writing date / location sorting yourself.