disktree --cli: print the tree to stdout, for agents and headless sessions #74

The gap

disktree is window-only today. Everything — the treemap, marking, the review screen, and the "copy as agent prompt" export — lives behind the GUI. That leaves out callers that can only read stdout:

• coding agents: they run commands and read the output; they cannot click a window. The project already leans this way — "Agent scratch" is a category, and the review screen hands the marked list to a coding agent — but without a human at a display, an agent can't reach any of it.
• SSH, containers, CI: no display, no window.
• and it would come to the artifacts people already have (including the Windows zip), with no new binary to ship.

I know #32 adds disktree-cli with JSON output for scan/check/trash, and #47 builds on that with an agent-focused du personality. Those are strong programmatic surfaces. This is the other half: a human- and agent-readable text tree from the binary people already have, so agent -> disktree --cli -> decide works in one step while the JSON surface stays JSON. (Agents are fluent in this shape — as #47 notes, they run du -sh * constantly.)

Proposal: disktree --cli

--cli scans through the existing pipeline and prints the tree to stdout instead of opening a window:

38 GiB  /home/me/work
├──     21 GiB  55%  builds/
│   ├──     12 GiB  57%  release/
│   └──    9.0 GiB  43%  debug/
├──    9.4 GiB  25%  node_modules/
└──    7.6 GiB  20%  datasets/

38 GiB in 512.0k files, 48.0k directories — 3 unreadable

• Largest first; each row shows its size and its share of the parent; directories are trailed with /.
• Every existing scan flag applies unchanged: --apparent-size, --follow-links, --no-hidden, --disk, --cross-filesystems, the power/thread flags, and --metric files (rank by count; the row shows the count).
• -d/--depth decides how many levels print (same meaning as in the window; 1 prints only the root's children).
• The tail merges like the treemap does: 96 children kept (the LayoutOptions::max_children default), then one … N more row with the summed size; a directory cut by the depth limit is marked with … rather than reading as empty.
• Names go through the same line-safety rules as export.rs — a name with control characters can never read as two rows.
• stdout only; a failed scan exits non-zero; a closed pipe (| head) is not an error.

Implementation

I have this working already, following the repo's conventions: a report.rs in disktree-core (re-ranks children by metric like the layout pass, keeps CHILDREN_KEPT = 96 to match the treemap, shares the escaping with export.rs, comments state the why, tests beside the promise) plus the --cli path in main.rs; the README gains a short "In the terminal" section and AGENTS.md's map gets the new module. Verified under Rust 1.97.1: cargo xtask lint clean, cargo xtask test green (one permission-sensitive scan test that also fails on unmodified main when run as root was excluded).

If this direction fits, I can contribute it as a PR — and if you'd rather have it as a format on #32's disktree-cli (--format text or a plain mode) instead of a flag on the window binary, the renderer moves over as-is. Same for the name: --cli is what I built, but --print/--text reads better next to #32 and I'm happy to rename.