[Feature Request] Non-interactive CLI commands so coding agents (Claude Code) can self-configure the status line

[axisrow]

Problem

ccstatusline is configured exclusively through the interactive TUI. That is great for humans, but it leaves one big workflow unserved:

1. Coding agents cannot configure their own status line. The natural workflow — ask Claude Code to add, remove, or rearrange a widget — dead-ends, because the only configuration interface is an interactive React/Ink app that an agent cannot drive. Hand-writing ~/.config/ccstatusline/settings.json instead is risky: the schema is not documented as a stable contract, and there is no CLI validation before an agent saves a possibly-broken config.
2. No scriptability. Dotfiles-style setup, scripted fleet installs, and config-as-code all currently require a human in the TUI.
3. TUI startup cost. For a single tweak, launching the whole React/Ink app is heavy.

Proposal

Non-interactive CLI subcommands that operate on the same settings file the TUI uses:

ccstatusline get [--json]                     # print current (post-migration) config
ccstatusline widget add <line> <widget> [--index N] [--option value ...]
ccstatusline widget remove <line> <index-or-type>
ccstatusline widget move <line> <index> --to <index>
ccstatusline set <option-path> <value>        # global options: padding, separators, etc.
ccstatusline validate [--file <path>]         # exit code 0/1 + machine-readable report
ccstatusline install / uninstall              # existing TUI actions, exposed non-interactively

Design notes / constraints:

• Reuse applyImport() and the v2.2.27 import/export logic where possible — validate + merge semantics largely already exist there.
• Preserve the current guarantees: atomic saves, symlinked settings.json written through the resolved target, invalid configs never silently overwritten. A validate-before-write step fits this naturally.
• get --json + validate + set/widget ... gives an agent a safe read-modify-write loop without ever opening the TUI.
• Error output should be single-line and machine-parseable (ideally --json) so agents can self-correct without guessing.
• The TUI stays the primary human interface; this is purely additive.

Related

• Visual drag-and-drop config editor (Flask web app) #471 (visual drag-and-drop web editor) is a human-first take on the same pain; this proposal is agent-first and adds no dependencies.

Happy to follow up with a PR if the proposed shape sounds acceptable.