aube run completion

[skoch13]

It seems like aube run/aubr doesn't produce shell completion for package.json defined scripts. Is there a solution for that?

[yhay81]

Not as shell Tab completion at the moment. The generated completion only knows the static CLI grammar.

In aube.usage.kdl, run declares a plain optional arg "[SCRIPT]" but no dynamic completion provider. aube completion delegates to usage g completion, so the generated Bash/Zsh/Fish script has no code that reads the current project's package.json.

The built-in solution today is to omit the script name:

aube run
# or
aubr

On an interactive TTY, prompt_for_script() finds the nearest package.json, preserves the order of its scripts entries, and opens a filterable picker. Select the script there and press Enter. In non-interactive use, pass the script explicitly.

So if you already installed/regenerated completions correctly and aube run <TAB> does not list package scripts, that is the expected current limitation; regenerating the completion file will not change it. Dynamic package.json script completion would need a new completion hook/provider (plus cwd/project-root handling), while the no-argument picker is the current supported workflow.

[skoch13]

Thank you for the reply. Please consider this as a feature request then if possible.

Speaking of interactive run: the view is broken in several ways:

• It shows the script contents which can be long enough to make it look ugly, and I doubt showing it that useful to be honest, so I’d personally like to hide it
• But what’s more important moving within the list or searching is completely broken as it scrambles the output, repeating lines and cursors

[yhay81]

Those are two distinct actionable items, and the first one is confirmed directly by the implementation. The picker currently constructs every display label as:

let label = format!("{name}: {cmd}");

So showing the full command is intentional today, and changing that to the script name alone (or truncating it) is a separate UX request from dynamic shell completion.

The scrambled/repeated lines are a rendering bug, not just a preference. The picker is provided by the demand terminal UI crate, so a useful reproduction report should include:

• exact aube version;
• terminal app and shell;
• echo $TERM (and whether tmux/screen is involved);
• terminal window width;
• a short recording or screenshot showing one arrow/search action producing duplicated lines.

Please also mention whether it still happens in a plain terminal outside tmux. Until that rendering issue is fixed, passing the script explicitly (aube run <script>) avoids the picker. The exact label construction is in prompt_for_script().

[jdx]

All three items here turned out to be actionable, and there are PRs open for each.

1. Script completion — #1108

aube run <TAB> / aubr <TAB> now completes the scripts in the nearest package.json, with each script's command as the description.

The diagnosis above was right about the cause: aube.usage.kdl is generated by clap_usage, which only sees the clap tree and can't express a dynamic completer. The missing piece is that usage does support one — a complete "script" run="…" node — so the fix is a hand-written spec fragment concatenated onto the generated spec, the same approach mise uses. Behind it is a hidden aube run --complete that prints the script list.

No change to how you install completions: aube completion <shell> already re-derives the spec from the binary at TAB time, so it picks up the right project automatically.

2. The command text in the picker — #1109

The script name is now the label and the command moved into an aligned second column, truncated to fit:

  build      tsc -p tsconfig.build.json && rollup -c rollup.config.mjs --enviro…
  test       vitest run --coverage --reporter=verbose --pool=threads --poolOpti…
❯ lint       eslint . --ext .ts,.tsx --max-warnings 0 --cache --cache-location …
  dev        vite --host 0.0.0.0 --port 3000

Names line up and the command reads as secondary, so it stops dominating the row without dropping information. On a terminal too narrow for any of it, the rows fall back to names alone.

3. The scrambled output — #1109 + jdx/demand#190

Not a preference issue and not specific to your terminal — a genuine rendering bug, and the same root cause as #2. No repro details needed in the end; driving the picker through a pty reproduced it directly.

demand sizes its redraw by counting logical lines but clears with clear_last_lines, which works in physical rows. Any row wider than the terminal wraps into rows the clear never erases, so each keypress stacks another copy of the frame. At 80 columns, after two arrow presses:

 0 |Select a script to run
 1 |package.json scripts
 2 |❯ build: tsc -p tsconfig.build.json && rollup -c rollup.config.mjs --environment
 3 |Select a script to run          ← stale
 4 |package.json scripts
 5 |  build: tsc -p tsconfig.build.json && rollup -c rollup.config.mjs --environment
 6 |Select a script to run          ← stale

The identical frames at 200 columns are clean, which is what pins it on wrapping. Filtering was hit for the same reason.

Two fixes, because aube's own change isn't sufficient on its own: keeping the rows inside the width fixes the picker, but demand's help footer can still wrap on a narrow window and retrigger the stacking. jdx/demand#190 makes the redraw wrap-aware across Confirm, Dialog, Input, List, Select, and Spinner. Verified end to end at both 34 and 80 columns.

Thanks for the report — the second half in particular was a real bug that had gone unnoticed upstream.

This comment was generated by Claude.

[skoch13]

Thanks. Is it also possible to start the search immediately in interactive mode by typing?

[skoch13]

Another two points if I may:

1. can interactive mode respect the current terminal theme and colors?
2. can one have a global switch not to show the script contents? I know what they do, just want to have an interactive way of executing one