Task Explorer 1.0.3

A third way of keeping the list up to date: only when you press Refresh or
F5. The list is a snapshot taken as the dialog opens and nothing moves until
you ask, so you can study it at leisure.

The Refresh button is now disabled while the list is updating itself, in either
of the continuous modes. It only ever repeated what the timer was already doing,
and a button that seems to do nothing is confusing. F5 in those modes says so
instead of silently refreshing.

Sorting and expanding no longer take a new measurement of their own. They
rearrange the numbers already on screen, which is what a still list needs and
what the continuous modes were about to do on the next tick anyway. Overall
totals on alt+T now come from the same measurement as the list, so the two
always agree.

The source is now indented with tabs, following NVDA's coding standards and the
add-on template, as suggested during the add-on store review. Nothing the add-on
does has changed because of it.

The readme now says plainly that parts of the add-on were written with AI
assistance.

SHA256: e42b9eaa2c56b61f7bf64a29d399be0e1048fd4a5f1dddb2506c6db1a4215e40 taskExplorer-1.0.3.nvda-addon

Task Explorer 1.0.2

The dialog now opens on NVDA+alt+E rather than NVDA+alt+R.

Typing in the list now searches by whole name instead of by first letter. s then l finds Slack, where before the l jumped to the first application beginning with l. Letters typed within about a second and a half of each other build up one search, and a pause starts a new one.

Part of a name works as well, so chrome finds Google Chrome. A name that begins with what you typed always wins first, so l still goes to Logic rather than to Slack.

Only the rows on screen are searched, so a search never jumps inside an application you have left collapsed. Pressing the same letter repeatedly still steps through everything beginning with it.

Task Explorer 1.0.1

Tested with NVDA 2026.2, and lastTestedNVDAVersion raised to match.

NVDA 2026.1 reset the add-on compatibility baseline, so a build last tested against 2025.3 is refused by every 2026 release. This single package now covers NVDA 2025.1 through 2026.2.

No functional changes from 1.0.

Task Explorer 1.0

Changelog

Version 1.0

First release.

Task Explorer shows which applications are using the most processor time, in
a dialog built to be explored by ear.

• A dialog on NVDA+alt+R listing running applications ordered by processor
  use, with memory use and process count on the same line, so one arrow key
  press tells you everything about an application.
• Applications are grouped by executable, so a browser's forty helper processes
  are one row. Right arrow or enter expands it into the individual processes,
  left arrow collapses it again, and left arrow from a process takes you back to
  the application it belongs to.
• Processor use is reported as a percentage of the whole machine, the same way
  Task Manager reports it.
• Sorting by processor use, memory use or name, from a combo box or from
  alt+1, alt+2 and alt+3 without leaving the list.
• alt+T announces total processor and memory use. F5 updates the list.
• End task asks the application to close first, exactly as clicking its close
  button would, so it can prompt you to save. Only if it is still running a few
  seconds later are you offered the choice to force it.
• NVDA itself is refused outright. Critical Windows processes warn first.
  Applications running with higher privileges than NVDA report that they cannot
  be closed, rather than failing silently.
• Three commands report usage without opening any window: the busiest
  applications by processor use, by memory use, and total machine utilisation.
  All three ship unassigned, to be bound from Input Gestures.
• Settings in NVDA's own settings dialog for update behaviour, update interval,
  how many applications the spoken commands announce, and whether they mention
  memory.

Notes on how it works:

• Process activity is measured directly rather than read from Task Manager,
  whose process list is a very large data grid that is slow to navigate with a
  screen reader.
• Every process is read in a single system call per sampling pass. Reading the
  same figures through psutil alone costs roughly 750 milliseconds per pass on a
  machine running 345 processes, because psutil falls back to a full system
  enumeration for each process it cannot open. A pass now costs about 6
  milliseconds, and processes a normal user cannot open stay visible instead of
  reading as zero.
• The fast path checks itself against psutil at startup and falls back to the
  slower psutil route if the two ever disagree.