v2.33.1: The skill says when a list or ranking is missing records

• You're told when a list is missing records. Some stacks return the same record on two pages,
  which leaves another out. The skill now says how many are missing, for a spreadsheet export too,
  instead of presenting the list as complete.
• A ranking that may be incomplete is labelled. It is shown as provisional, never as "the top
  10", and the skill offers to narrow the question so the ranking can be read in full.

v2.33.0: Faster answers on slow stacks, a lighter start to every session, and tighter handling of secrets and suspicious data

• Faster answers on large stacks. Rankings and lists now read fewer records, in smaller pages
  fetched together. On one large stack a top-10 ranking went from about 16 seconds to about 5, with
  the same result.
• A list tells you when records are missing. Some stacks return the same record on two pages,
  so another is left out. Lists and rankings now drop the repeats and say how many are missing.
• Sessions start lighter. The connector summary you see on connecting is now written by the
  skill itself, so it reads the same every time, and each session loads about a fifth less
  instruction text.
• Secrets stay put, and data stays data. Your token and alert settings are no longer passed to
  the browser that makes reports. Names from your stack are never run or shown as links, and any that
  look like a command or an instruction are pointed out.

v2.32.0: Readable severity, a designed report on every install, and report styling that can't hide what a report says

• Severity is easier to read in reports. High, medium and low severity words and graph markers
  are darker, so each meets the standard contrast for small text. High and medium used to fall
  below even the large-text minimum.
• Every install gets a designed report. Unbranded reports used to be near-unstyled text. They
  now use the full layout in a neutral theme, and each file is about 96% smaller. Branded reports
  are unchanged.
• A report's styling can't hide what it says. It now comes from a theme the skill checks first,
  so all text stays readable, including how current the data is. A theme that fails is set aside,
  and the report says why.

v2.31.1: Saved tokens stay with their own stack, more secrets hidden, and safer stack and update handling

• Your saved token only goes to its own stack. Pointing MERIDIAN_FQDN at another address no
  longer sends it there, and stacks move accepts only the stack's own new address unless you
  confirm another, with certificate checks on.
• Connector credentials are harder to expose. Messages and errors hide more secrets, such as
  quoted passwords, client secrets and AWS key ids, and raw API requests can't reach credentials
  via a redirect or another spelling of an address.
• Safer stacks and updates. A damaged stacks file stops the skill instead of being saved over,
  switching stacks keeps machine-wide settings such as a proxy certificate, and self-update never
  installs an older version, even when forced.
• Quicker failures and sturdier reports. A failed request reports straight away instead of
  waiting 2 seconds first, PDF reports work in folders with # in the name, and Slack alerts no
  longer expand link previews.

v2.31.0: Windows token encryption and the move to meridian-cyderes.com

• Your saved Meridian tokens are encrypted on Windows. The first time the skill connects, it
  encrypts them to your Windows account, so a copied ~/.meridian folder is useless on another
  machine or account.
• Ready for the move to meridian-cyderes.com. lucidum.cloud addresses stop working on
  2026-12-05. The skill tells you when a stack is still on one and moves it for you, keeping its
  token, history and alert rules.
• A stack that has already moved is recognised. When an old address redirects to its new one,
  the skill says so and offers the move instead of reporting an error. Your token goes to the new
  address only once you agree.
• Scheduled snapshots can be stopped cleanly. schedule rm retires a job, so a task you removed
  is no longer reported as stopped at every session start. Its history is kept.

v2.30.0: Smaller downloads, retries on a busy stack, stacks rm cleanup and private-key scrubbing

• Large inventories download much less. The skill now asks Meridian to compress its responses: a
  page of 100 assets shrank 28 times on a test stack. How much faster a question gets depends on how
  fast your stack itself answers.
• A busy stack slows an answer down instead of failing it. When Meridian says it is getting too many
  requests, the skill waits as long as it asks, up to a minute, says so, and tries again.
• Removing a stack stops using its credentials. Before, its token stayed in use until you switched.
  Another saved stack now takes over and the skill says which; its cached data is deleted, and
  --purge deletes its history too.
• Private keys are kept out of connector messages. An error that quoted an uploaded key file, such
  as a cloud service-account file, is now cleaned by recognising the key itself.

v2.29.0: Snapshots on a schedule, and a job that stops gets noticed

• Snapshots can now be taken on a schedule. Ask the skill to keep tracking something and it sets
  up a daily or weekly job for Windows, macOS or Linux. You see it and approve it before anything is
  registered.
• You hear about a scheduled job that stops or fails. Each run is recorded, failures included, and
  the next session says when a job has stopped, keeps failing, or is waiting on another stack.
• A job collects for one stack only. While another stack is active it writes nothing, and
  switching stacks tells you so.
• "Start tracking X" now says whether anything will measure it. Before, a metric could be
  registered with nothing ever taking a snapshot.

v2.28.5: Risk answers follow Meridian's risk model; numeric breakdowns fixed

• Risk answers follow Meridian's own risk model. Rankings order by Risk_Score, tiers come from
  Risk_Level, and a record with no risk factor is reported as "no risk factors identified", never
  as low risk.
• Breakdowns by a number field now count correctly. summary --by on a numeric field or
  SmartLabel used to report every group as 0.
• The reference docs use illustrative figures throughout.