v0.10.4

Fixed

• The AI capture is now something you can see. Clicking AI capture
  closed the window and then looked like nothing had happened. The analyst
  session was created — it ran for minutes and created the card — but on a
  current DSH host the session tree groups conversations per workspace and draws
  only a short window of each group. The analyst runs in the idea's
  workspace, not the one you were looking at, so its session landed out of
  sight; the natural reaction (clicking again) started more analysts in
  parallel, and each one wrote another card that then had to be merged away. The
  board now switches to the analyst session as soon as the analysis is
  queued, exactly like Open session on a launched card. Re-analyze and
  Find similar do the same — their report is the thing you asked for. A
  shell that cannot navigate still starts the analysis (its console says so),
  and a failed launch never opens an empty session.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.10.4

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.10.4/CHANGELOG.md

Full Changelog: v0.10.3...v0.10.4

v0.10.3

Fixed

• Launch works again when the TaskBoard itself is disabled. Turning the
  TaskBoard's master switch off hides its board but not its API: the ideas plugin
  kept choosing the card backend, and every Launch click died on the board's
  task board is disabled refusal — a red error, nothing started. A board that
  refuses every action is now treated like a board that is not installed: the idea
  runs in a fresh session instead. A refusal that concerns ONE card (a run already
  in flight, an archived card, a permission confirmation) still shows the board's
  own reason, and the idea keeps its card, so a re-enabled board is used again on
  the next launch.
• A fresh run is no longer overwritten by its old card. While such a fallback
  run is in flight, the card the idea still carries no longer clears the running
  badge, replaces the session link with an older card execution, or opens the
  review gate from a frozen card status. The session owns the run until it
  settles; the card keeps mirroring its status for display.
• Mirrored cards follow the deployment's permission again on newer DSH hosts.
  A newer DSH host reports the TaskBoard's session default at a new place in its
  snapshot. Reading only the old place did not fail: it silently made every newly
  mirrored card read-only, so its first run started without write access. Both
  places are read now, newest first.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.10.3

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.10.3/CHANGELOG.md

Full Changelog: v0.10.2...v0.10.3

v0.10.2

Fixed

• A launch that runs in a fresh session reaches the session again. On a newer
  DSH host, clicking Launch on an idea with no TaskBoard card failed before any
  session existed: the host requires every request to name who is speaking, and this
  plugin's direct-session path named nobody. The launch window showed the raw failure
  text and the idea stayed in the backlog, with nothing started. It now speaks as one
  owner of its own — created once for the whole board and kept for the life of the
  plugin, never retired early — so the session is created, named after the idea,
  given the chosen model and prompted, on a host that enforces this rule and on one
  that never asked for it. Card-backed launches are unchanged, and nothing about the
  board's data changed.
• The launch window now says why the reasoning effort is missing. The effort
  selector is offered only when the run reaches a fresh session, so an idea that
  runs through its TaskBoard card shows the model picker and no effort selector —
  a card pins the model and nothing else. That omission was silent, and on a
  board whose ideas all carry a card it read as a broken feature. The window now
  names the reason where the choice would have been, and the default line no
  longer prints a stored level beside a model when the run cannot use it.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.10.2

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.10.2/CHANGELOG.md

Full Changelog: v0.10.1...v0.10.2

v0.10.1

Added

• Choose how hard a model thinks, in the launch window. When you pick a
  model that supports a reasoning effort, the launch window now offers a second
  selector with the levels that model declares — its own default is preselected,
  and Model default pins no effort at all, so the run keeps whatever the
  model would have done anyway. The levels come straight from the model catalog
  (they are adapter-owned names, not a fixed list), and the same selector appears
  in the capture, re-analyze and find-similar windows, where a fresh session
  always reaches the model. The effort travels with the model you chose and is
  pinned on the session before the run starts.
  The selector is honest about where it can go: it is offered only when the run
  reaches a fresh session. A TaskBoard card carries the model and nothing else,
  so on a card-backed idea the window shows the model picker but not the effort
  selector — better to omit a choice that could not be kept than to offer one.

• The workspace default remembers the reasoning effort too. Save as the
  workspace default stored the model alone, so an effort you had just chosen
  was dropped the moment you saved: the next launch fell back to the model's own
  default. The default now stores the pair — model and effort — and a launch
  that picks nothing lands on both. The save is a gesture on the pair, so
  changing only the effort is a real change and stays available, and the stored
  line names the level the run will use.

Fixed

• A refused button no longer looks like a broken one. The launch window's
  Save as the workspace default stays unavailable until the model you picked
  actually differs from the one already stored for that workspace. It used to
  keep the same colours and the same pointer cursor as a live button, so the
  click landed on a disabled control and simply did nothing. A disabled button
  now looks disabled, and while the save is refused the window names the missing
  gesture — Already the default for … when the pick is the stored one, or
  Pick a model above … when no model is picked yet — instead of leaving you
  to guess at a dead click.

• A refused save now says what to do. A settings write refused by the
  revision fence — another tab, another client, or a page left open while the
  settings moved — answered The settings changed elsewhere — please retry,
  which sends you straight back into the same refusal: the page still holds the
  old revision, so retrying the identical save fails again. The message now
  names the gesture: The settings changed since this page loaded: refresh it,
  then try again.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.10.1

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.10.1/CHANGELOG.md

Full Changelog: v0.10.0...v0.10.1

v0.9.0

Added

• The AI analysis can now relate an idea to its neighbours. When the analysis
  establishes a real link, the analyst records it: relates to for a shared
  subsystem or constraint, and blocks with the direction that means — the
  card that depends on another is blocked by it, never the reverse. Only
  explicit relations, using resolved ids, at most three per run, and named in
  the report. An empty relation graph stays a normal state.

• A re-analysis no longer erases your tag instructions. The analysis rewrites
  the tag list, and a tag also carries the line shown to you before each launch.
  A tag the analysis keeps now comes back with that line intact.

• Six agent tools: drive the board from any session. When your DSH deployment
  serves agent tools, any chat can now work the ideas board directly — no shell,
  no hand-written JSON, no browser. ideas_list reads a filtered page of your
  backlog, ideas_get reads one idea in full, ideas_capture captures an idea
  together with its priority opinion, ideas_triage records value / effort /
  rationale / rank and re-ranks a workspace group, ideas_launch starts the
  execution, and ideas_review settles the review gate with approve,
  follow-up or decline. They go through exactly the same gate as the web
  board, so a call that would be refused in the interface is refused the same way
  here, and an idea written by an agent shows up on your board at once — marked as
  the agent's work in that idea's history. An agent still cannot write a run state
  or claim a task card; those stay the host's. If your deployment serves no agent
  tools, nothing changes: the board keeps every feature and the HTTP channel.

• Every idea now remembers what happened to it. Each idea keeps a short
  activity log — who did what, and when — and the editor shows it as a compact
  timeline under the description, right above the Approve / Follow-up / Decline
  buttons. It is written by the board itself: the capture, the ranking changes,
  the launch, the run that finished or failed, the approval, the follow-up, the
  decline with its reason. Entries are attributed, so you can tell your own
  changes from an agent's and from the run's. The log keeps the last 50
  entries and nothing more, and an idea that has recorded nothing yet shows
  nothing at all rather than an empty box — so "nothing has happened" is never
  mistaken for "something is missing".

• Your ideas travel with their history. The JSON export and import carry the
  activity log, and the markdown export prints it as a short Activity block
  under each idea — so an archived document can still answer why was this
  declined?, which a last-state record never could.

• Re-analysis reads the real past. When you ask for a fresh analysis of an
  idea, the analyst is now handed what actually happened to it before — the
  decline, the delivery, the archive-and-restore — and is told to read it before
  writing. A re-analysis can no longer re-propose something the board already
  refused.

• Every finished run leaves a delivery note. When a run ends, the board keeps
  the closing words of the run right under the description of the idea it worked
  on — on the card in the To review column, in the idea editor above the
  Approve / Follow-up / Decline buttons, and on the row of a delivered idea in the
  Delivered list. It is the last thing the run actually said, not a summary
  written for you: the board never invents one. When a run leaves nothing behind
  — it was interrupted before answering, or the workspace it ran in is gone — the
  note says exactly that, so an empty delivery can never be read as a successful
  one. The note is short by design (a couple of lines; longer answers are cut) and
  never changes what the run did: Open the session remains the way to watch
  the whole thing. The note follows the idea through the markdown export too, and
  nothing you type can overwrite it: it belongs to the run, not to you.

• The sidebar entry now tells you what is waiting. A small amber count sits on
  the Ideas icon and shows how many ideas of the current workspace scope are in the
  To review column. It reads the board's own list already loaded in the page, so
  it costs nothing, needs no extra polling, and disappears as soon as the gate is
  empty. With Remember the workspace turned on, the count follows the
  workspace you last worked in rather than mixing every workspace together.

• A quiet Stale marker on ideas nobody has touched. An open idea with no
  update for a while wears a discreet Stale badge next to its state, on the
  Overview cards and in the Priorities list. How long counts as too long is yours
  to set under Stale after (days), which ships at 30 days; set it to 0 to turn
  the marker off entirely. The marker is a display aid only: it is drawn when the
  board paints and nothing is written to your ideas. It never appears on an idea
  that is already under review, delivered or declined — those have a state that
  says more than a date does.

• A new idea card carries your TaskBoard's own permission. Until now every
  mirrored card was stamped read-only whatever your deployment allowed, so an
  idea could never write anything when its card ran — while a card whose
  permission sat above the TaskBoard's session default refused to start until
  you walked to the board and confirmed it. A new card now takes the level your
  TaskBoard is configured with, never more, and launching raises an older card to
  that same level before the run. On a deployment left at read-only nothing
  changes at all; the plugin follows whatever you set, and never asks for
  anything above it, so a launch stays one click — no detour to the board to
  confirm a permission. A card you raised yourself is never lowered, and a card
  whose permission the board did not report is never written to.

• A setting for the launches that have no card. When you launch an idea while
  the TaskBoard plugin is absent, the run happens in a brand-new session. That
  session now starts at the level you pick in Direct-launch permission
  (read-only, workspace-write or danger-full-access), applied before the
  very first turn so the run never starts fenced. The default is
  workspace-write: the run brief asks for an implementation, so a read-only
  session would answer with a plan and settle the run having written nothing. If
  the elevation cannot be applied, the launch stops with the reason instead of
  quietly starting without it. Card-backed ideas ignore this setting.

Changed

• An update now replaces the installed analysis prompt, and keeps what it
  replaced. The prompt was installed once and never replaced, so an instance
  that had installed an older version kept using it after every upgrade — the
  new features simply went unused, with nothing to show for it. Every start now
  writes this version's prompt, and the file it replaced is kept beside it
  (SKILL.md.<timestamp>.bak, the five most recent), so a prompt you wrote by
  hand is never lost: the start-up log names the copy, and the README says how
  to put it back.

• A long Open column stays responsive. Once the Open column grows past a few
  hundred ideas, the board draws only the cards you can actually see and leaves
  the rest behind the scrollbar, so scrolling, searching and dragging no longer
  get slower as the backlog fills up. Everything you can do still works on the
  whole column, not just on the part of it that happens to be on screen.

  • Drag & drop still lands where you point. You can drop a card before one
    you have scrolled past and one you have not: the written order is built from
    the entire column, not from the cards that happen to be drawn.
  • A shift-click range still covers the whole block. It selects every row
    between the two you clicked in the column's own order, including the ones
    scrolled out of view, and the bar keeps stating the truth — n selected of
    m shown, where m is the filter, never the window.
  • A card keeps its place while you read. Cards are not a fixed height, so
    their real height is measured as they appear — and every measurement is
    anchored to what you are looking at, so the column never slides under a
    resting hand.
  • "Go to idea" still works on a card you have never scrolled to. The board
    brings it into view first, then opens and rings it, exactly as before.
  • A standing notice appears above the Open column once it holds 300 or
    more open ideas, naming the number. It tells you the size of what you are
    looking at before you start scrolling, rather than pretending the board is
    struggling.
  • Short columns are untouched. Below that size the board draws every card
    exactly as it did before, so a normal backlog behaves identically.
  • Nothing changes for your data. The board does this entirely on its own:
    no new stored field, no new API call, and the JSON export/import, the
    snapshots and the markdown export are byte for byte what they were.

• A workspace can carry its own launch model. The launch dialog used to ask
  which model every single run should use. Set one for a workspace and it stops
  asking: the dialog just tells you which model the run will use, and your click
  confirms.

  • Remember this model for {workspace} is offered right where you make the
    choice, in the launch dialog. From then on every idea launched in that
    workspace runs on it.
  • Change… brings the picker back over the remembered model, so one run can
    use another model without changing what every later run will use.
    Save as the workspace default is what makes a new pick stick.
  • Forget this model removes it, and the dialog goes back to asking.
  • **A workspace with no default behaves exactly as befo...

v0.7.6

Fixed

• The header gear and the "Open the TaskBoard" button work in the Desktop app. In the Desktop app, Settings lives in the account menu instead of behind a standalone button, and the TaskBoard switch relied on a window call that shell does not always answer. The gear now opens Settings through whichever entry your window offers and still lands on the Ideas section. The TaskBoard button now switches to the board the same way clicking its sidebar entry does, then filters on the idea title as before. The browser version behaves exactly as before.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.7.6

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.7.6/CHANGELOG.md

Full Changelog: v0.7.5...v0.7.6

v0.7.5

Fixed

• The Ideas board works in the DSH Desktop app again. In the official
  Desktop app every ideas request answered 403 and the board showed
  Host operation failed: forbidden — the panel was unusable, while the same
  install worked in an ordinary browser tab. The desktop shell talks to the
  Host on your behalf and rewrites the request on the way, which made the
  panel's own calls look like a stray local script to the fence. The fence now
  also accepts the shell's own sign-in credential as proof of an application
  running on this machine, so the board, the search, the settings and the
  launches all answer again. Nothing is loosened for anything else: a plain
  curl, a request from another machine, and a request naming another site
  are still refused with 403, exactly as before.
• Installing this version replaces the whole plugin, so a hand-patched copy of
  the plugin in a profile is superseded by this release rather than kept
  alongside it.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.7.5

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.7.5/CHANGELOG.md

Full Changelog: v0.7.4...v0.7.5

v0.7.4

Fixed

• The Ideas row is a real panel row now. It sits with Plugins, Task Board
  and Skill Center, in their row box, font and highlight, and it behaves like
  them: clicking another panel closes the board instead of leaving it stuck open
  behind the new one. Clicking Ideas again, or Back to chat, still closes it.
  Previously the row was a button this plugin injected itself and hid the
  conversation with a stylesheet, which is why only Ideas behaved like a toggle.
• A launch refused by the card's permission gate now tells you what to do.
  The board showed the Host's own English sentence and stopped there. It now
  names the card, explains that the permission is above the session default, and
  offers Open the TaskBoard — which switches to the board panel with the
  filter already set on the idea title, so the card is the one waiting in front
  of you. Copy the title is there for the case where the board does not
  answer. The permission itself is still yours to confirm, in the board: the
  plugin never confirms it for you.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.7.4

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.7.4/CHANGELOG.md

Full Changelog: v0.7.3...v0.7.4

v0.7.3

Added

• The list rows now show the run state. The Priorities ranking and the
  Delivered log carry the same pills as the Overview cards — Running,
  Task failed, follow-up of #N, and Open session — in the row's
  top-right corner. The "is this one already being worked on?" answer is
  available on the list you actually read, without opening the card.
• The Open column can be ordered by creation date. The new Open column
  order setting offers Creation date (oldest first), Creation date (newest
  first) and Rank.
• A separate switch floats the running work to the top. Show running ideas
  at the top (on by default) lifts the ideas whose run is in flight above
  whichever of the three orders you picked, without changing that order.

Changed

• The Overview's Open column now reads oldest-first by default, so a backlog
  scans like the diary it is. Pick Rank to get your hand-set ranking back; the
  setting applies to that one column, and the Priorities tab always stays on the
  rank it prints.
• Only the running work is floated. A failed run keeps its red Task
  failed pill but stays where the selected order puts it — a date-ordered
  backlog that jumped every failure to the top would stop reading as a diary.
• Neither setting stores anything: the board refreshes every 2.5 s, and it can
  never overwrite the ranking you set.
• While the Open column displays something other than the stored rank, its drag
  grip is off and says why. The card buttons still move an idea to another
  column, and choosing Rank with the running float off brings the grip back.

Fixed

• A restored idea no longer shows a stale Delivered {date} pill in the
  Priorities list. Restoring clears the archive stamp but keeps the delivery
  date, so the row was announcing "this one is done" on an idea you were about
  to pick up.
• Dropping a card into a reordered Open column no longer writes back the ranking
  it already had. The anchor you saw came from the displayed order while the
  write resolves in rank order, so the drag did nothing visible while still
  calling the host.

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.7.3

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.7.3/CHANGELOG.md

Full Changelog: v0.7.2...v0.7.3

v0.7.2

Fixed

• The npm and changelog links at the bottom of a release note pointed at a
  doubled version prefix, so they did not open. (The v0.7.1 release page still
  shows the wrong links: it was published before the fix.)

npm: https://www.npmjs.com/package/dsh-plugin-ideas-manager/v/0.7.2

Changelog: https://github.com/EiffelBS/dsh-plugin-ideas-manager/blob/v0.7.2/CHANGELOG.md

Full Changelog: v0.7.1...v0.7.2