You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Idea: a quiet transcript - persist thinking-hidden and allow hiding tool cards
#2044
Two display ask that today have no persisted setting:
Hide reasoning (thinking) rows for good./thinking already does exactly this (it filters kind === 'reasoning' out of the transcript), but it is UI state only — not persisted, so it is
re-enabled on every launch. This was already requested as point 2 of 关于工具调用过程折叠和思考隐藏功能的建议 #858, and the maintainer reply
there said it was out of [Feature] 过程可读性:补任务进展叙述 + 压缩工具卡篇幅 #981's scope and should be pushed as its own request.
The end state asked for is a transcript that reads as: the ⏵ status line from dsh-working-activity (the "what I am doing now" line), then the assistant's answer — nothing else
unless the user expands it.
Current state (verified on dsh-tui 0.20.1, installed profile)
thinkingFold is a two-option select: preview (2–3 live lines, folded when settled) / full
(expanded until turn end). No "hidden".
The hide path already exists in rendering: components/MessageList.tsx filters (thinkingVisible || row.kind !== 'reasoning'), and thinkingVisible starts as useState(true) in screens/Chat.tsx and is toggled by /thinking (documented as "UI state only,
not persisted" in the user guide §/thinking).
Tool cards keep TEXT_BODY_MAX_LINES = 3 (components/messages/AssistantToolUseMessage.tsx); foldTerminalCommand only folds a multi-line command header. No setting hides the card.
thinkingDisplay: hidden = persist the current thinkingVisible React state as a channel preference
(same shape as thinkingFold: config schema → channel state/preferences → /settings item → Chat.tsx initial value). The filter in MessageList.tsx is unchanged.
Running and failing tool cards stay visible in every mode (a hidden failure would leave the user with
no idea why the turn stopped). Ctrl+O / click still expands whatever the mode hides, so nothing is
lost — this is a default-view preference, not an export filter.
Why a setting instead of a default change
#858 wants less noise, #981 wanted more progress narration; the two pull in opposite directions, so the
defaults should stay as they are and the verbosity become a gear. dsh-working-activity already gives
the "what is happening now" line, which is what makes a fully quiet transcript readable in the first
place.
Related
关于工具调用过程折叠和思考隐藏功能的建议 #858 — folding tool calls + persisting the thinking-hidden setting (maintainer: push point 2 separately)
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Summary
Two display ask that today have no persisted setting:
/thinkingalready does exactly this (it filterskind === 'reasoning'out of the transcript), but it is UI state only — not persisted, so it isre-enabled on every launch. This was already requested as point 2 of 关于工具调用过程折叠和思考隐藏功能的建议 #858, and the maintainer reply
there said it was out of [Feature] 过程可读性:补任务进展叙述 + 压缩工具卡篇幅 #981's scope and should be pushed as its own request.
toolBodyLines(0–5, default 3), which givesheader + "… +N lines (ctrl+o to expand)"at0.That is a big improvement, but the card header still costs a line (plus its block margin) per call.
The end state asked for is a transcript that reads as: the
⏵status line fromdsh-working-activity(the "what I am doing now" line), then the assistant's answer — nothing elseunless the user expands it.
Current state (verified on dsh-tui 0.20.1, installed profile)
thinkingFoldis a two-option select:preview(2–3 live lines, folded when settled) /full(expanded until turn end). No "hidden".
components/MessageList.tsxfilters(thinkingVisible || row.kind !== 'reasoning'), andthinkingVisiblestarts asuseState(true)inscreens/Chat.tsxand is toggled by/thinking(documented as "UI state only,not persisted" in the user guide §/thinking).
TEXT_BODY_MAX_LINES = 3(components/messages/AssistantToolUseMessage.tsx);foldTerminalCommandonly folds a multi-line command header. No setting hides the card.toolBodyLinesis not onmain: PR feat(tools): make the collapsed tool-card body budget configurable #2019 was closed unmerged (it looks like the contributorwhitelist auto-close, not a "no"), and [功能] 工具卡正文行数可配置:toolBodyLines(0-5,默认 3) #2018 was closed as not planned.
Proposal (opt-in, defaults unchanged)
A single display preference, no system-prompt change (so no "out of scope" prompt concern):
Minimal implementation, reusing what exists:
thinkingDisplay: hidden= persist the currentthinkingVisibleReact state as a channel preference(same shape as
thinkingFold: config schema → channel state/preferences →/settingsitem →Chat.tsxinitial value). The filter inMessageList.tsxis unchanged.toolCardDisplay: header= thetoolBodyLines: 0case from [功能] 工具卡正文行数可配置:toolBodyLines(0-5,默认 3) #2018/feat(tools): make the collapsed tool-card body budget configurable #2019 (capLines()alreadyaccepts 0);
hidden= one more filter next to the reasoning one(
row.kind !== 'tool'), applied before virtualization like the thinking filter and the job-groupfold already do.
no idea why the turn stopped).
Ctrl+O/ click still expands whatever the mode hides, so nothing islost — this is a default-view preference, not an export filter.
Why a setting instead of a default change
#858 wants less noise, #981 wanted more progress narration; the two pull in opposite directions, so the
defaults should stay as they are and the verbosity become a gear.
dsh-working-activityalready givesthe "what is happening now" line, which is what makes a fully quiet transcript readable in the first
place.
Related
toolBodyLines0–5 (closed not planned / unmerged)Environment
thinkingFold: preview,/thinkingused manually every sessionAll reactions