- pins: CI mount-gate CLI, dsh.plugin.json engines floor, and the
@deepseek-ai/* peer floor / devDependencies baseline move to
0.1.2-alpha.5 (dsh-client-locale trails at 0.1.2-alpha.3 — no such
release upstream; ^0.1.2-alpha.3 already accepts the alpha.5 runtime)
- code: migrate the alpha.4 Session.events removal to snapshotEvents()
(sidechat transcript live reads, fork inheritance, jobs.output replay,
subagent activity) and drop the seedLength meta field the host removed
from CreateAgentOptions.meta
- gates: pnpm peers check clean, typecheck + 1137 unit tests green,
real-host mount smoke 14/14 on npm-published 0.1.2-alpha.5
getEntriesByType widens to PerformanceEntry; transferSize/
decodedBodySize live on PerformanceResourceTiming (the e2e specs sit
outside tsconfig.build.json, so the bundler never saw this).
The bottom-push rule anchored on #root :has(> [data-slot=conversation]):
a :has() match cache invalidates on every #root subtree mutation, and
the streaming chat mutates that subtree constantly — every layout-push
frame (the bottom drag writes --dsh-sidebar-height per frame) paid a
whole-document selector re-evaluation. The shell's locator (which
already holds the measured column node for the ResizeObserver) now
tags it with [data-dsh-center-col]; the rule evaluates once per node.
The tag retargets with the ref on boot swap / HMR node replacement and
is removed on unmount; the drag lane pins exactly-one-tagged-node and
its identity as the conversation slot wrapper's parent. The dead
[data-pane] branch (renamed away in alpha hosts) is gone.
TextEditor: the CodeMirror updateListener re-stringified the WHOLE
document into React state on every keystroke (O(docLength) per key,
re-rendering the shell per key). The listener now only flips the dirty
flag; the preview's draft is snapshotted from the live view when
entering preview (or on a content swap), and save() already reads the
live document.
The core client bundle statically imported every third-language
dictionary (ja/de/fr/…/zh-HK/MO/TW, ~640KB of source — roughly half
the bundle's weight) for a code path most sessions never enter: the
sidebar's own t() only consults zh/en plus the better-locale override
store, and the dicts' ONLY consumer is that store's register() call.
They now ride the existing lazy-chunk machinery as lib/client-locale.js
(fetched from /sidebar/bundle only when better-locale is actually
installed; a generation counter keeps the async registration moot when
the integration effect re-runs or disposes mid-flight). Until the chunk
lands, t() keeps the zh/en chain; the store's own revision bump on
register re-renders the chrome once the dicts arrive.
Core client.js: 1.46MB → 822KB (-44%). The chunk-boundary key-set
check moves to the existing runtime test (every shipped dict stays
key-set-equal to zh).
SideChat transcript: every 2s poll rebuilt ALL row objects, so every
row (markdown parse + DOMPurify included) re-rendered even when the
log had not changed. transcriptRows now takes the previous poll's
rows and re-adopts the position-aligned rows whose content is
identical (reuseRows) — only the changed tail rebuilds. The view
feeds the memoized array back through a ref; unit tests pin the
identity contract (unchanged rows keep object identity, changed rows
and everything past the change rebuild, tool-result landings never
re-adopt).
FileTree: expanded/revealed membership was an includes() scan per
rendered row (O(rows × expanded) per render on deep trees); one Set
per render keeps it O(1) per row.
Shell: buildNewTabOptions was inlined in the two Workbench JSX
sites — two fresh arrays per render (drag state, resizes) re-rendered
every LeafView's + affordance. One shared memo now.
DiffPane resize / TerminalView ResizeObserver: pointer and resize
streams fired a setState / glyph-measuring fit() per event; both ride
per-frame batching now (frame-batcher in the core, a local rAF gate
in the lazy terminal chunk). SessionLens sizes each row's content
with a Blob (a UTF-8 encode of the whole edit) — that moved from
per-row-per-render into one memoized pass per changed fold.
Boot: loadPrefs and loadExternalDisable each fetched the settings
document, serially, before the first paint — and the second had no
timeout, so a stalled wire kept the sidebar unmounted forever. Both
decisions now ride ONE fetch (loadBootDecision) under the existing
2s race.
GitLens: the 2s silent poll re-listed worktrees on every tick (a
second git process per tick for a list that almost never changes, and
a fresh worktrees array re-render each time). Silent ticks now fetch
only the selected checkout's status; every 15th tick (~30s) re-lists
and re-runs auto-selection, so an agent-created linked checkout is
still picked up mid-session.
ChangesTab: extractFileOps parses every accumulated tool/call (up to
4000 events); the poll ran it once for the badge and the render ran
it AGAIN per tick, and the fresh array defeated the downstream memo.
The fold now runs once per poll and its result rides a ref shared by
badge and render (reset on session switch).
A measurement harness, not a pass/fail gate (beyond the mount sanity
it shares with the other lanes): an init-script probe samples the
navigationStart → host-attach latency, a longtask collector runs
through the full built-in tab sweep, sidebar-shaped resource entries
report transfer/decoded/duration, and a frame sampler records the
bottom-strip drag's pacing (p50/p95/max) plus the width-leak guard.
Results print as single-line PERF_JSON records so a shell run can tee
and grep them; before/after comparisons land in
docs/plans/2026-08-31-perf-optimization.md. Baseline (main @ d70db8f,
3 runs): mount ≈ 377ms, 2 longtasks ≈ 161ms, drag p95 = 10ms,
width leak = 504px.
The bottom drag applied the panel's persisted width preference to
--dsh-sidebar-width regardless of panelOpen (only the idle push effect
gated it through layoutPushSize). With the right panel closed, the
first drag frame jumped the variable 0 → ~400px: #root lost 400px
instantly, the host frame's viewport dropped across its 1024px
auto-collapse breakpoint, and the NATIVE left sidebar snapped to its
56px rail — then snapped back on release. The bottom strip is the only
drag reachable with the panel closed, which is why only it showed.
applyDrag now rides the same gate as the push effect for the PUSHED
width; the panel-side DOM writes (panelRef/bottomRef/lastDragSize)
keep the raw width — the bottom panel's right edge derives from it.
The e2e regression pins the pose: panel closed + bottom open, drag the
strip, and every sampled frame must keep the pushed width at 0, the
host grid template, the left column width and #root's right edge
unchanged. Verified red on the old code (leaked 504px).
Point-by-point audit of all 117 commits between alpha.2 and alpha.3:
every host contract this plugin relies on (token auth, slash RPC,
MarkdownText labels, the event-stream/persistence APIs behind
sidechat.events, SettingsNamespaceInput, SUBAGENT_DESCRIPTOR_VERSION
(still 3), dsh-client-store, the profile loader, the node-pty pin) is
unchanged, so no plugin code needed rewiring. alpha.3's breaking
changes (required BeginSubmissionInput.mode, the renamed
attachment-invalid subagent error, the removed SQLite persistence
backend, the identity-gated projection change feed) do not touch this
plugin.
What did need adapting was the dependency tree: the alpha.3 npm
artifacts ship with workspace-era metadata that does not match their
published code —
- dsh-subagent@0.1.2-alpha.3 imports `admitPromptContent` from
@deepseek-ai/dsh-attachment without declaring that dependency, so
resolution floats to a stale 0.1.1-rc.1 copy that lacks the export
(broke 5 test files at load time);
- six transitive peers (dsh-scope, dsh-session-projection,
dsh-system-prompt, dsh-user-approval, dsh-util-time, dsh-code-runtime)
drifted to pre-alpha.3 versions against the ^0.1.2-alpha.3 ranges the
alpha.3 packages declare;
- the alpha.3 line peers on @deepseek-ai/cordis ^4.0.2 (was ^4.0.1).
Fix: pin the seven stragglers via pnpm-workspace.yaml overrides (with
a note to drop them once upstream fixes its metadata), narrow the
cordis peer to ^4.0.2, and bump the baseline pins — CI mount gate,
dsh.plugin.json engines, peer/devDependencies, README badges, plus
SIDEBAR_SERVICE_VERSION in lockstep.
Verified on the npm-published @deepseek-ai/dsh@0.1.2-alpha.3:
typecheck clean, 107 test files / 1137 tests green, `pnpm peers check`
reports zero unmet peers, and the real-host mount smoke passes 14/14.
Rendering-layer upgrade for the Side Chat transcript (data already rides
the raw events; the mapping just stopped discarding it):
- Tool rows with a derivable structure now render the host's
ui-primitives atoms instead of a generic double <pre>: bash calls map
onto TerminalBlock (command from call args, output + exit pill
recovered from the result's trailing exit marker), edit/write map onto
DiffBlock (call-time literal replacement, refined by the result meta's
contextual hunks), read maps onto ReadBlock (line-window meta, same
semantic narrowing the host applies). Every narrowing declines to the
generic row on malformed input.
- turn/end now emits a turnSummary row: token usage aggregated from
assistant/message.usage (output summed across steps, input from the
last request) plus the envelope-time wall duration, printed with the
main conversation's formatTokens/formatDurationMs cadence.
- The client entry injects 'connection' (DSH 0.1.2-alpha.2+) and the
thread view shows the host's ConnectionIndicator while the Remote
transport is down; the reconnect control calls reconnect() and a
disconnected -> connected transition pulls the transcript immediately
instead of waiting for the next poll tick.
19 new locale keys propagated across zh/en and all 19 third-language
dictionaries; unit coverage in sidechat-transcript.spec and a new
sidechat-view render spec (jsdom). Real-host mount smoke: 14/14.
DSH itself is still on its 0.1.2-alpha line (npm `alpha` dist-tag), so the
plugin mirrors it: v0.18.0-alpha.0 publishes under the npm `alpha`
dist-tag and supports DSH 0.1.2-alpha.x only. Stable-DSH (0.1.0-rc.8 ~
0.1.1-rc.2) users stay on v0.17.1 (npm `latest`). Verified end-to-end
against a real 0.1.2-alpha.2 host: full vitest suite, typecheck,
build/pack, consumer type surface, and the headless mount smoke 14/14
(the prerelease version also passes `dsh plugin add file:`).
- peer floor raised `^0.1.0-rc.8` -> `^0.1.2-alpha.2` (the mount install
no longer warns: the old floor could not semver-match a prerelease
host); dsh.plugin.json engines.dsh `>=0.1.2-alpha.2`.
- Dropped the pre-alpha compatibility layer:
- e2e host RPC collapsed from the dot/slash dual dialect to slash-only
(rpcAttempt; the 404-probe negotiation and its cache are gone), and
parseLaunchUrl now requires the ?token= launch URL;
- MarkdownText labels collapsed to the nested single shape (the cast
became unnecessary against the alpha.2 declarations);
- the defunct `@deepseek-ai/dsh-client-runtime` row is gone from the
chunk externals allowlist (chunk-loader / tsdown / two specs) and
from dsh.client.inject.
- market-manifest guard now allows an exact prerelease version (alpha
cuts are not the market's verified `latest` target); release.yml picks
the npm dist-tag from the version (prerelease -> `alpha`, else
`latest`) so an alpha can never hijack `latest`.
- Docs retargeted: AGENTS.md §2/§3, README/README_EN badges + v0.18.0
-alpha.0 section + install note (`dsh-better-sidebar@alpha`), guide.
DSH 0.1.2-alpha.2 is on npm under the `alpha` dist-tag, so the CI mount
pin and the @deepseek-ai/* devDependencies baseline move up from
0.1.1-rc.1/rc.2 (the dead dsh-client-runtime devDep goes away with
them). Verified end-to-end against a real 0.1.2-alpha.2 host: full
vitest suite (1106), typecheck, build/pack, consumer type surface, and
the headless mount smoke 14/14 — the token URL, slash RPC and
MarkdownText labels contracts are unchanged from alpha.1.
- dsh-settings dropped the runtime `settingsNamespace` export:
namespaces are validated at compile time (SettingsNamespaceInput), so
the host passes the 'dsh-better-sidebar' constant directly and the
fake settings service in smoke.spec casts to the branded type.
- dsh-subagent bumped SUBAGENT_DESCRIPTOR_VERSION 2 -> 3: the version is
stamped by the host package; the sidechat seed assertion now follows
the constant instead of pinning a literal.
- Restored SessionEvent.ignorable and the RemoteError wrapping were
verified to have no impact on this plugin.
- e2e scratch profiles (e2e-mount.sh / e2e-aggregate-mount.sh) add
'@deepseek-ai/*' to minimumReleaseAgeExclude (mirroring the repo
root): alpha releases are often <24h old when the lane runs, which
pnpm 11's default minimumReleaseAge would refuse to install.
- vitest excludes .worktrees/** — local task worktrees carried stale
code against the shared node_modules and broke on the devDeps bump.
- Version lockstep 0.17.2 (package.json, dsh.plugin.json,
SIDEBAR_SERVICE_VERSION, README/AGENTS/guide docs).
DSH 0.1.2-alpha.1 removed the client `ctx.connection.api` face (Remote
gateway migration), so every Side Chat transcript poll (`sessions.history`)
threw TypeError into the view's catch and was swallowed — the tab rendered
an empty transcript forever, silently (no console error, so the mount lane
never caught it). The successor APIs are not a drop-in either:
session/follow|page reject plain session addresses for origin:'subagent'
logs (agent-busy fence) and paging cannot read past the current cursor.
Move the transcript to the plugin's own wire — one transport for every
DSH generation:
- new `sidechat.events` route: live threads read the in-memory
`agent.session.events`; cold threads read `sessionPersistence.inspect`
(both seams exist since 0.1.0-rc.8; alpha.1's inspect returns the
expanded logical log, never the packed chunk-row encoding). The
inherited seed is cut host-side (`threadOwnLogEvents`), polls ride an
`afterSeq` delta, and the response is tail-capped at 8000 events (the
old client walk ceiling).
- SideChatView.fetchThread is single-path; collectOwnEvents (the client
walk) is gone, and so are the ctx.connection mirror and inject entry
(no consumer left).
- transcriptRows keeps re-cutting at session/end-seed as defense in
depth.
Verification:
- unit: sidechat.events route (seed cut, afterSeq delta + never
resurrecting the seed, cap, cold/live sources, 404/503/bad-request),
full suite 1104 passed; typecheck and build (purity gate) green.
- mount e2e vs a real dsh-v0.1.2-alpha.1 source build: 14/14, including
the Side Chat tab sweep polling sidechat.events (performance resource
assertion) and live + post-dispose (persisted) route reads.
- mount e2e vs PATH dsh 0.1.0-rc.8: 14/14 (no regression).
- New persisted built-in pref workspaceFence (default true) under the
files card's secondary settings modal; host reads it per request via
the settings seam (fenceEnabledOf — absent/malformed stays ON).
- ensureWorkspacePath/ensureWorkspaceWritePath gain a fence param:
disarmed skips the containment assert but keeps realpath resolution;
wired through fs.tree/fs.read/fs.write, the media + html routes, and
the upload path (writeWorkspaceUpload fence input).
- Editor load failures and file-tree level errors show the localized
reason plus a one-click global off (FenceErrorNotice) that flips the
pref, adopts it into the store, and auto-retries the failed operation
(reloadSeq / retryDir) — the raw wire text is never shown.
- GitView's two client-side pre-checks (hidden open action + no-op
defense) respect the pref; store threaded through TreePanel/FileTree.
- i18n: 4 new keys across all 21 dictionaries; tests: fence-off route
coverage, prefs parsing, schema default, editor toggles list;
docs: guide §8/§8.1 + plans/2026-08-29 design record.
A programmatic setDraft commit resets the controlled textarea's caret
(observed landing at the start of the value), so every later insert probed
the reset position and the stack drifted (A|B + C + D became |DACB).
placeComposerCaretAfterInsert puts the caret right after the inserted text
once the value commit lands - the index accounts for the left-side
separating space - and is guarded by a value-sync check plus a two-frame
retry, so a newer edit or a different composer wins untouched.
spliceInsert now reports the caret index together with the spliced draft
so the placement math cannot drift from the join logic. Nothing else moves:
the pure insertAtCaret surface is unchanged.
appendToDraft always spliced at the end of the draft. The conversation
service exposes no caret API, so probe the live composer textarea selection
from the DOM (value-sync guarded) and splice at that position with
whitespace-aware joins; an unknown or stale caret falls back to appending.
The explorer @-reference shares this path, so it inserts at the caret too.
The portaled position:fixed add-to-conversation button stays pinned to its
viewport anchor while the sidebar keeps every tab mounted (tab switches only
set display:none; collapsing the panel translates it off-screen), so the
popup could outlive its editor surface. Extract the popup into
useSelectionPopup and close it on any mousedown outside the button, Escape,
document hidden, window blur, and - the tab-switch/collapse path - when an
IntersectionObserver reports the editor surface off-viewport.
A reveal on a deep tree shifted the whole panel (tab bar included) out
of the viewport: the reveal used native scrollIntoView, which scrolls
every scrollable ancestor, and the fixed panel host
([data-dsh-panel-host], overflow hidden) is still a programmatically
scrollable container — the host scrolled instead of (or besides) the
tree, with no way to scroll back. Reported on DSH 0.1.2-alpha.1.
Two layers:
- FileTree: compute the centered, clamped target and scrollTo the tree
body only; scrollIntoView is never called.
- panel host: overflow hidden -> clip. A hidden box remains
programmatically scrollable; clip keeps the same viewport-edge
clipping (collapsed panels sliding off-screen) while making the host
impossible to scroll, killing the whole leak class (md-toc jumps,
future consumer scrollIntoView calls).
AGENTS.md (546 -> 63 lines) now carries only project-global development
rules: hard constraints, mount-smoke CI, DSH 0.1.2-alpha.1 adaptation
notes, npm release, a quick-rule list, and a docs/tests map. The
consumer plugin API reference lives solely in
docs/external-plugin-guide.md.
Reference sync:
- README/README_EN: collapse the dual 'full integration docs' links to
the guide; AGENTS.md stays linked as the repo rules
- shell-presets.ts / mount.e2e.ts comments: skinning contract pointer
moved from AGENTS.md §8 to guide §12
Dual-version compatibility: the plugin now works identically on
0.1.0-rc.8 ~ 0.1.1-rc.2 and 0.1.2-alpha.1 (the latter verified against a
source build of the dsh-v0.1.2-alpha.1 tag: mount smoke 14/14 + aggregate
double-mount; 0.1.1-rc.2 re-verified 14/14).
- MarkdownText: alpha.1 renamed the flat optional `codeLabels` prop to a
required nested `labels: { code: {copyLabel, copiedLabel}, footnotes }`;
the old prop alone crashes fence rendering with "reading 'code'". All
four render sites (mermaid / MarkdownHtml / TextEditor / SideChatView)
now go through src/client/markdown-labels.tsx (dual-shape object passed
under both prop names).
- e2e mount lanes speak both host dialects (tests/e2e/host-protocol.ts +
host.ts, unit-locked by tests/e2e-host-protocol.spec.ts):
- one-time-token browser auth: the launch URL grep extends past the
`/?token=` query (truncating at `/` loses the token), goto navigates
the token URL (browser cookie exchange), and stamped navigations
pre-seed the context cookie via addCookies — the exchange 303 drops
sibling query params;
- Remote gateway: dot endpoints first, slash endpoints on 404 fallback,
with args keyed by the controller parameter name (workspace/create ->
{request}, session/list -> {_request}; verified on a live host).
- package.json: drop the @deepseek-ai/dsh-client-runtime peer (package
removed upstream in alpha.1; zero references in the source — the chunk
externals entry stays and is resolution-tolerant).
- scripts/e2e-mount.sh: fix a latent set -u crash in the npx fallback
branch ($DSH_CMD followed by a full-width comma parsed as part of the
variable name).
- CI pin stays at @deepseek-ai/dsh@0.1.1-rc.2 until alpha.1 publishes to
npm (comment updated); README/AGENTS document the new contracts.
- Version lockstep bumped to 0.17.1.
* fix(cwd): resolve detached session cwd via persistence index, not host process cwd
sessionCwdOf fell back to process.cwd() when a session was not yet
attached and the client did not send a cwd — on Windows the host
process cwd is the DSH source root (dsh.cmd's pushd %DSH_ROOT%),
so every user-project path failed the workspace containment guard
with 'path "..." is outside workspace'. The bug surfaced when
clicking chat file links at first paint (session still hydrating)
or any flow that hit a sidebar route before the client's
fetchedCwd effect resolved.
Fix: thread ctx.get('sessionPersistence') as the fallback between
the client cwd and the host process cwd. The persistence index
holds every persisted session's header (including cwd), so a cold
(detached) session resolves its real project cwd instead of the
DSH source root. The host process cwd remains the final fallback
for deployments without persistence (tests / stripped-down hosts);
production always provides persistence.
sessionCwdOf is now async; all 8 call sites await it. The two tool
resolvers (registerTools / registerOpenTool) take an async
resolveCwd callback; the one execute path that used it
(terminal_create) is now async.
Tests: two new regression cases in smoke.spec.ts cover the cold
session via persistence path and the persistence-without-cwd
fallback. Existing stubs updated to the async resolveCwd signature.
* Address Copilot review: fix JSDoc accuracy and validate persistence cwd
- tools.ts / agent-opens.ts: correct the resolveCwd JSDoc — the resolver
does fall back to the host process cwd as a final tier (tests /
stripped-down hosts); the previous "never falls back" wording
contradicted the actual sessionCwdOf contract.
- index.ts: route the persistence meta.cwd through requireAbsolute,
matching the clientCwd path. A buggy/corrupt persistence layer that
stored a relative cwd would otherwise flow into the workspace guard,
be resolved against the host process cwd, and potentially recreate
the original "outside workspace" misclassification in a harder-to-
debug way.
- smoke.spec.ts: use resolvePath() for the cold-cwd fixture so it is
absolute on both POSIX and Windows; add a regression test for the
relative-cwd rejection path.
M4 of pinned terminals (v0.17.0):
- New PinnedRail component: compact strip above the tab bar, renders
pinned terminals from other sessions (pin icon + terminal icon + title
+ tooltip with kind/scope/cwd)
- Click navigates to the home session (ctx.sessions.open) and activates
the tab (or raises its float) via the pinnedJumpRef effect
- Right-click menu: Unpin / Close Terminal; middle-click close
- IconPinOutline16 added to icons.tsx
- Rail styles in sidebar.module.css (token-driven, no-drag opt-out)
- Sidebar: collectPinnedTabs on store snapshot + session list, railRevision
state forces re-render after reduceFor (which doesn't notify)
- Cross-session close: reduceFor(closeTab/closeFloatByTab) + ptyClose/
agentPtyClose; cross-session unpin: reduceFor(setTabPin null)
M3 of pinned terminals (v0.17.0):
- reconcileAgentTerminals skips removal of agent tabs with pin !== undefined
(the agent closed them / pty exited, but the user pinned them)
- The xterm reconnect-failure banner is the user-visible 'disconnected'
signal (design M3 convergence: no title suffix, no meta write)
- A later push reviving the same uuid is a no-op (id already exists)
- Unpinned agent tabs still remove as before; toAdd path unaffected
M2 of pinned terminals (v0.17.0):
- TabBar: onPinTab prop (optional, hidden when unset for legacy callers)
- Terminal tabs get a 'Pin ▸' submenu (workspace/global) or 'Unpin' row
- UI vs Agent label distinction via isAgentTabId
- Non-terminal tabs get no pin entry (5-item legacy menu preserved)
- WorkbenchActions.pinTab threaded through split-pane LeafView
- Sidebar.onPinTab wires store.reduce(setTabPin) with cwd snapshot
- 12 new i18n keys synced across all 19 language dictionaries
Opening the source-control panel could freeze the whole sidebar, and the
frozen layout restored itself after every reload, leaving no way out
(issue #369). Three unbounded layers compounded:
- `git status --porcelain -z --untracked-files=all` shipped every row of a
pathological untracked set into one response; the browser main thread
froze on response.json() plus the unvirtualized change list, and the 2s
poll replayed the freeze. status() now caps entries at 2000 and reports
`truncated` (the panel shows a notice, mirroring fs.read's truncation
semantics); worktree change counts inherit the bound for free.
- repoRoots() probed every visible child directory of a non-repo cwd
(e.g. ~) serially, once per git.* call, each with the full 30s command
budget. Discovery now uses a 3s probe timeout, caps probes at 200
directories, and shares one in-flight scan plus a 60s TTL cache per cwd.
- A hung layout persisted `panelOpen: true` with the git tab active, so a
reload remounted straight back into the freeze with no escape hatch.
Loading with `?dsh-sidebar-reset` now drops the persisted layout (and
the shared width) and starts from the default.
statusTruncated copy synced across all 19 shipped dictionaries.
Adapted to HEAD's unified editor+explorer tab: the reveal gesture now
opens the editor home tab (the files window) instead of the deleted
explorer type. PR's revealInExplorer implementation, produced-files
engine-data selector, and tests are retained verbatim.
Atlas Zeng