- Restore file-tree Add to chat and keyboard menu behavior (#1322).
- Preserve nested OpenAI cache hits beside zero-valued cache fields (#1327).
- Allow exact development renderer CORS preflights while retaining auth (#1335).
- Prevent stale transcripts from restoring explicitly cleared goals (#1339).
- Recover interrupted auto-dream attempts while preserving legacy stamps (#1349).
- Add Zed to the existing editor discovery and launch menu (#1346).
Refs #1322, #1327, #1335, #1339, #1349, #1346
Validation: local before/after reproductions, browser smoke, server 5537
and desktop 6086 tests passed; provider/chat/agent-flow/persistence checks
passed. Changed-line coverage is 98.23%. The full coverage gate remains
blocked by the unchanged IM slow-drip test; native signing/package
validation is incomplete.
Resolve optional click animation assets from the running helper bundle.
Avoid SwiftPM Bundle.module's fatal fallback to a build-machine path so
visible clicks can return successfully and keep the daemon alive.
Keep procedural feedback when optional frames are absent or unreadable.
Add an input-free resource probe and run it against relocated release and
staged helper apps. Cover standard Resources copying and reject probes
that load from an external build directory.
Validation: 524 XCTest and 15 Swift Testing cases passed, including eight
new resource regressions. Build fixtures, compiled sidecar smoke, Electron
checks, local packaging and final relocated-package probes passed.
Connecting Feishu meant creating a bot by hand on the open platform and
pasting an App ID and App Secret back. Feishu also exposes an RFC 8628
device-authorization flow, so the desktop can now render a QR code, and
confirming it in the app creates the bot and stores its credentials
directly. `adapters/feishu/registration.ts` implements that protocol
rather than importing `registerApp` from `@larksuiteoapi/node-sdk@1.73`:
the repository pins 1.60 for the chat client, and the SDK runs the whole
poll inside one un-cancellable promise where the desktop needs the
stateless begin/poll pair the DingTalk registration already uses. The
scan is create-only, so it can never rewrite the configuration of a bot
the user already runs, and it pre-fills exactly the scopes, events and
callbacks this adapter calls. International tenants finish on Lark's
domain, which is now persisted and honoured by the client.
WeCom, QQ and Slack join the same session model. WeCom and QQ bind by
scanning; Slack has no scan flow, so it uses an app manifest that
pre-fills the scopes and Socket Mode. All three run over long
connections, so no public callback URL is needed, and all three accept
private chats only — pairing authorizes one person, and answering in a
group would extend that authorization to everyone else in the room.
They are built on a new `adapters/common/chat-runtime.ts` instead of a
fourth copy of the loop the five existing adapters each carry. A platform
supplies a `ChatPort` — how to say something, how to open a streaming
reply, optionally how to send an image — and the runtime owns pairing,
command routing, session restore, permission bookkeeping and the
translation of the server's stream. The existing five are deliberately
left on their own copies; migrating them is a separate change with its
own regression surface.
Attachments are downloaded through a deferred loader that runs after the
pairing gate and inside the per-chat queue. Resolving them eagerly would
let an unpaired stranger make the adapter fetch bytes and write them
under ~/.claude/im-downloads — on Slack with the bot token attached —
and would let a slow attachment overtake a text message sent after it.
The sidecar launcher's per-adapter branches become one table. It is
declared above the mode dispatch on purpose: `runAdapters` is hoisted and
runs at module top level, so a table declared below it is still in its
temporal dead zone when the adapters mode reads it — which type checks,
lints and unit tests all miss, and only the compiled binary reveals.
Verified with the checks `check:impact` selects: adapters, server,
desktop, electron, policy, chat-contract, agent-flow, docs, native
(sidecar compile, packaging and an adapters-mode smoke against the real
binary) and coverage. The scan flows themselves are not verified against
live platforms — that needs real WeCom, QQ and Slack accounts and would
create real bots.
Claude-Session: https://claude.ai/code/session_01CCGoP316AK7wdQG3Ms6Uwq
Rebuilt against main so the branch carries the Computer Use work and no other
divergence. Three unrelated efforts had been sitting uncommitted in this
worktree and were swept into an earlier commit; they are preserved on
cu-worktree-full-backup and belong on their own branches — adapter control
credentials, Electron asar sealing, and the sidecar code-loading audit. Every
file outside Computer Use now matches main exactly.
The engine
A Swift helper drives apps through the accessibility tree, with coordinate
actuation for the Chromium and Electron apps whose tree is a bare window
frame. Ten primitives matching the shape Codex uses, so an app's guidance and
the model's habits transfer.
Coordinate actions resolve their target window once and refuse when none can
be named. The unbound event they used to fall back to is discarded by custom
renderers, so a minimized target produced a whole session of "Action
completed" with nothing behind it.
Input acceptance is established for typing and key presses as well as clicks:
each MCP call is seconds apart, so the keyboard cannot inherit the focus a
click established. The synthetic focus notification is gated on the target
not already being active — sent unconditionally it names window 0 at an app
that already owns a key window, and nine window-bound clicks were discarded
with the traffic lights fully lit.
State the model can trust
An off-screen target says so, and says which tools still reach it: element
actions need no on-screen geometry, so an app with a real tree can still be
driven from the Dock. A fully covered window is recovered once, then left
alone — burying it again is the user wanting their screen back. A repeated
capture is reported with the cause that actually applies rather than both,
because coverage is something we compute.
Signing
The helper is signed under a stable identity before electron-builder sees it,
and excluded from re-signing: macOS ties Accessibility and Screen Recording
grants to the signing identity, so rotating it drops both on every update.
Discoverability
The desktop slash menu falls back to a directory scan while a session's CLI
has not started, which is when the menu is first opened. Built-ins and
bundled skills live in the binary, so /computer-use was absent until after
the first message.
`Verify Windows installer execution` failed twice in a row on the v0.5.0
tag, both times at the same assertion:
Elevated default-mode reinstall without CLR expected process exit
code 20, received 22.
20 and 22 are different answers to "why did setup stop". 20 is legacy
recovery refusing to continue; 22 is "a matching process is running". The
stage breaks the CLR on purpose so the installer cannot run PowerShell,
which is exactly when CcHahaFindInstallProcess degrades from resolving
paths to matching bare image names -- `Claude Code Haha.exe`, the three
`claude-sidecar*` names, `OpenConsole.exe`, `winpty-agent.exe`, `rg.exe`.
Any process on the runner carrying one of those names answers for the
stage, whoever started it.
The stage before it expects 22, so a stray match there is indistinguishable
from a pass; the stage after it expects 20, so the same stray match is a
failure. That asymmetry is why this reads as "one flaky assertion" rather
than "the whole no-CLR group is unguarded".
This does not fix the installer, because the installer is not wrong:
refusing to delete user data when it cannot confirm what is running is the
intended fail-closed behaviour, and neither installer.nsh nor this script
changed between v0.4.11 (green) and v0.5.0 (red). What changed is what was
running on the runner. So the script now controls that instead of assuming
it:
- Both stages that expect 20 first clear any process matching the same
name list. `WaitForExit` only covers the PIDs this script started; the
fallback matches names, so it also sees leftovers from earlier steps of
this job -- the compiled-sidecar smoke starts 20 sidecars -- and any
child a probe spawned.
- Clearing warns instead of throwing when something survives. A survivor
is runner-owned and out of reach, and failing there would replace the
stage's own failure with a less informative one.
- A mismatched exit code now prints every matching process with PID and
path, and a baseline is printed before the first install. Between those
two, a future failure says whether the runner was dirty or the installer
regressed, which this run had no way to answer.
The name list is duplicated from installer.nsh by necessity -- NSIS
compile-time state is not readable from PowerShell -- and is commented on
both sides to be kept in sync.
Not verified locally: this needs Windows and an ephemeral runner (the
script refuses to run unless CI=true, since it mutates installer registry
state). Reviewed for Windows PowerShell 5.1, which is what the workflow
invokes: no pwsh-only syntax, and the file is kept pure ASCII as it was,
so 5.1 cannot mis-decode it.
Importing an animated pet required a file that was exactly 1536x2288, laid
out as 88 seamless cells, with the last two rows holding sixteen distinct
gaze angles. No image model emits that. Whatever a user got back from Jimeng
or ChatGPT was some fixed size like 1024x1536, so the path ended at "the
animation atlas must be exactly 1536x2288 pixels" every time. The third card
was worse: "AI-generate full animation" was hardcoded `disabled`, so the one
entry point named after what people actually wanted to do was dead.
The fix was already in the tree. `scripts/assemble-generated-pet-atlas.py`
landed in the same commit as the four built-in pets, which is to say the
built-ins were produced this way — it takes an action sheet at any size,
slices it on an 8x9 grid, fits each cell to 192x208, mirrors the run row to
make run-left, and reuses rows to reach eleven. That capability was never
wired to anything a user could reach.
`petAtlasNormalize.ts` reimplements it on a canvas in the renderer, so an
author draws nine rows and the app derives the rest. Verified against the
reference assembler by reversing dada-code's atlas into a nine-row sheet and
re-normalizing it: every difference lands on semi-transparent antialiased
edges (2314 pixels, max channel delta 14/255) and opaque regions are
identical. That residue is canvas premultiplied-alpha round-tripping, not a
slicing bug.
Three contract details worth stating. Row frame counts are now derived from
`PET_ANIMATION_DEFINITIONS` rather than typed out a fourth time; they come
out equal to the assembler's `(6,8,8,4,5,8,6,6,6,8,8)`. A sheet already at
1536x2288 passes through byte-for-byte instead of being resliced, because
resampling finished artwork buys nothing. And since the validator never
inspects the alpha channel, a flattened white background used to import
happily and render as a rectangle on the desktop — the renderer now rejects
sheets whose atlas is under 5% transparent (the built-ins sit near 78%) with
a message that names the actual problem.
The copy stops describing the implementation. "Animate one image" and
"Import professional animation atlas / exact 1536x2288 v2 PNG" become "use a
picture you already have" and "I already have an action sheet"; the dead AI
card becomes a three-step walkthrough carrying a copyable prompt, a labelled
8x9 reference grid that can be saved locally, and the checks that catch the
common failures. Reference images are generated by a script rather than hand-
placed, in both languages. All five locales move together.
Caught while reviewing the real dialog in Electron: after finishing the
walkthrough the form heading fell through to the atlas branch and announced
"I already have an action sheet" to someone who had just been walked through
drawing one. Covered by a test now.
Not done: docs/images/desktop_ui/15_pet_create_methods.png still shows the
old dialog and needs a fresh capture from a running app to match the styling
of the shots around it.
The Windows x64 build job went red on `Verify compiled Windows sidecar
startup`, asserting that a loopback request without
`CC_HAHA_LOCAL_ACCESS_TOKEN` returns 403 while it returned 200. Nothing
about x64 is involved — that step carries
`if: matrix.smoke_platform == 'windows' && matrix.arch == 'x64'`, so it is
the only job in the whole matrix that runs the smoke at all. Any regression
in this area can surface nowhere else.
The 200 is correct. `7d2a8a3cd keep loopback trusted without the desktop
process token` deliberately made the token additive again: gating every
local request behind it turned the Grok OAuth success page, `/preview-fs`
links and plain `curl` into 401s, because none of that traffic can ever
carry the token. Loopback is trusted on its own; the token is demanded only
on the `/api/h5-access` control plane, where another program on the same
box must not be able to publish the user's sessions to the network. The
assertion, written before that change, was still guarding the path that had
been intentionally opened.
So the probe moves to the boundary that is actually enforced, and gains a
positive assertion — loopback without a token must be 200 — so the additive
model is pinned down rather than merely no longer contradicted. Reverting to
the pre-`7d2a8a3cd` behaviour now fails the smoke instead of passing it.
Three copies of the stale assertion existed; all three are updated. Only the
compiled-sidecar smoke runs in CI, but `local-index-benchmark.ts` and its
corpus test were already failing the same way for anyone running them
locally. The benchmark also cancels the probe response bodies now: an unread
body holds its connection open, and that would land in the event-loop delay
and RSS samples taken immediately after.
Verified with the CI parameters — `bun run build:sidecars` then
`CC_HAHA_COMPILED_SIDECAR_SMOKE_STARTS=20 bun run test:compiled-sidecar-smoke`,
8/8 — and by running the benchmark directly, which now reports
`loopbackAuth` as 200/403/403/200 with validation intact.
The resolveElectronExecutable case asserted against the real
node_modules/electron/dist, which bun never downloads without the
electron postinstall, so it failed on macOS/Linux dev machines.
Point the tests at a self-built fixture desktop root instead and
extend coverage to the darwin/linux candidates and the missing-
executable error path.
Follow-up to #1096.
`new URL('..', import.meta.url).pathname` yields a leading-slash path like
/E:/... on Windows, so Bun.spawn failed with ENOENT. Resolve desktopRoot via
fileURLToPath, start Vite through its Node API instead of spawning `bun run
dev`, and resolve the Electron binary directly per platform.
Recovered from the discarded wip/capabilities snapshot (cb1208c9) as a
standalone fix.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Avoid the bunx launcher for local Electron packaging after it can be terminated before Electron Builder starts. Use the installed Electron Builder CLI through Node for the macOS package script and desktop package shortcuts.
Tested: SKIP_INSTALL=1 SKIP_PACKAGE_SMOKE=1 bash ./scripts/build-macos-arm64.sh
Tested: bun run test:package-smoke --platform macos --package-kind release --artifacts-dir desktop/build-artifacts/macos-arm64
Tested: bash -n desktop/scripts/build-macos-arm64.sh
Tested: git diff --check
Not-tested: full bun run verify was not run because this is a narrow local packaging launcher fix.
Confidence: high
Scope-risk: narrow
- Rewrite release-notes/v0.4.0.md to match the historical style (no emoji,
Highlights/Fixes/Notes), keep it user-facing only (drop release-process
notes), and correct Linux (supported since the Tauri builds, not new).
- README.md / README.en.md: replace Tauri references with Electron, list
macOS / Windows / Linux, and point first-launch approval to the guide.
- Rewrite docs/desktop/04-installation.md for Electron: Electron asset
names, Linux section, and the unsigned-macOS flow (clear the DMG quarantine
before double-clicking; drop the WebView2/right-click-open leftovers).
- install-macos-unsigned.sh: keep the same-folder DMG flow, no online download.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bump the Electron desktop package version to 0.4.0, publish a macOS unsigned install helper, and let the release workflow continue when Developer ID signing is not configured.
Tested: bash -n desktop/scripts/install-macos-unsigned.sh
Tested: bun test scripts/pr/release-workflow.test.ts
Tested: bun run scripts/release.ts 0.4.0 --dry
Tested: git diff --check
Tested: bun run check:docs
Tested: bun run check:policy
Confidence: high
Scope-risk: moderate
- mac.notarize=true + hardenedRuntime + entitlements so a signed CI release
actually notarizes (gatekeeper smoke + Squirrel.Mac auto-update need it)
- entitlements grant disable-library-validation for the Bun sidecar/node-pty
- local unsigned build passes -c.mac.notarize=false so electron:package still
works without an Apple account
- release signing-preflight now hard-requires only the Apple secrets; Windows
cert is optional (unsigned NSIS still auto-updates, just SmartScreen warning)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Introduce the Electron desktop shell alongside the existing React renderer and local Bun server boundary. The migration keeps the DesktopHost contract explicit across Tauri, Electron, and browser runtimes while adding Electron main/preload services for dialogs, shell, notifications, updates, tray/window lifecycle, terminal, preview WebContentsView, app mode, and release/package validation.
The commit also carries the latest local main desktop command updates, including agent slash entries and hidden-by-default markdown thinking details, so the packaged Electron build matches the current main UX surface.
Constraint: React renderer, local Bun server, REST/WebSocket, and sidecar boundaries must remain reusable during the migration
Constraint: macOS dev packages are ad-hoc signed and cannot prove Developer ID notarization or Gatekeeper release launch
Rejected: Browser-only smoke validation | it cannot exercise native dialogs, keychain prompts, notification behavior, or packaged app startup
Confidence: medium
Scope-risk: broad
Directive: Do not remove Tauri host support until signed Electron release artifacts pass native OS smoke on macOS, Windows, and Linux
Tested: bun run check:desktop
Tested: cd desktop && bun run check:electron
Tested: CSC_IDENTITY_AUTO_DISCOVERY=false bun run electron:package:dir
Tested: bun run test:package-smoke --platform macos --package-kind dir --artifacts-dir desktop/build-artifacts/electron
Tested: Computer Use read packaged Electron app window at desktop/build-artifacts/electron/mac-arm64/Claude Code Haha.app
Not-tested: Developer ID signed/notarized Gatekeeper launch
Not-tested: Real OS notification click-to-session action
Not-tested: Windows and Linux packaged app smoke on real hosts
Prepare the desktop release metadata and concise release notes while keeping
tagging and release publishing for a later step. The staged local build-script
updates keep desktop commands on checked-in local toolchain paths and avoid
rewriting preview-agent output when the built content is unchanged.
The persistence-upgrade gate now runs the focused desktop Vitest migration
suite in non-watch mode, matching the broader desktop quality lane and avoiding
pre-push termination during release preparation.
Constraint: Release publishing is intentionally deferred per request
Constraint: Desktop release metadata must keep package, Tauri config, Cargo metadata, and Cargo.lock aligned
Confidence: medium
Scope-risk: moderate
Directive: Do not tag v0.3.2 until release dry-run and final release verification are rerun on the release candidate
Tested: bun run scripts/release.ts 0.3.2 --dry
Tested: cd desktop && bun run lint
Tested: bun run check:persistence-upgrade
Tested: git diff --check
Not-tested: Full bun run verify
Desktop users need to see automatic context compaction as a transient transcript state instead of a stray stdout bubble or large card. The UI now collapses prior visible content into a compact timeline divider, shows the compacting phase, then updates the same divider with the completed summary affordance.
Constraint: CLI compaction emits both status and synthetic transcript artifacts that must be normalized before rendering
Rejected: Keep the existing small pill marker | it did not show the in-progress state and looked like chat content
Confidence: high
Scope-risk: moderate
Directive: Do not render local compact stdout as a user-visible chat message
Tested: cd desktop && bun run test -- src/components/chat/MessageList.test.tsx src/stores/chatStore.test.ts
Tested: bun test src/server/__tests__/ws-memory-events.test.ts
Tested: bun run check:desktop
Not-tested: full root bun run verify after the final UI iteration
Bring back the narrow, previously validated sidebar recovery and Windows x64 sidecar compatibility fixes that were present on worktree branches but absent from the canonical local main checkout. The broader worktree branch is intentionally not merged because it would replay unrelated history and risk dropping current main work.
Constraint: The local main worktree already contains unrelated Settings and portable storage edits, so only the recovery hunks are staged.
Rejected: Merge 1506086927/main wholesale | its history includes broad rollback/replay commits that would remove current main changes.
Confidence: high
Scope-risk: narrow
Directive: Recover future worktree fixes by cherry-picking or staging narrow hunks; do not merge drifted worktree branches wholesale.
Tested: cd desktop && bun run test -- --run src/components/layout/Sidebar.test.tsx scripts/build-sidecars.test.ts
Not-tested: Full bun run verify due unrelated dirty Settings/portable-storage worktree changes.
The notification bridge fix was correct, but the local macOS build script still left Tauri's raw app and DMG outputs available with the executable's ad-hoc signing identifier. Opening that raw artifact made macOS treat notification authorization as a different app identity, so the UI could fall back to the unsupported environment state even though the canonical output worked.
Normalize the Tauri-produced app bundle in place before copying it to the canonical output, then replace the Tauri-generated DMG with the rebuilt canonical DMG. This keeps every path printed by the build script on the same bundle identity and prevents testers from accidentally opening the unstable raw artifact.
Constraint: Local test builds use ad-hoc signing, but macOS notification authorization still depends on stable app identity.
Rejected: Only tell testers to open canonical output | the script still printed and preserved misleading usable-looking artifacts.
Confidence: high
Scope-risk: narrow
Directive: Keep Tauri raw app and canonical app signing identifiers aligned before reporting build artifacts as runnable.
Tested: bash -n desktop/scripts/build-macos-arm64.sh
Tested: SKIP_INSTALL=1 ./desktop/scripts/build-macos-arm64.sh
Tested: codesign -dv --verbose=4 shows com.claude-code-haha.desktop for both target bundle app and canonical app
Tested: hdiutil verify target DMG and canonical DMG
Tested: Computer Use opened target bundle app and verified Settings -> General shows notifications as authorized.
Not-tested: Fresh first-time notification prompt after resetting macOS notification permissions for this bundle id.
Desktop sessions can run against different saved providers at the same time, so the runtime selection must remain scoped to the session from WebSocket startup through the CLI request path. This adds parallel-session regression coverage, a browser-driven fake-provider E2E, and refreshes idle connected sessions after provider saves so stale provider edits are reapplied without disturbing active turns.
Constraint: Desktop provider credentials must stay session-scoped and must not leak across concurrent CLI child processes
Rejected: Rely on provider test connectivity alone | it does not exercise the desktop WebSocket, session runtime, or CLI request path
Confidence: high
Scope-risk: moderate
Directive: Do not change runtime selection persistence or provider env injection without rerunning the parallel provider browser E2E
Tested: cd desktop && bun run test src/stores/providerStore.test.ts src/stores/chatStore.test.ts
Tested: CLAUDE_CONFIG_DIR=/Users/nanmi/.claude bun test src/server/__tests__/conversations.test.ts --timeout 30000
Tested: desktop/scripts/e2e-parallel-providers-agent-browser.sh
Tested: bun run quality:smoke --provider-model custom:main:custom-main --provider-model minimax:main:minimax-main
Tested: cd desktop && bun run lint
Tested: git diff --check
Not-tested: bun run verify passes fully; coverage lane still fails on the existing agent-utils ratchet below baseline
The desktop test package was able to reuse stale Tauri target output even after the frontend and sidecar were rebuilt, which made UI changes look missing after a build. The local macOS build script now clears that cache by default while retaining an opt-in incremental path.
The project selector also no longer repeats the active branch because branch and worktree are now first-class launch controls beside it.
Constraint: Local macOS test packages must prioritize freshness over incremental build speed
Rejected: Keep branch text in the project chip | duplicated the dedicated branch selector and made the launch row harder to scan
Confidence: high
Scope-risk: narrow
Directive: Keep branch selection behavior in RepositoryLaunchControls; DirectoryPicker should only identify the project
Tested: cd desktop && bun run test -- DirectoryPicker.test.tsx
Tested: cd desktop && bun run lint
Tested: SKIP_INSTALL=1 ./desktop/scripts/build-macos-arm64.sh
Tested: Opened built app and confirmed sidecar /health returned ok
Provider model context windows now feed the composer-level context indicator, so users can see current usage without opening slash panels. The indicator refreshes when the session runtime model changes, keeps a stable loading placeholder, and marks estimate fallback data during reconnect or compaction windows.
Constraint: The desktop UI must use the real session inspection API and session-scoped runtime selection instead of mocked or static context values.
Rejected: Keep context details only in /context or /status | the user explicitly wanted an always-visible composer indicator.
Rejected: Reuse stale context after model switches | different provider windows make identical percentages misleading.
Confidence: high
Scope-risk: moderate
Directive: Do not remove the runtimeSelectionKey refresh path without real multi-provider browser verification.
Tested: cd desktop && bun run test -- pages.test.tsx
Tested: bun run check:desktop
Tested: KEEP_ARTIFACTS=1 desktop/scripts/e2e-context-usage-live-agent-browser.sh
Not-tested: Live browser E2E with every configured provider in the committed script; separate agent-browser sweep verified four providers locally.
DeepSeek-compatible Anthropic endpoints reject requests that combine an explicit disabled-thinking block with an effort value. The desktop UI can create that shape when users disable thinking and keep medium effort selected, so the request builder now omits effort whenever it will send explicit disabled thinking.
Constraint: DeepSeek-compatible Anthropic endpoints reject disabled thinking with effort
Rejected: Only patch Desktop provider settings | hidden title and side request paths still share the request builder
Confidence: high
Scope-risk: narrow
Directive: Keep explicit disabled-thinking providers from sending effort unless the upstream contract changes
Tested: UI repro on pre-fix DeepSeek provider path reproduced the 400
Tested: desktop/scripts/e2e-deepseek-thinking-ui-agent-browser.sh
Tested: bun test src/utils/__tests__/thinking.test.ts src/server/__tests__/provider-presets.test.ts src/server/__tests__/title-service.test.ts
Tested: git diff --check
Not-tested: Real DeepSeek API key live call
The chat surface now relies on the current-turn changes card for file rollback, so the older per-message hover rewind affordance was removed to avoid two competing rollback models.
Constraint: Current-turn undo still depends on the existing checkpoint rewind API.
Rejected: Keep both rewind entry points | duplicate rollback affordances make the product harder to explain and test.
Confidence: high
Scope-risk: moderate
Directive: Prefer adding rollback behavior through the current-turn change card instead of restoring per-message hover rewind.
Tested: bun test src/server/__tests__/sessions.test.ts
Tested: cd desktop && bun run test
Tested: cd desktop && bun run lint
Tested: cd desktop && bun run build
Not-tested: Browser E2E scripts were updated but not executed in this commit.
Desktop rewind previously used the visible user-message index as the primary selector. That can drift from the persisted active chain when hidden or non-rendered user messages exist, so the API now prefers a stable user message id and checks the selected prompt text before mutating transcript or files.
Constraint: Desktop UI can hide transcript entries that still exist in the persisted session chain
Rejected: Keep index-only rewind | can target an earlier user message after UI/server chain drift
Confidence: high
Scope-risk: moderate
Directive: Do not remove targetUserMessageId or expectedContent guards without reproducing shifted visible-index sessions
Tested: bun test src/server/__tests__/sessions.test.ts --timeout 20000
Tested: cd desktop && bun run test MessageList.test.tsx chatStore.test.ts -- --run
Tested: desktop/scripts/e2e-rewind-agent-browser.sh
Tested: desktop/scripts/e2e-rewind-complex-agent-browser.sh
Not-tested: Full desktop Vitest suite still has unrelated locale-sensitive failures
The local Apple Silicon packaging script produced app bundles that could
fail strict bundle validation because the outer .app had no sealed
resources. The fix shallow-signs only the copied canonical app bundle and
checks that the claude-sidecar code-signature hash stays unchanged.
Constraint: GitHub Actions release packaging must remain on the existing tauri-action path
Constraint: macOS Keychain ACLs are sensitive to sidecar code-signature hashes
Rejected: Set macOS signingIdentity in release CI | Tauri re-signs nested sidecars and can change Keychain caller identity
Confidence: high
Scope-risk: narrow
Directive: Do not deep-sign claude-sidecar without proving existing Keychain ACLs still work after upgrade
Tested: bash -n desktop/scripts/build-macos-arm64.sh
Tested: SKIP_INSTALL=1 desktop/scripts/build-macos-arm64.sh
Tested: codesign --verify --deep --strict --verbose=2 desktop/build-artifacts/macos-arm64/Claude\ Code\ Haha.app
Tested: open -n desktop/build-artifacts/macos-arm64/Claude\ Code\ Haha.app and /health returned ok
Not-tested: GitHub Actions release workflow, intentionally unchanged
Desktop sessions can switch provider and model while a CLI subprocess is already alive, so the server now serializes runtime restarts and marks provider-managed launches to prevent stale settings env from overriding the selected provider. Provider settings also write API key env consistently and clear stale managed keys before syncing.
This includes the related desktop/docs brand asset refresh and keeps the desktop locale default in Chinese, with tests updated to match the current provider semantics.
Constraint: Session-scoped model selection must win over cc-haha/settings.json and inherited ANTHROPIC_* values.
Rejected: Store the selected model as a global provider activeModel | chat runtime selection is per session.
Confidence: high
Scope-risk: moderate
Directive: Do not remove CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST without validating Desktop provider switching against stale settings env.
Tested: bun test src/server/__tests__/conversation-service.test.ts src/server/__tests__/conversations.test.ts src/server/__tests__/providers.test.ts src/server/__tests__/providers-real.test.ts
Tested: cd desktop && bun run test src/stores/settingsStore.test.ts
Tested: cd desktop && bun run lint
Tested: git diff --check
Not-tested: Full desktop production package/signing.
Desktop plugin details now route into the shared Skills, Agents, and MCP management surfaces instead of maintaining separate read-only drilldowns. This also extends the desktop/server skill aggregation so plugin-provided skills appear in the shared list, groups MCP entries by source, and preserves detail-view back navigation based on where the user entered the page.
The implementation keeps plugin detail as the high-level capability hub while pushing real inspection into the existing management pages. Disabled plugins no longer expose false navigation paths into shared views, and the agent-browser regression script was expanded to exercise the new end-to-end flows.
Constraint: Shared Agents data only includes enabled plugin agents, so disabled plugins cannot deep-link into agent detail
Rejected: Keep duplicating full Skills/Agents/MCP detail inside Plugin detail | creates divergent UI flows and stale data paths
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: If a detail view can be opened from multiple entry points, keep the return target in store state rather than hardcoding a single back destination
Tested: desktop vitest for plugins/skills/agents/mcp; desktop tsc --noEmit; desktop vite build; server skills API test; agent-browser web regression on plugin->skill/mcp and plugin->agent back navigation
Not-tested: packaged desktop app regression after rebuilding the Tauri bundle
The desktop app already had solid session streaming, permission, and tool-rendering flows, but extension setup still forced users into manual forms or external shell work. This change adds an Install Center in Settings that reuses the session chat pipeline for natural-language installs, adds installer-specific guidance for plugin and skill URLs, and includes an agent-browser E2E script for real UI validation.
Constraint: Must reuse the existing session/chat execution path instead of introducing a second install runtime
Constraint: Plugin installs need real CLI commands while skill installs may come from published install commands on third-party pages
Rejected: Separate terminal-only install surface | duplicates session UX and weakens permission/tool visibility
Rejected: Pure form-based installer expansion | too much friction for plugin, MCP, and skill onboarding
Confidence: medium
Scope-risk: moderate
Reversibility: clean
Directive: Keep installer prompts aligned with the actual CLI install surfaces; do not let the installer fall back to slash-command syntax inside Bash
Tested: cd desktop && bun run lint
Tested: Real UI automation via agent-browser for Telegram plugin install flow through Settings > Install, verified Plugins page shows telegram enabled
Tested: Real UI automation via agent-browser for ui-ux-pro-max skill install flow through Settings > Install, verified ~/.claude/skills/ui-ux-pro-max and Skills page visibility
Not-tested: Full e2e-install-center-agent-browser.sh script as a single uninterrupted green run after the latest stability tweaks
The desktop client already had a conversation-level rewind UI concept on the
CLI side, but the web/desktop surface lacked the protocol, session trimming,
and file checkpoint restore path needed to make rewind trustworthy. This change
adds a desktop-specific rewind API, wires the message-level UI affordance and
confirmation modal, enables SDK file checkpointing for desktop sessions, and
covers the restore path with service tests plus a real agent-browser workflow.
Constraint: Desktop sessions run the CLI in SDK/print mode, so file checkpointing had to be enabled explicitly for that path
Constraint: main branch is checked out in a separate worktree, so merge-back must happen from the primary worktree after commit
Rejected: UI-only rewind that only trims local state | would leave persisted transcript and disk state inconsistent after refresh
Rejected: Reuse getLastSessionLog as the sole snapshot source | active rewind must read file-history metadata directly from the session file
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep desktop rewind keyed to persisted user-message order unless the UI model starts carrying stable transcript UUIDs end-to-end
Tested: bun test src/server/__tests__/conversation-service.test.ts src/server/__tests__/sessions.test.ts; desktop MessageList vitest; desktop tsc no-emit; live agent-browser E2E on isolated ports with file edit then rewind
Not-tested: Browser E2E matrix for multi-file and second-edit scenarios is still covered at service-test level rather than full UI level
Windows releases were shipping both NSIS exe and MSI bundles, which made the
release page harder to understand and increased the chance that users would
pick the wrong asset. Restricting Windows output to MSI narrows the installer
surface while the release asset naming now spells out platform, architecture,
and bundle type directly in every published filename.
Constraint: Remote desktop packaging is triggered by GitHub Actions releases, not local uploads
Constraint: Updater metadata must keep latest.json stable for existing clients
Rejected: Keep publishing both NSIS and MSI with better docs | still leaves the higher-risk installer on the release page
Rejected: Rename assets manually after upload | easy to drift from workflow output and updater metadata
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: If Windows exe bundles are reintroduced later, reassess SmartScreen and AV impact before publishing them again
Tested: YAML parse for both desktop workflows; git diff --check on modified files; manual review of PowerShell changes
Not-tested: End-to-end GitHub Actions release run; PowerShell parser validation on a Windows host
Concurrent writes to the same settings file could interleave and leave user
preferences in a partially updated state, and a restarted conversation session
could still be torn down by an older process exit callback. This serializes
per-file settings writes, hardens temp-file handling, adds regression coverage
for both service behaviors, and drops an unused sidecar external from the
build exclude list.
Constraint: Settings writes must remain atomic while allowing multiple service entry points to update the same JSON file
Rejected: Keep fire-and-forget writes with unique temp names only | still allows stale reads and last-writer races between callers
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Preserve the per-file write lock when adding new settings mutation paths so concurrent writes keep one serialization point
Tested: bun test src/server/__tests__/settings.test.ts src/server/__tests__/conversations.test.ts (new settings/conversation cases passed)
Not-tested: Root lint script is unavailable; Models API tests in src/server/__tests__/settings.test.ts still fail in this checkout with changed default model/effort expectations
程序员阿江(Relakkes)