* Refactor agent chat operations to resolve chats uniformly
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Refactor chat ops to use concrete chat URIs
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: tolerate a session URI at the chat-context chokepoint (fix CI)
The chat-addressing refactor removed the agents' tolerance for addressing the
default chat by its session URI (the AHP convention: default chat URI == session
URI). This broke integration tests that address the default chat by the session
URI (Malformed AHP chat URI) and restore via the mock agent (0 restored turns).
- Claude/Copilot `_getChatContext`: accept either a chat channel URI or a bare
session URI, normalizing to the default-chat URI in the single resolution
chokepoint (not re-derived per operational method). Codex already tolerant.
- MockAgent.getSessionMessages: normalize a default-chat channel URI back to the
session URI, mirroring the real agents.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: remove the dead onDidEndChat membership channel
onDidEndChat was declared by Claude and Copilot but never fired: a completed
subagent chat stays live and is removed only on session teardown (via a direct
removeChat), so subagent_completed intentionally has no end event. Remove the
unused channel end-to-end — the agents' emitters, the optional
IAgent.onDidEndChat member, the orchestrator subscription + _onChatEnded handler,
and the synthetic test. onDidSpawnChat (which is fired) remains as a spawn-only
membership channel.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: rename IResolvedAgentChat.session to chatSession
Address CCR feedback: `session` overwhelmingly means the owning session URI in
this codebase, so `IResolvedAgentChat.session` (the resolved chat session
instance) was easy to misread. Rename to `chatSession` and update the two call
sites in claudeAgent.ts / copilotAgent.ts.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: quick chats — workspace-less single-chat sessions
Adds quick chats to the Agents window: lightweight chats not scoped to a
workspace, backed by an agent-host session. The host infers workspace-less
from an absent workingDirectory (forks excluded) and assigns a stable scratch
dir; the workspaceless tag rides the generic _meta bag. Quick chats are
single-chat, use the normal session presentation (Done hidden), render in an
always-visible in-list "Chats" section, and persist across reloads.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: New from a quick chat opens another quick chat (Round 23)
The primary "New" action gated quick-chat routing on `isCreated &&
isQuickChat`, but a quick-chat draft is Untitled (isCreated=false), so it
fell through to the workspace composer seeded with a throwaway scratch dir
(no session-type picker, "No models available").
Route on `isQuickChat` alone so a quick chat — draft or committed — opens
another quick chat mirroring its harness. Extract the routing into a pure,
side-effect-free `openNewChatOrQuickChat` helper so it is unit-testable
(chat.contribution.ts is not test-importable). Supersedes Round 14(2) and
updates Round 22(3); the Round 14(2) discard branch and Round 17 picker
re-parent are kept as internal defense.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: quick-chat list & layout polish (Round 22 items 1-2)
- Suppress the redundant per-row chat icon when a quick chat is rendered
under the always-visible "Chats" section (the section header already
carries a chat icon); keep it in Pinned/custom/date groups where the
chat identity is useful.
- Disable the "Toggle Side Panel" command for quick chats via
precondition IsQuickChatSessionContext.negate(), since a quick chat has
no side pane (the empty aux bar is hidden and the chat is full-width).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: persist empty peer-chat catalog sentinel to avoid re-running legacy migration
When a session has no legacy peer chats, write an empty catalog so
_readPersistedPeerChatCatalog returns [] on subsequent restores and
_migrateLegacyPeerChats never re-runs.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: Cmd+N always creates a new session; New Quick Chat gets Cmd+K Cmd+N (Round 24)
Per user feedback, Cmd+N must always open a NEW SESSION — never a quick
chat. Drop the context-aware quick-chat routing from
NewChatInSessionsWindowAction (rename its title "New Chat" -> "New Session",
keep the id) so it unconditionally calls openNewSession from the active
session; the helper is renamed openNewChatOrQuickChat -> openNewSessionFromActive.
Quick chats are created only via the Chats-section "+" (NewQuickChatAction),
which now has a default Cmd+K Cmd+N chord. The peer-chat "+" (Cmd+T) is
unaffected. Supersedes the Round 22(3)/23 mirror routing; the Round 14(2)
discard branch and Round 17 picker re-parent are untouched.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: Cmd+N never inherits a quick chat's folder into the workspace composer (Round 25)
New from a quick chat must always land on the clean New Session composer with
a visible session-type picker. Gate openNewSessionFromActive's folder
inheritance on isQuickChat so a quick chat never carries a (possibly leaked
scratch) workspace URI into openNewSession.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: re-seed workspace draft when composer swaps out of quick-chat mode (Round 25b)
Cmd+N from a quick chat reuses the new-session composer and only
_activate(undefined), leaving it session-less. The session-type picker hides
itself when it has no folder types (no active session), so no picker showed.
Re-run the constructor's workspace-draft seed from an autorun when the composer
transitions out of quick-chat mode with no active session, matching a freshly
opened new-session composer (folder + visible picker).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: quick-chat untitled title falls back to "New Chat" via shared helper (Round 26)
An untitled quick chat's titlebar showed "New Session" because the empty-title
fallback was hardcoded and not quick-chat aware. Add getUntitledSessionTitle(isQuickChat)
to the common layer and route all 5 fallback sites (titlebar, session header x2,
list hover, sessions picker) through it.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: track isQuickChat in titlebar re-render autorun (Round 27)
The SessionsTitleBarWidget re-render autorun read the active session's
title and workspace but not isQuickChat, which _render() consumes for the
untitled title fallback. Track it as a reactive dependency for
forward-safety and consistency with other reactive render sites.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: move workspace-less marker ownership to the AH service
Each agent used to persist and re-emit its own workspace-less (quick chat)
marker (copilot.workspaceless / claude.workspaceless) in the shared session
database, and agents that persist nothing (Codex) lost the marker on restart.
Make the AH service the single owner: AgentService persists a single
agentHost.workspaceless key at create/materialize (from the value it already
infers in _buildInitialSummary) and overlays _meta.workspaceless onto every
agent's summary in listSessions. Agents no longer write or namespace the
marker; Copilot reads the shared key for its resume system prompt, and the
now-dead workspace-less plumbing is removed from the Claude session. This
fixes restored quick chats for every agent (including Codex) with no
per-agent code.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions/agentHost: address review feedback (naming + inline)
- Rename the workspace-less launch-plan flag from isQuickChat to
workspaceless in CopilotSessionLaunchPlan and IAgentHostPromptContext (and
the disposeSession local) so the flag matches the workspaceless marker it
flows from throughout the AH layer. Feature-descriptive names
(COPILOT_AGENT_HOST_QUICK_CHAT_INSTRUCTIONS, _quickChatScratchDir) are kept.
- Inline openNewSessionFromActive back into NewChatInSessionsWindowAction.run
and remove the single-caller seam module + its test.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: make the AH service the sole owner of the workspaceless marker
Following the earlier ownership move, the agents still read + re-emitted
_meta.workspaceless in their metadata projections, which was redundant on the
listSessions path (AgentService overlays it centrally) and only load-bearing
on the single-session restore path.
Centralize the restore overlay in AgentService.restoreSession (reads
agentHost.workspaceless in its existing batch metadata read and merges it into
the restored summary _meta), then drop the per-agent re-emit: remove it from
the Claude metadata store entirely (Claude has no runtime need) and from the
Copilot listSessions/getSessionMetadata projections. Copilot keeps reading the
AH key for its resume system prompt and scratch-dir cleanup. Codex is now
covered centrally with no Codex code.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: inline openQuickChatAndFocus into NewQuickChatAction
Single-caller helper folded into the action's run().
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: make ISessionsProvider.createQuickChat mandatory
Replace the optional createQuickChat with a mandatory method that throws when
the provider does not support quick chats; callers now gate solely on the
supportsQuickChats capability instead of probing for the method. Workspace-bound
providers (Copilot chat, local chat) get an explicit throwing implementation.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: use "workspace-less chat" terminology instead of "quick chat"
Rename the agent-host-internal quick-chat identifiers, prompt tags, and prose to
workspace-less: COPILOT_AGENT_HOST_QUICK_CHAT_INSTRUCTIONS ->
COPILOT_AGENT_HOST_WORKSPACELESS_INSTRUCTIONS, the <quick_chat> system-message
tag -> <workspaceless_chat>, and the scratch-dir helpers
(_quickChatScratchDir/_ensure*/_cleanup*/_withQuickChatScratch). The workbench
UI term "Quick Chat" is kept only where the agent host documents that mapping.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: address Copilot Code Review feedback on quick chats
- sessionsList: select the row chat icon from isQuickChatSession(), not from
workspace === undefined, so a workspace session with a transiently-undefined
workspace no longer briefly shows the chat icon.
- sessionContextKeys: correct the isQuickChat comment to reflect that the key
is sourced from the isQuickChat tag, never inferred from workspace absence.
- Agents window accessibility help: document the New Quick Chat command
(Cmd/Ctrl+K Cmd/Ctrl+N) and the Chats section plus button, and note that the
workspace picker does not apply and Toggle Side Panel is disabled for quick chats.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* wip on mcp auth through AH
* Preserve live MCP server state across customization re-syncs
The Agents window showed a connected GitHub MCP server flipping back to
'Starting' when navigating away from and back to a session. A client
re-subscribe re-published the session's customizations, and the
SessionCustomizationsChanged full-replace reset each MCP entry's state to
the 'Starting' default baked into makeMcpServerCustomization.
- Skip no-op customization re-syncs in SessionPluginController.sync so an
identical re-publish (e.g. navigating back) does no work.
- SessionPluginController now overlays live MCP runtime state/channel onto
every published customization via _projectForPublish, so a genuine
single-customization change no longer resets otherwise-unchanged MCP
servers. The overlay is driven by an ISettableObservable kept up to date
by the session from McpCustomizationController.
- McpCustomizationController._live is now an observable and exposes
runtimeStates as a derived; mutations are batched in transactions.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Persist agent-host MCP auth and improve auth-required prompts
- Key remembered MCP auth on a stable id (session authority + server name +
resource URL) instead of the unstable customization id, so grants survive
reloads and don't require re-auth.
- Record agent-host metadata (authority + host label) on allowed MCP servers
and surface agent-host servers in their own section of the Manage Trusted
MCP Servers picker instead of filtering them out.
- Make the auth-required chat prompt reactive: servers is now an observable so
servers whose auth requirement surfaces later join the existing prompt, and
the part marks itself used once hidden so later requirements re-prompt.
- Show an 'Authenticating <server>...' progress state while each server auths.
- Drop the never-serialized mcpAuthenticationRequired part from the serialized
response-part unions.
- Add unit tests for the stable-id helper, agentHost metadata persistence, and
query-service exposure.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Address CI failure and Copilot review on MCP auth prompt
- Don't emit an empty mcpAuthenticationRequired progress part (was adding a
stray part and breaking AgentHostChatContribution tool-progress tests).
- Guard the async auth filter with a run id so out-of-order completions can't
overwrite a newer server list.
- Group agent-host servers in the Manage Trusted MCP Servers picker by stable
authority (sorted by label) instead of label, which could collide.
- Scope the authenticate link to the #authenticate target and give it button
semantics (role=button, cleared href).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Adds ChatInteractivity (Full/ReadOnly/Hidden) to IChat and hides the composer
for non-interactive chats (gated via chatIsReadonly context key). Surfaces agent
host subagent (tool-origin) chats as read-only peer tabs with a lock icon,
persisted across restart. Adds an inline "Open Subagent" pill in the transcript
and a "Subagents" dropdown above the chat input to reveal them.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: remove Claude dead legacy migration + drop redundant session param
PART 1: remove the unreachable claude.chats legacy peer-chat migration
(listLegacyChats, _readPersistedChats, _META_CHATS and the two dead
fallback call sites). Copilot's copilot.chats migration and the
orchestrator's generic migration are unchanged.
PART 2A: drop the redundant session parameter from IAgentChats.createChat
and fork. Each agent derives the parent session from the chat URI.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: demote default chat to a uniform chat-map entry
Restructure the shared AgentSessionEntry so a session's default (main)
chat lives in the SAME per-session chat map as its peers, keyed by its
default-chat URI. Send/abort/model/agent/history now resolve any chat
via one uniform getChat lookup with no default-vs-peer storage split.
- agentPeerChats.ts: single _chats map + _defaultChatKey; getChat is the
uniform lookup; defaultChat/setDefaultChat/clearDefaultChat replace the
separate default-session slot; peer accessors read the same map.
- Claude/Copilot seed the default chat into the map (setDefaultChat) and
read it via defaultChat; _findChat is a single uniform lookup.
- Two lifecycle guards kept at the session boundary: createSession seeds
the default entry; disposeChat/createChat refuse the default chat.
- Codex (single-chat) unchanged; still resolves through the session URI.
No protocol/state changes; behavior identical (Agent* suites green).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: resolve per-chat effective working directory in session projection
Generalize the mergeSessionWithDefaultChat projection so a chat's resolved
session-context view (working directory, active clients, config,
customizations/MCP scope, …) applies to ALL chats, not just the default one.
The projection already spreads the full SessionState onto every chat, but it
ignored ChatState.workingDirectory — a per-chat override the protocol allows so
a subordinate/peer chat can run in its own git worktree. Operational consumers
(changeset/commit/discard/git/PR/sync services + agentService) read the effective
cwd via getSessionState(...)?.workingDirectory, so a chat with its own worktree
incorrectly reported the session default. Layer chat.workingDirectory over the
session default in the projection so no code path has to walk back to the session.
Pure orchestrator-side projection: no wire-protocol, IAgent, or persisted-shape
change. Docs generalized from "default chat" to per-chat effective context.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: route MockAgent createChat/fork through the resolved chat URI
Address CCR feedback: createChat/fork resolved the session via
_resolveChatTarget but passed the original chat argument, so a default-chat
call (session URI) would use the session URI instead of the resolved
default-chat channel URI. Destructure and pass the resolved { session, chat }
to match the other chat-addressed methods in the mock.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Add Responses API cache control markers
* Refactoring code
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* Gate Responses API cache breakpoints by model support
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* sessions: add explicit "Open this folder" action to the mobile workspace picker
Tapping a folder in the mobile workspace picker drills into that folder, which re-renders the search results and removes the visible checked state for the selected folder. That made it unclear that Done would select the browsed folder.
Add a confirm-close sentinel to the reusable mobile picker sheet so stay-open pickers can explicitly confirm a row. Use it from the workspace picker to prepend an "Open this folder" row for the currently browsed folder, while preserving child-folder drill-down and Done confirmation.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: redesign mobile workspace picker "Select folder" as a single pinned action
Replace the synthetic checked "Open {folder}" row (which produced two
checkmarks and a Done/Open redundancy) with one tinted-accent, pinned
"Select {folder}" primary action above the folder list. Folder rows now
carry a chevron to signal drill-down, recents show only at the root, and
the header button is a plain Cancel.
Adds reusable primitives to the mobile picker sheet:
- `IMobilePickerSheetItem.navigates` renders a trailing chevron and is
excluded from the radio-toggle.
- a pinned primary action driven declaratively by the search source via
`IMobilePickerSheetSearchSource.getPrimaryAction(query)`.
- `doneLabel` to override the dismiss button label.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Review comments
---------
Co-authored-by: osortega <osortega@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* policy: resolve managed settings per-key across delivery channels
Managed settings previously used a single authoritative source: the first
non-empty delivery channel (native MDM > server > file) won wholesale and the
others were ignored entirely. Switch to per-key precedence: the same order is
honored, but resolved key-by-key. A key locked by a higher-precedence channel
still cannot be overwritten, while keys a higher channel leaves unset are now
filled in by a lower channel.
Centralize the resolution in a new pickManagedSettings() (replacing
selectManagedSettings) that returns the merged bag, per-key provenance, and the
active sources, so policy evaluation and the Policy Diagnostics report share one
implementation. Build the merged bag with Object.fromEntries so an untrusted
__proto__ key cannot corrupt its prototype chain.
Rework the Policy Diagnostics report to show every contributing source, a per-key
Resolution table (effective value, winning source, and struck-through overrides),
and per-key policy attribution.
Add unit coverage for the per-key merge (provenance, fill-down, falsy values,
ordering, prototype-pollution guard) plus an end-to-end test proving two keys can
win from two different channels and both reach policy evaluation.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* policy: iterate managed-settings bags with Object.keys over own props
Address PR feedback: switch pickManagedSettings from `for...in` to a guarded
`Object.keys` loop so only own enumerable properties are visited (managed-settings
bags are untrusted input) and absent channels are skipped explicitly.
Note: `for...in` over an undefined bag was already a no-op (never threw), so this is
a robustness/clarity improvement rather than a crash fix.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* policy: tidy pickManagedSettings unit suite
Drop two redundant cases (distinct-keys and standalone activeSources ordering,
both already covered by the headline and empty/absent tests) and fold the
non-contributing-middle-channel ordering check into the empty/absent snapshot.
Verified against the live node unit runner (all green).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Re-syncs common/state/protocol/* from the agent-host-protocol repo now that
AgentCapabilities has been upstreamed and merged there. PR #323625 had
hand-edited the generated channels-root/state.ts to add AgentCapabilities +
AgentInfo.capabilities?; this regenerates it from AHP so it is no longer a
DO-NOT-EDIT hand-edit that the next sync would silently overwrite.
The upstreamed shape evolved from the hand-edit: instead of flat
`supportsMultipleChats`/`supportsFork` booleans, the generated
AgentCapabilities now exposes an MCP-style nested `multipleChats?:
MultipleChatsCapability { fork? }` (presence signals support, absence =
unsupported). Consumers that produced/read the protocol capabilities were
adapted to the new shape with equivalent semantics; the vscode-internal
ISessionCapabilities (supportsMultipleChats/supportsFork) is unchanged.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Add support for system-wide (OS global) keybindings
Allow user keybindings in keybindings.json to be marked with "systemWide": true so they register as operating-system global shortcuts that fire even when the window is not focused.
- Thread the systemWide flag through IUserFriendlyKeybinding, ResolvedKeybindingItem and KeybindingIO (read + serialize)
- New GlobalKeybindingsMainService owns Electron's globalShortcut, reconciles per-window registrations, resolves conflicts deterministically and routes triggers through the existing vscode:runAction path
- New renderer contribution syncs opted-in bindings to the main process, gated behind the experimental, off-by-default setting keyboard.enableSystemWideKeybindings with a one-time confirmation dialog
- Enable the GlobalShortcutsPortal feature on Linux/Wayland
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Avoid focusing the routing window on system-wide keybinding trigger
Force-focusing the routing (main) window before dispatching the command
pulled it to the foreground even when the command opens/reveals a different
window (e.g. openAgentsWindow reveals the agents window), producing a visible
flicker. Remove the force-focus and let the invoked command control what is
surfaced/focused, matching every other vscode:runAction sender.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Add workbench.action.focusWindow to raise the current window
Adds a generic command that brings the current window to the foreground and
focuses it using FocusMode.Force (which works even when the application is not
the active app). This lets users compose system-wide keybindings via
runCommands to reveal the window before running a command that surfaces UI in
it, e.g. [workbench.action.focusWindow, workbench.action.quickOpen].
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Address Copilot review: schema default, mnemonic comment, test grammar
- keybindingService.ts: correct 'systemWide' schema default to false to match
KeybindingIO parsing (defaults to false when absent/invalid)
- systemWideKeybindings.contribution.ts: add '&& denotes a mnemonic' translator
comment to the Enable button label
- keybindingEditing.test.ts: fix test title grammar (a user)
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Always enable system-wide keybindings; drop the enablement setting
Removes the experimental `keyboard.enableSystemWideKeybindings` setting so the
feature is always active: any user keybinding with "systemWide": true is a
candidate. The one-time confirmation dialog is retained and now serves as the
opt-out - its Enable/Disable choice is persisted as a tri-state consent
(unset -> ask, granted -> register, denied -> stay off and never re-ask).
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Make the first-run dialog an informational notice, not a permission prompt
The system-wide keybindings feature is always on, so the first-run dialog no
longer needs to grant/deny permission. Replace the Enable/Disable confirm dialog
with a single-button informational notice ("I Understand") shown once before the
first registration. Collapses the tri-state consent to a boolean acknowledged
flag; the feature has no decline path, so there is no stuck off-state.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
The bundled codex app-server advertises the hosted `image_generation`
tool whenever the developer's local `~/.codex` login uses a codex
backend (ChatGPT / agent-identity / PAT). `CodexProxyService` forwards
the request verbatim to the Copilot CAPI, which rejects it with
`400 unsupported_value` ("The requested tool image_generation is not
supported."). Disable the feature so codex never emits the tool.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Agents mobile (prototype): phone-aware dialog handler
Render IDialogService confirm/prompt as native bottom sheets on phone
layout, covering every modal in the Agents window in one place instead of
converting call sites individually.
The Agents window now registers a MobileDialogHandlerContribution (in place
of the standard web dialog handler) that drains the shared dialog model
through a MobileAwareDialogHandler. That handler delegates to the standard
BrowserDialogHandler off-phone and for text input / about, and on phone maps
the dialog model (message + detail + ordered buttons + optional checkbox)
onto a bottom sheet via the existing showMobileContentSheet primitive:
full-width stacked buttons at 44px, the safe action styled secondary, focus
moved into the sheet and restored on close.
Because confirmations from the sessions layer AND inherited chat / workbench
dialogs all flow through one handler, this covers them uniformly (e.g. delete
session, delete chat, discard changes, sign out) without per-call-site work.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Agents mobile: drop the bespoke archive confirm sheet
The phone-aware dialog handler now renders every IDialogService.confirm as
a bottom sheet on phone, so the per-call-site mobileArchiveConfirmSheet is
redundant. Collapse ArchiveSectionAction back to a single dialogService.confirm
(matching the group "Mark All as Done" action) and delete the bespoke helper.
This also brings the section archive-all in line with the group archive-all:
both now flow through the handler and get the same sheet on phone, including
the "Do not ask me again" checkbox (safe to surface now that it renders in a
proper bottom sheet rather than the previously cramped desktop dialog).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Address review: live checkbox state + dialog aria-describedby
- Track the chosen button and checkbox state across the whole interaction so
a backdrop / Escape dismiss (or a checkbox toggle) returns the live values
instead of a snapshot captured only on button click.
- Wire the message / detail to the sheet's role="dialog" element via
aria-describedby so screen readers announce them when the sheet opens.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: open the session when using Open in VS Code
The "Open in VS Code" action in the Agents window only opened the
session's workspace folder, leaving the user to manually open Chat and
click into the session. Hand off the active session resource through a
new chatSessionToOpen open-window option so the opened window restores
both the folder and the session via the existing vscode:openChatSession
IPC, matching the web variant that already forwards the session.
Fixes#323200
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: address CCR feedback for Open in VS Code session handoff
Only send vscode:openChatSession when exactly one window is opened so the
session handoff is not sent to an ambiguous target, and condense the
inline handoff comment to a single line.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* feat(automations): add session runner — dispatches automation runs via sessions layer
Implements AutomationRunner:
- Creates sessions with configured prompt, mode, model, and permission level
- Dispatches via createAndSendNewChatRequest (background session)
- Tracks run lifecycle: recordRunStart/updateRun on IAutomationService
- Error notification on run failure
- Cancellation handling for mid-flight stops
- Sessions titled with automation name via renameChat
Supporting changes:
- ICreateNewSessionOptions: added modelId, modeId, permissionLevel, isolationMode, branch
- ISessionsProvider: added optional setMode/setPermissionLevel/setIsolationMode/setBranch
- ISendRequestOptions: added optional source field
- createAndSendNewChatRequest now returns ISession | undefined
- Implementation applies createOptions to provider before sending
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* fix(automations): title the session before the agent host launches
Thread an optional title through ISendRequestOptions and apply it in
_sendFirstChat before sending, falling back to the query's first line.
The automation runner passes the automation name, so launch-time prompts
(e.g. worktree file transfer) show the automation name instead of the
prompt text. The post-send renameChat still sets the committed CLI title.
* Signing commit
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Support multiple chats for Claude agent-host sessions
Enable a single Claude (agent-host `provider === 'claude'`) session to own
multiple peer chats in the Agents window, matching the Copilot CLI experience.
- ClaudeAgent: add `_chatSessions` map plus `createChat` / `disposeChat` /
`getChats`, per-chat persistence, lazy resume of restored peer chats, and
per-chat routing on `sendMessage` / `abortSession` / `changeModel` /
`changeAgent`. Fork a peer chat from a source chat's SDK conversation at a
turn, falling back to a fresh chat when the fork anchor can't be resolved.
- Agent-host sessions provider: advertise `supportsMultipleChats` for the
`claude` logical session type in addition to `copilotcli`.
- Update SESSIONS.md and ClaudeAgent tests.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: fix Claude peer-chat signal routing and harden multi-chat lifecycle
Fix additional (peer) chats in Claude agent-host sessions getting stuck in
progress: a peer chat passes its `ahp-chat` channel URI as the session's
`sessionUri`, but `ClaudeAgentSession` derived its routing channel via
`buildDefaultChatUri(sessionUri)`, double-encoding it so the renderer never
matched the channel. Use the chat URI directly when `sessionUri` is already an
`ahp-chat` channel.
Also harden the peer-chat lifecycle per code review:
- serialize all catalog read-modify-write on the parent session id (createChat /
disposeChat / _updateChatCatalogModel) to avoid lost updates
- hold the per-chat lock across both materialize and send so disposeChat /
disposeSession serialize against an in-flight turn (no use-after-dispose)
- make _disposeChildChats async + per-chat serialized to avoid zombie entries
- abort provisional peer chats up front during shutdown
- route setPendingMessages steering to peer chats
- shape-guard the persisted catalog model
Refs https://github.com/microsoft/vscode/issues/322776
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: thread chat channel through setPendingMessages for peer-chat steering
Address CCR feedback: peer-chat steering was non-functional because
`AgentSideEffects._syncPendingMessages` always dispatched the parent session URI
to `agent.setPendingMessages`, so the Claude peer-chat routing branch was never
reached and steering landed on the default chat.
Add an optional `chat?` param to `IAgent.setPendingMessages` (mirroring
sendMessage/abortSession/changeModel), dispatch the chat channel from
`_syncPendingMessages` (undefined for the default chat), and route via it in
ClaudeAgent. Copilot/Codex 3-param implementations remain valid and unchanged.
Adds an AgentSideEffects dispatch test asserting the peer chat URI is forwarded
as the `chat` arg (and is undefined for the default chat).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* sessions: unify Claude session/peer-chat plumbing into one entry container
Address review feedback (connor4312): stop overloading session/chat URIs and
the parallel-map split that special-cased peer-chat dispatch.
- ClaudeAgentSession now takes an explicit `chatChannelUri`; its `sessionUri`
is always the real session URI and is never a chat URI
(`isAhpChatChannel(sessionUri)` can no longer be true). Per-chat resources
(db, overlay, config scope, server-tool advertise) key off a derived
`_storageUri` so peer chats stay isolated without overloading `sessionUri`.
- Drop the parallel `_chatSessions` map: a single `_sessions` map of
`ClaudeSessionEntry` containers now holds each session's default chat plus its
peer chats. Dispatch resolves a chat via `_findChat(session, chat)` / the
entry, and teardown disposes the whole entry (main + peers) via
`_teardownEntry`.
- Unify peer-chat message reconstruction with `getSessionMessages` via a shared
`_reconstructTurns(sdkId, routingUri, primeOn)`; remove the duplicated
`_getChatMessages`.
No behavior change to storage keying (main -> session URI, peer -> chat URI).
All ClaudeAgent / AgentSideEffects / CopilotAgent / AgentService node tests pass
(425 claude tests).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: add opaque providerData to chat catalog + multi-chat tests
Wave A + gate G-B1 of the multi-chat unification:
- AgentHostStateManager: add an opaque, agent-owned `providerData?: string`
to peer-chat catalog entries (addChat/restoreChat) plus getChatProviderData.
Stored verbatim and never parsed; the default chat carries none. This becomes
the single source of truth for a peer chat's backing-conversation token,
replacing the agents' private copilot.chats/claude.chats persistence.
- Add characterization tests for the StateManager catalog (default chat, add/
remove/restore, summary roll-up) and peer-chat + restore round-trip tests for
CopilotAgent and ClaudeAgent, guarding the upcoming de-dup waves.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: orchestrator owns the peer-chat catalog (Wave B de-dup)
Make AgentHostStateManager's catalog the single source of truth for peer
chats, removing the agents' private copilot.chats/claude.chats persistence:
- agentService: restore peer chats by enumerating the orchestrator's own
catalog (using the opaque providerData blob) instead of agent.getChats;
call materializeConversation(chatUri, providerData) before getSessionMessages
so the agent re-attaches its conversation backing; persist providerData on
createChat and re-persist on onDidChangeConversationData.
- IAgent: createChat returns IAgentCreateChatResult { providerData? };
add materializeConversation + onDidChangeConversationData.
- CopilotAgent / ClaudeAgent: stop writing their private *.chats catalogs;
shrink _chatSessions to a live-only map; decode providerData to rebuild the
chatUri -> sdkSessionId mapping; emit onDidChangeConversationData on per-chat
model/fork change. A one-time legacy *.chats READ (triggered by an undefined
providerData blob) migrates in-flight sessions.
Typecheck, valid-layers-check, and the agentService/Copilot/Claude/StateManager
suites (511 tests) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: add scope/conversation IAgent surface + dispatch mapper (gate G-C1)
Introduce the orchestrator-owned scope/conversation vocabulary on IAgent,
additively alongside the legacy (session, chat?) surface (kept as a compat
shim until waves C2-C5 migrate each agent):
- IAgent: add createScope/disposeScope and an IAgentConversations surface
(createConversation/disposeConversation/getMessages/fork, conversation-
addressed sendMessage/abort/changeModel/changeAgent).
- AgentService: map feature-level (session, chat) -> (agent, scope,
conversation) and own default-chat resolution; resolveConversationUri helper.
Typecheck, valid-layers-check, and the AgentService dispatcher suites
(112 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: agents adopt scope/conversation surface (Wave C)
CopilotAgent, ClaudeAgent and CodexAgent now implement the new scope/
conversation IAgent surface (createScope + conversations:
createConversation/disposeConversation/getMessages/fork and conversation-
addressed sendMessage/abort/changeModel/changeAgent), and agentSideEffects
threads it through where straightforward. The legacy (session, chat?) compat
shim is intentionally retained for now; it is removed centrally in gate G-C2.
Typecheck, valid-layers-check, and the Copilot/Claude/Codex/AgentService
suites (563 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: remove legacy (session, chat?) shim from IAgent (gate G-C2)
With every agent migrated to the scope/conversation surface (Wave C), drop
the agent-facing legacy methods — sendMessage(session,chat,...)/createChat/
disposeChat/getChats and the chat?-suffixed abort/changeModel/changeAgent —
leaving only the conversation-addressed surface on IAgent. AgentService,
agentSideEffects and the three agents migrate their remaining call sites; the
mock agent is updated to the new surface. The orchestrator-facing
IAgentService/IAgentConnection (session,chat) API and the wire protocol are
unchanged — they remain the (session,chat) -> conversation mapping boundary.
Net -209 lines. Typecheck, valid-layers-check, and the Copilot/Claude/Codex/
AgentService suites (559 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: add harness spawn-conversation channel + catalog routing (gate G-D1)
Generalize the subagent_started/subagent_completed signals into a first-class
membership channel: IAgent.onDidSpawnConversation({ scope, conversation,
parent? }) / onDidEndConversation(conversation). AgentService subscribes on
provider registration and routes spawned conversations straight into the
chat catalog (addChat/removeChat), so harness-spawned chats (teams, fleet,
subagents) and user-driven chats share ONE catalog path, preserving the
parent relation. Per-agent emission of these events lands in Wave D.
Typecheck, valid-layers-check, and the AgentService suite (107 passing) pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: agents emit spawn events + capability-driven UI gating (Wave D)
- CopilotAgent / ClaudeAgent emit onDidSpawnConversation/onDidEndConversation
from their subagent/fan-out paths, so harness-spawned chats flow into the
shared catalog via the G-D1 channel (carrying the parent relation).
- IAgentDescriptor advertises IAgentCapabilities { supportsMultipleChats,
supportsFork, supportsTeams }; the agent-host sessions provider maps these
onto ISessionCapabilities instead of the hardcoded
supportsMultipleChats(logicalSessionType) session-type check, and exposes
supportsFork/supportsTeams context keys so UI gates generically with no
per-harness branches.
Typecheck, valid-layers-check, and the agentHost + sessions provider suites
(1725 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: test default-chat rename is restored on restoreSession
Re-add coverage for restoring a default chat's independently-persisted custom
title (customChatTitle:<defaultChatUri>), homed in the dedicated restoreSession
suite using the localService + TestSessionDatabase pattern. A version of this
test arrived via a merge but was misplaced in the createChat suite; this puts it
in the right place. The behavior itself lives in AgentService.restoreSession.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Remove unused supportsTeams capability
The supportsTeams capability was fully plumbed (protocol to agents to
ISessionCapabilities to SessionSupportsTeamsContext) but had zero
consumers: no when-clause and no widget read it. Harness-spawned
teams/subagents surface automatically via onDidSpawnConversation
regardless of any flag, so this was speculative dead weight. Remove all
13 references across 9 files. supportsMultipleChats and supportsFork are
left untouched as they are actually consumed.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: unify subagent catalog membership onto the spawn channel (DR1)
Make the spawn-conversation channel the single owner of subagent catalog
membership, removing the duplicate add path:
- AgentSideEffects._handleSubagentStarted no longer calls addChat; it keeps
only the subagent lifecycle (ChatTurnStarted, _subagentChats tracking, parent
tool-call Subagent content, buffered-signal drain, teardown).
- AgentService now sequences a subagent_started/subagent_completed signal onto
the spawn-channel handlers (_onConversationSpawned/_onConversationEnded) via a
new onDidSessionProgress subscription registered BEFORE the side-effects
progress listener. This deterministically guarantees the subagent chat exists
in the catalog before its turn is started, independent of when the agent
registers its own subagent->spawn bridge (addChat/removeChat are idempotent).
- Extract the subagent-signal -> spawn-event mapping into shared helpers
(subagentSpawnConversationEvent/subagentEndConversation) reused by the agents'
bridges and the AgentService sequencer.
Adds a "subagent membership sequencing" suite: exactly one catalog entry with
parent origin/title/started turn regardless of order, buffered inner-signal
drain, and completion teardown. Typecheck, valid-layers, and the agent suites
(567 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: add multi-chat architecture spec
Living architecture spec for the agent-host multi-chat design (scope/session vs
conversation/chat, orchestrator-owned catalog, opaque providerData, unified
spawn channel, capability gating) with mermaid diagrams. Kept in sync with the
implementation, like SESSIONS.md.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: legacy peer-chat migration (BC1) + Copilot session container (F2)
Two changes to the Copilot/Claude agents and the orchestrator restore path:
BC1 - backward-compatible restore of legacy peer chats: sessions whose
additional chats were persisted only in the old agent-owned copilot.chats /
claude.chats format (no orchestrator peerChats catalog) previously restored
with those chats invisible. AgentService now performs a one-time migration when
the orchestrator catalog is absent (undefined, not []): it enumerates the
agent's legacy chats via a new migration-only IAgent.listLegacyChats, restores
them through the normal catalog path, and writes the peerChats key so the drain
runs once. Fixes the stale JSDoc that claimed a fallback removed in G-C2.
F2 - collapse CopilotAgent's default-vs-peer _sessions/_chatSessions two-map
split into a single _sessions map of a CopilotSessionEntry container (mirroring
ClaudeSessionEntry): the entry holds the default chat plus a nested _peerChats
map. Removes the special-casing Connor flagged on #323625.
Typecheck, valid-layers-check, and the AgentService/Copilot/Claude suites
(570 passing, incl. the migrate-once / empty-catalog / new-format restore
cases) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: rename agent scope/conversation surface to session/chat (N1)
Collapse the agent-facing vocabulary back to session/chat, so the whole stack
(protocol, orchestrator, UI, agents) speaks one language. The scope/conversation
terms were a 1:1 veneer over concepts already named session/chat elsewhere (the
create* methods even returned IAgentCreateSessionResult). The sessionUri vs
chatChannelUri TYPE separation is preserved — this is a naming change only.
- IAgent: createScope/disposeScope -> createSession/disposeSession; the
conversations surface (IAgentConversations) -> chats (IAgentChats) with
createChat/disposeChat/getMessages/fork + conversation-addressed send/abort/
changeModel/changeAgent now chat-addressed; materializeConversation ->
materializeChat; onDidSpawn/End/ChangeConversation* -> onDidSpawn/End/ChangeChat*.
- Types: IAgentSpawnConversationEvent -> IAgentSpawnChatEvent,
IAgentConversationDataChange -> IAgentChatDataChange; drop
IAgentCreateConversationOptions (reuse IAgentCreateChatOptions).
- Helpers: resolveConversationUri -> resolveChatUri and the private
_*Conversation* members across AgentService/agents renamed to _*Chat*.
- IAgentService/IAgentConnection/protocol/UI names unchanged (already session/chat).
- Reconcile agentSideEffects tests to the renamed chat surface (mock URI
normalization) and update MULTI_CHAT_ARCHITECTURE.md terms/diagrams.
Typecheck, valid-layers-check, and the agent suites (686 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* docs: align architecture diagram label with chat terminology
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* docs: align multi-chat spec terminology with the session/chat rename
Refine MULTI_CHAT_ARCHITECTURE.md wording after N1: the default chat's backing
SDK *session* (not "SDK chat") is the session, peer chats are backed by their
own sdkSessionId, and clarify the (session, chat) -> (agent, session URI, chat
URI) mapping label and the per-chat state description.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* agentHost: group refactor-added helpers into logical units (NS1)
Reduce loose top-level exports in common/agentService.ts introduced by the
multi-chat refactor (the pre-existing config/env helpers are left untouched):
- Move resolveChatUri to common/state/sessionState.ts next to its sibling
chat-URI helpers (buildChatUri/buildDefaultChatUri/isDefaultChatUri/
parseChatUri) — its logical home.
- Group the subagent signal -> spawn-channel mappers into an
`export namespace SubagentChatSignal { toSpawnEvent, toEndChat }` (mirroring
the existing AgentSession namespace), updating the Copilot/Claude bridges and
AgentService._sequenceSpawnedChat call sites.
Pure move/regroup, no behavior change. Typecheck, valid-layers-check, and the
agent suites (686 passing) all pass.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Make ISession.capabilities observable so late-hydrating capabilities reconcile
The agent-host adapter exposed `capabilities` as a live plain getter reading
the connection's root state. When `rootState.agents[].capabilities` hydrated
after a session's first `SessionState`, existing sessions were never
reconciled: a multi-chat catalog processed while `supportsMultipleChats` was
still `false` stayed collapsed to `[defaultChat]`, and the
`supportsMultipleChats`/`sessionSupportsFork` context keys stayed stale because
a plain getter cannot be tracked by the `setActiveSessionContextKeys` autorun.
Change `ISession.capabilities` to `IObservable<ISessionCapabilities>`. The
agent-host adapter derives it from `connection.rootState` (bridged via
`observableFromEvent`) with `derivedOpts` + `structuralEquals`, and re-applies
the last `SessionState` catalog in an autorun when capabilities change. Static
providers wrap their capabilities in `constObservable`; consumers read
`.read(reader)` (context keys) or `.get()` (one-shot).
Adds a regression test and updates SESSIONS.md and the sessions skill.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Keep completed subagent chats live (fix subagent integration tests)
A DR1 regression conflated 'subagent turn completed' with 'chat removed':
a subagent_completed -> removeChat path tore the child subagent chat out of
the catalog on completion, so subscribing to it after the parent turn
completed failed with 'Resource not found'. A completed subagent chat must
stay live and subscribable (merely hidden from listSessions), with its turn
completed via AgentSideEffects.completeSubagentSession; subagent chats are
removed only on session teardown via removeSubagentSessions.
- agentService._sequenceSpawnedChat: handle spawn only (no removal on completion)
- copilotAgent/claudeAgent spawn bridges: stop firing onDidEndChat on completion
- remove now-unused SubagentChatSignal.toEndChat (keep toSpawnEvent)
- keep onDidEndChat as a generic membership-removal hook
- tests: assert the subagent chat survives subagent_completed and that
completion does not fire onDidEndChat
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Add non-opaque backingSession to IAgentCreateChatResult
Introduces a first-class, non-opaque backingSession URI on the peer-chat
create result so the orchestrator can correlate and suppress a peer chat's
backing SDK session. Kept distinct from the opaque providerData blob so the
providerData opacity invariant is preserved.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Report peer-chat backingSession from Claude and Copilot agents
ClaudeAgent._createChat mints a fresh top-level SDK session per peer chat in
the same store its listSessions enumerates, so it now returns that session as
backingSession for the orchestrator to suppress. CopilotAgent sets it too for
uniformity (harmless — its peer sessions already don't leak).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Filter peer-chat backing sessions from the top-level session list
createChat now stamps a persisted peerChatBacking marker into the backing
session's database, and listSessions drops any enumerated session carrying it
(batched into the existing metadata overlay, mirroring the subagent filter).
Fixes Claude peer chats leaking as separate top-level sessions. Adds a unit
test covering the filter and its persistence across a restart, plus doc
invariant I7.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Share peer-chat scaffolding across Claude & Copilot agents
Extract the near-verbatim multi-chat peer scaffolding shared by the Claude
and Copilot agents into a new node-target module
`src/vs/platform/agentHost/node/agentPeerChats.ts`:
- Move the opaque `providerData` codec (`IPersistedChat`,
`encodeProviderData`, `decodeProviderData`) into the shared module and
export it. Use Claude's stricter `model` validation, which is a superset
of Copilot's unconditional cast. Both agents import it and drop their
private copies.
- Add a generic `AgentSessionEntry<TSession extends IDisposable>` container
holding the optional default session plus the peer-chat map. Rewrite
`CopilotSessionEntry` as an empty subclass and `ClaudeSessionEntry` as a
subclass that narrows `session` to non-optional.
Behavior-identical refactor; existing Agent* suites stay green.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Persist migrated legacy peer chats in a single atomic catalog write
`_migrateLegacyPeerChats` wrote the migrated peer chats to the orchestrator
catalog one entry at a time in a loop. Each `_persistPeerChat` is a separate
read-modify-write of `PEER_CHATS_METADATA_KEY`, so after the first write the key
is present containing only the first entry. If the agent-host process crashed
(OS kill, power loss, forced restart) after write 1 but before write N, the
catalog was left partial; on the next restart `_readPersistedPeerChatCatalog`
returns that subset (not undefined), the catalog-present branch short-circuits,
and migration never re-runs -- chats 1..N-1 are lost forever.
Write the whole migrated set in a single atomic `_enqueuePeerChatCatalogWrite`,
so the key is absent before and complete after; no partial catalog can survive a
crash mid-migration. Adds regression tests asserting the full set is persisted in
one write and that a rejected write leaves the key absent (never a subset).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The Customizations section in the Agents sidebar always reset to expanded
on reload because AICustomizationShortcutsWidget initialized _collapsed to
false and never persisted the state.
Read and store the collapsed state via IStorageService (StorageScope.PROFILE)
so the section restores its collapsed/expanded state on reload. Apply the
persisted state to the DOM after each render instead of hard-resetting to
expanded.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
