Removes permission patterns already shown as the rendered bash command
Keeps distinct patterns visible in the permission card
Adds tests for the pattern filtering behavior
Compare normalized attachment MIME types with the selected model's declared input modalities and show a non-blocking warning for incompatible files.
Recheck newly added attachments, restored drafts, async metadata, and existing files after model changes while avoiding warnings when capability metadata is unavailable. Summarize affected filenames and localize the warning across every supported locale.
Add focused modality compatibility coverage, document the composer behavior, and keep model metadata subscriptions stable to prevent startup render loops.
Users can now attach Microsoft Office documents (.docx, .pptx, and .xlsx) and OpenDocument files (.odt, .odp, and .ods) from the shared web, desktop, mobile, and VS Code chat surfaces.
Document text is extracted locally and sent as a text/plain file part with the original filename, keeping the visible user message clean. Supported embedded PNG, JPEG, GIF, and WebP images are sent as separate image parts, with matching [filename] citations preserved near their source paragraph, slide object, spreadsheet cell anchor, or OpenDocument position. Presentation notes, spreadsheet values, headers, and footers are included where available.
Document expansion is metadata-validated and bounded against oversized entries, excessive uncompressed data, unsafe paths, invalid image signatures, attachment-name races, and dangling citations after truncation. Generated document parts are published to the composer atomically.
Add fflate for worker-backed ZIP extraction and narrowly allow blob workers in the VS Code webview CSP without permitting blob scripts. Include focused fixtures for every supported format, extraction limits, positional citations, collision recovery, atomic attachment state, and CSP behavior.
Add picker and MIME support for more source-code, diff, notebook, structured-log, SVG, and Draw.io formats.
Convert Jupyter notebooks into readable text while omitting binary outputs, and sanitize HAR credentials, cookies, sensitive query parameters, and request/response bodies before files enter chat state.
Convert HEIC and HEIF images to JPEG up front, centralize attachment preparation in a focused module, and cover the new validation and transformation behavior with regression tests.
Attachment pickers now share an allowlist for supported file types
Local attachments are normalized to consistent MIME types before upload
VS Code file picker now respects extension filters and larger files are allowed
Removes the separate review check and relies on review labels/comments instead
Marks workflow/tooling failures with review:automation-failed
Updates review guidance to reflect the advisory verdict model
Adds a dedicated `review` check for the exact pull request HEAD
Keeps manual review commands reported separately from the default branch job
Documents the new review check behavior in contributing guidelines
Adds a General settings switch to show or hide the macOS menu bar item.
Persists the new desktopMacMenuBarEnabled setting and passes tray state into Electron.
Disables tray sync and native tray setup when the menu bar is turned off.
Require reading .github/PULL_REQUEST_TEMPLATE.md alongside AGENTS.md and CONTRIBUTING.md
Tighten guidance for loading matching skills and task-required references before editing
Add PR handoff expectations for concrete evidence and rollback considerations
Allow Desktop to skip its in-process OpenChamber server with OPENCHAMBER_SKIP_LOCAL_SERVER=1 while continuing to load the packaged UI shell.
Carry local runtime availability through the boot contract so unavailable or unconfigured remotes enter a remote-only chooser instead of offering broken local recovery actions. The chooser can select saved instances, add a server by URL, or redeem an OpenChamber pairing link over direct or E2EE relay transports.
Keep additional windows, Mini Chat, background startup, and unreachable-host recovery functional without a local origin. Render boot and recovery surfaces with the active theme background rather than exposing the native vibrancy backing.
Document the environment variable and cover serverless boot routing plus malformed pairing imports with focused tests.
Improve OpenChamber responsiveness under large session workloads while fixing
cache, synchronization, and persistence correctness across runtimes, projects,
directories, and worktrees.
- prioritize selected and visible sessions during bootstrap and defer
non-critical enrichment work
- reduce redundant message loading, event processing, store publication, and
hidden sidebar work
- prevent stale session and message requests from overwriting newer
authoritative state
- preserve existing data when authoritative fetches fail instead of treating
failures as successful empty responses
- scope session materialization, messages, drafts, queues, todos, pins,
permissions, folders, tabs, Git state, and pull request data by runtime and
directory identity
- harden runtime switching, reconnect, cleanup, mutation reconciliation, and
persisted-state ordering
- preserve live subagent Task linkage when metadata arrives after an older
message request or while streaming parts are suspended
- coalesce overlapping tail refreshes without losing newer refresh demand
- improve cold-session loading by moving deferrable work out of the critical
bootstrap path
- isolate URL authentication, mobile credentials, native secrets, and other
runtime-owned state across endpoint changes
- bound long-lived caches and remove avoidable allocations from event and
rendering hot paths
- limit virtualization to archive collections where it improves rendering
without disrupting active sidebar layout
- stabilize session folders, pin ordering, expanded state, and persisted
sidebar behavior
- open skill files through the same secure editor and outside-workspace grant
flow used by file navigation, including worktree sessions
- expand regression coverage for stale completions, runtime collisions,
reconnect behavior, persistence races, authoritative empty results, and
subagent refresh ordering
- document the updated synchronization, cache ownership, performance, and
runtime-isolation invariants
Adds a new `hmr-react-scan` web dev option in `oc-dev`
Turns on React Scan via `VITE_ENABLE_REACT_SCAN` for that mode
Documents Electron workspace trust so fresh installs fetch the runtime automatically
- Assistant turn bottom padding pb-8 -> pb-2, desktop user message top
padding pt-6 -> pt-4 (footer-to-bubble gap 60px -> 28px)
- Drop the recap note's -mt-6 that compensated the old large padding
- Working indicator shows provider icon and model name with the live
status (e.g. 'Fable 5 is reading file'), localized via
chat.statusRow.modelStatus
- Pin auto-follow to the exact fractional scroll maximum and re-pin on
every passive follow instead of skipping within tolerance
- Round message text line-height to whole pixels so streamed content
grows on the pixel grid; kills the 1px vertical jitter of
bottom-anchored rows during streaming
- Remove assistant message headers; show provider icon, model, agent,
thinking variant, duration and time in the turn footer (metadata left,
hover-revealed actions right)
- Merge turns started by hidden user messages (subagent nudges) into the
previous turn so Activity, footer and spacing stay continuous
- Treat compaction summary text (info.summary) as justification activity
in sorted mode and skip it when picking the turn summary
- Interleave activity segments with standalone tool rows so Agent Task
sits chronologically between activity sections
Use dedicated access and refresh token labels for Cursor quota credentials.
Add localized token placeholder strings across supported settings languages.
Loads sessions from the active workspace directory
Fetches session messages using both session ID and directory
Resets expanded raw messages when the directory changes
Add a root-session action that creates a generated worktree from the session directory's current branch, transfers uncommitted changes, and moves the parent session plus its descendants through OpenCode's control-plane API.
Reuse existing project/worktree topology and quick-create behavior, keep the UI non-blocking, reconcile live and global session state across directories, and roll back partial moves and failed worktree creation safely.
Split worktree bootstrap readiness into directory-created, git-ready, and setup-ready phases across web and VS Code. Session moves wait for Git readiness while existing setup-aware flows continue waiting for full setup completion, and worktree removal is serialized with active bootstrap tasks.
Expose the move only for idle root sessions, show localized progress and explanatory tooltips in the sidebar, and keep pending/ready worktree metadata synchronized with authoritative session attachments to avoid stale setup indicators.
Add coverage for control-plane payloads, session-state migration, bootstrap phase ordering and compatibility, removal races, progress metadata, and fast-ready attachment races.
Retries transient index.lock conflicts during worktree population
Removes unchanged stale locks automatically and continues bootstrap
Adds coverage for stale lock recovery
Updates the SDK dependency across web, UI, and VS Code packages.
Refreshes the lockfile and package versions to match the release.
Adds unreleased changelog notes for the current set of user-facing changes.
Replace the legacy terminal flow with a shared authenticated WebSocket
runtime used across web, desktop, relay, and mobile surfaces.
- introduce the v3 terminal protocol with scoped attachments, snapshots,
ordered output, bounded replay history, reconnects, and explicit lifecycle
- harden PTY creation, restart, resize, close, force-kill, idle cleanup,
shell selection, login mode, environment sanitization, and appearance sync
- add runtime-aware terminal APIs with relay authentication and Electron parity
- add a fullscreen mobile terminal workspace with touch scrolling,
long-press selection, safe-area controls, quick keys, and Ctrl/Alt input
- add terminal selection attachments, preview detection, project actions,
shell settings, and localized UI
- harden Ghostty rendering, resize recovery, Unicode handling, block
characters, line height, and stale-row behavior
- remove the obsolete terminal SSE path and update reverse-proxy guidance
- expand terminal runtime, transport, input, selection, and store coverage
- avoid duplicate web builds when preparing mobile assets in root CI builds
Add a VS Code-specific foreground permission responder while preserving the server-owned implementation for web, desktop, and mobile runtimes.
Persist the authoritative VS Code policy in extension global state and expose matching GET/PUT bridge routes. Broadcast policy updates to the sidebar, session editor panels, and agent manager so every active webview observes the same explicit per-session policy.
Resolve missing child-session lineage through OpenCode, honor nearest explicit ancestor overrides, deduplicate concurrent requests, retry transient replies, and reconcile pending permissions after enablement, bootstrap, and reconnect. Treat resolved requests as handled and route notification suppression through the same responder outcome.
Keep post-toggle reconciliation failures non-fatal after policy persistence and fail closed when lineage or replies cannot be confirmed. Document that auto-accept intentionally cannot run while every OpenChamber webview is closed or suspended.
Regenerate the shared UI icon sprite from current source usages. The video-chat icon is no longer referenced anywhere under packages/ui/src, so the generator removes its stale SVG path from the generated asset.
Keep the new-session draft open until the backend confirms that the session was created successfully. This prevents a failed request from closing the draft and discarding the user's creation context.
Restore the exact submitted composer text after a new-session send failure and persist it as the draft. Only restore when the composer is still empty or unchanged, so text entered while the request is pending is not overwritten.
Add regression coverage that simulates an offline session creation request and verifies that the draft remains open with its title intact.
Add pin and unpin actions for user and assistant text messages, with clear compaction-survival labels, localized tooltips, status-info active styling, and VS Code gating where the server runtime is unavailable.
Persist pinned message IDs, creation timestamps, and roles under the OpenChamber session metadata namespace using fresh-read merge updates so goal, review, and other metadata remain intact.
Introduce a server runtime that reacts to OpenCode's dedicated session.compacted event, fetches pinned messages by ID, extracts and chronologically orders their text parts, and injects them as hidden synthetic context through prompt_async. The restoration prompt tells the agent to use the context silently while work remains and limits idle summaries to one short paragraph.
Track the last handled compaction summary to avoid replay duplication, tolerate individually missing pinned messages, integrate runtime shutdown, document ownership and limitations, and cover metadata round trips plus compaction injection behavior with focused tests.
Supports {env:NAME} and {file:path} apiKey substitutions in provider config.
Keeps resolved credentials and file contents server-side.
Adds coverage for env and file-based credential resolution.
Uses thinkingLevel for Gemini 3 Flash models
Keeps older Gemini Flash models on thinkingBudget: 0
Updates docs and tests for the new Google request payload
New drafts now receive the active project ID from sidebar project actions.
Project-group draft creation also forwards the selected project.
Aligns draft opening behavior across project and session selection flows.
Shows the evaluation provider/model used for the latest successful goal audit in the UI.
Persists evaluation model metadata on session goals and covers it in tests.
Adds diagnostic logging for small-model calls and session-goal audit parsing.
The editor font size setting controls code editors and should not also resize the chat composer. Code and conversational input have different readability and layout requirements, so forcing them to share one size is not appropriate.
Delete the share opinion dialog and its sidebar trigger.
Remove the related toast and footer action wiring.
Clean up unused i18n entries in all supported languages.
Re-checks authoritative session status after the quiet window
Skips auditing while a direct child session is still busy or retrying
Retries the quiet window when live status data is unavailable
* fix(chat): anchor prompt navigator to last turn at chat bottom
The scroll spy activates the last turn whose top edge crossed the reading
line (100px below the container top). A final turn short enough to fit
below that line could never become active — sitting at the very bottom of
the chat kept the previous prompt highlighted on the navigator rail.
When the container is scrolled to the bottom (within 8px), force the last
turn active.
* feat(chat): enable prompt navigator by default
Users who already persisted an explicit preference keep their choice.
Prefers a real APK asset when the update API points to an AAB
Keeps direct APK links from the API unchanged
Uses the GitHub releases API as a fallback for Android mobile downloads
Use the Task part's state.metadata.sessionId as the only live identity contract for child sessions. Remove timestamp, status, and ordering heuristics that could misassociate concurrent subagents or repeatedly scan directory sessions while parts stream.
Extract Task metadata parsing and child-summary projection into a focused model with cached projections for unchanged message records. Preserve output and part-level metadata parsing only for legacy persisted records, and keep standalone Task rows visible when sorted activity groups are rendered.
Validated with focused Task and turn projection tests, UI type-check, lint, and dead-code analysis.
* feat(chat): prompt navigator list preview with prompt filtering
The hover preview is now an interactive scrolling mini-list of prompts:
rows render as bordered two-line cards, the highlighted row stays inside
a center dead zone and the list glides only near the window edges, wheel
steps the highlight, and the panel stays open when the pointer moves into
it so a click can be corrected inside the list.
Rail entries are filtered to real prompts: previews are built from
normalized user display parts (synthetic context stripped), fully
synthetic user messages are excluded, and shell-mode messages show their
extracted command via the shared shell bridge helpers.
* fix(chat): render shell command status transitions
The injected /shell text part carries live state in shellAction, which
the render-relevant part comparator ignored — a running→completed update
reached the store but never re-rendered the message row until the next
send. Compare shellAction command/output/status for text parts.
* fix(sync): stream shell bridge part updates while running
Streaming suspension keeps part updates out of the static message records
while an assistant message streams, relying on the live streaming-tail
path to render it. Shell-mode bridge messages are hidden from the
timeline and rendered inside the user row, so they have no live path —
suspension froze their output chunks and left the card without a Show
output action until the run finished. Exempt shell bridges (single bash
tool part parented to a synthetic shell-marker user message) from
suspension; their updates arrive at command-output pace, not delta pace.
* feat(chat): syntax-highlight shell command card
Render the shell-mode command and its output through the shared
WorkerHighlightedCode (Shiki) with bash grammar, matching the bash tool
part presentation, instead of plain pre blocks.
Keep always-on instructions concise and route specialized work through focused skills. Split large skills into progressive references and add dedicated change, desktop, sync, and performance guidance.
Replace repeated project-by-session directory matching across sidebar hooks with a shared ownership index that resolves each unique directory once and exposes direct project and folder-scope buckets.
Gate destructive folder reconciliation on authoritative session data and topology readiness, preserve last-known worktrees after discovery failures, and retain nested-project, VS Code, active/archive dedupe, and Windows drive-root semantics.
Narrow cross-directory subscriptions to session and status slices so streaming deltas no longer trigger global aggregation. Reuse a cached session ID index for permission lineage checks instead of rebuilding it on every session switch.
On the reported 15-project, 67-worktree, 14,561-session shape, ownership indexing averages 3.81 ms versus roughly 450 ms for the cache-only hotfix.
Validation: 28 targeted tests, UI type-check, UI lint, and dead-code analysis.
* fix(chat): rework prompt navigator rail into sliding tape with hover preview
- Pin the active indicator to the target during programmatic scrolls so the
scroll spy's intermediate reports don't drag it backwards mid-animation
- Replace the visibility-ratio active-turn picker with a stable reading-line
rule (last turn whose top is above the line), dropping IntersectionObserver
- Replace the list panel with a Codex-style gutter: the whole strip is one
hover/click target mapped to the nearest tick, with a per-prompt preview
card that follows the cursor
- Cap the rail at a fixed window of ticks; hovering the edges carousels
through the rest, with gradient masks hinting at more content
- Render ticks as a tape that glides to keep the active prompt centered,
remounting on history prepend to avoid spurious slide animations
- Keep load-earlier as a compact button aligned over the tick column
* fix(chat): shrink navigator gutter when message column sits under it
On narrow windows the centered message column extends under the rail's
full-width invisible hover zone, which swallowed clicks on the right edge
of user bubbles — including the expand/collapse control. Measure the
column against the gutter and switch to a narrow hit zone when they
overlap.
* fix(sync): stop runaway history auto-load on sessions with empty assistant messages
An assistant message fetched with zero parts (e.g. a run aborted before any
output) was stored as absence — indistinguishable from parts that were never
fetched. getSessionMaterializationStatus therefore reported the session as
never renderable, so the ensure-renderable effects (ChatContainer,
ModelControls) retried syncSession forever; each retry refetched the whole
grown window and fired another background prepend, progressively loading the
entire history of large sessions on open.
Commit an explicit empty [] snapshot for assistant messages so fetched-empty
counts as renderable, while non-assistant messages keep the absent
representation and its no-op commit behavior.
Reproduced and verified headless against a real 857-message session: before,
20 message fetches escalating to limit=857; after, one initial page and a
single progressive-mount prepend.
Omits the craft-goal command from draft starters in VS Code
Prevents the starter from being resolved or pinnable in that runtime
Keeps non-VS Code behavior unchanged
Adds section headers and tooltip help for the goal settings area
Reorganizes chat settings into Session Assistance, Reasoning, Message Appearance, Tools & Files, and Composer groups
Adds localized labels and search entries for the new section headings
Relay connect used to serialize a dead LAN probe (up to 8s per stale address
on mobile, 2-4s on desktop) in front of the relay attempt, then paid a second
WebSocket connect + E2EE handshake because the probe tunnel was thrown away.
- mobile probeConnectionCandidates: race the relay probe against the direct
chain with a 1.5s direct headstart; a live LAN still wins, a dead one no
longer delays startup
- relay probes adopt their tunnel as the runtime tunnel (adoptRelayTunnel)
instead of dialing a fresh one — applies to auto-connect, pairing redeem,
password login, and the desktop host switcher's relay fallback
- relay probe drops the /health round-trip: the E2EE handshake already proves
the server identity, /auth/session alone proves liveness and auth
- desktop restoreDesktopRelayRuntime: same headstart race; a late direct
success hot-switches back (stable runtimeKey); startup probe now passes
expectedServerId so a re-leased LAN address never sees the token
- launch splash shows 'Connecting to device: <label>' with animated dots
under the (still centered) logo, translated in all locales
- editing a saved instance no longer rebuilds it from the URL field alone:
the id is passed through, relay/https candidates are preserved, and a
token-key change migrates the Keychain token instead of orphaning it
All local instances share the data dir and therefore the relay identity
(serverId), so concurrent relay hosts evicted each other at the relay worker
(4001: Control replaced) and paired devices landed on whichever local process
won last — often a stale dev server, surfacing as 'Unable to reach server'
and devices stuck on relay with 503s on newer endpoints.
- relay/host-lock.js: per-machine claim file (relay-host.lock, {pid}); stale
claims from dead pids are ignored; unwritable data dir falls back to
pre-lock behavior
- relay/service.js: start only when the claim is free or ours, otherwise
'standby' with the holder pid in lastError; 30s watcher takes over when the
claimant dies and stands down when another process claims; pairing-link
creation and explicit /relay/enable force-claim (user intent wins)
- mobileConnections.ts: log candidate-refresh skip reasons and the refresh
result instead of failing silently
Candidates refresh (server + mobile + desktop clients):
- GET /api/client-auth/connection/candidates returns the server's current
LAN URLs, relay candidate, and serverId for already-paired devices
- /health and /api/version expose serverId so clients can verify a learned
address belongs to the expected server before sending their bearer token
- mobile: refresh saved candidates over the live transport after every
connect/wake, hot-switch relay->LAN when a fresh address is reachable;
serverId gate on direct probes; token no longer sent to /health
- desktop: refresh stored host apiUrl after a relay connect and hot-switch
back to direct; electron probe verifies serverId before authenticated fetch
Fixes found while debugging a dead pairing:
- settings: strict reader that throws on corrupt/unreadable file instead of
returning {}; relay signing/encryption key generation is now gated on it,
so a swallowed read failure can no longer mint a new server identity and
orphan every paired device (loud log when a keypair IS generated)
- SessionAuthGate: bounded auto-retry for transient session-check failures
(initial request racing the relay tunnel's first WS attempt, startup 5xx)
- the goal dialog renders as the shared MobileOverlayPanel bottom sheet
on mobile instead of a centered dialog
- tapping the target button to ARM keeps the soft keyboard open (the next
message is the objective; same guard as the attachment/mic buttons),
while opening the manage sheet lets the keyboard close as usual
- capacitor: bottom-sheet overlays cap their height by the keyboard inset
AND the top safe area — with the keyboard raised while typing inside a
sheet, 100dvh does not shrink (native resize is off) and the panel could
slide under the notch/status bar; applies to every MobileOverlayPanel
The distillation moved from PlanView into setSessionGoal — the shared
choke point for every goal source (composer, fork dialog, plans) — and
into the scheduled-tasks goal creation server-side. Any objective over
5000 chars is distilled into completion criteria by the small model; on a
transient failure a head+tail excerpt keeps the intent (top) and the
acceptance criteria (bottom) with a trim marker, plus a toast. The
working agent always received the full prompt in chat — only the auditor
is bound by the limit.
Plan-goal objectives are capped at 5000 chars for the auditor while the
working agent reads the full plan from its file. Plans over the limit are
now distilled by the small model into completion criteria (end goals +
per-phase verification, no implementation steps), prefixed with a header
pointing back at the plan file so every continuation re-anchors on the
live source of truth. If distillation fails (transient small-model
hiccup), a head+tail excerpt keeps the plan's intent (top) and acceptance
criteria (bottom) with a trim marker between — sacrificing the
implementation middle the agent reads from the file anyway — and a toast
tells the user the objective is degraded.
Removes recent-project grouping from the sidebar project list
Keeps project ordering based on manual and default sort rules only
Reduces memoized session-derived state in the sidebar
Move per-session permission auto-accept policy ownership from the UI to the
OpenChamber server so enabled sessions continue running when clients disconnect
or the server restarts.
- persist explicit per-session policies in OpenChamber settings
- inherit the nearest explicit policy across subagent session hierarchies
- allow child sessions to opt out of an inherited parent policy
- immediately accept matching global and directory-scoped pending requests
- process future requests without requiring a connected UI client
- reconcile pending permissions after startup and event-stream reconnects
- deduplicate concurrent requests and retry transient reply failures
- synchronize policy updates across connected clients
- migrate existing browser-persisted policies to server storage
- suppress auto-accepted permission cards before they enter UI state
- show deduplicated permission toasts for inactive sessions
- preserve foreground-only permission handling in VS Code
- integrate directory-aware notification routing from main
- add coverage for persistence, inheritance, retries, reconciliation, pending
requests, client hydration, and inactive-session toasts
Use authoritative session.idle and session.error events for notifications
while retaining legacy message.updated handling for compatibility.
Classify sessions through targeted, directory-aware session lookups instead
of fetching the full session list. Cache confirmed root and parent session
relationships without treating failed lookups as root sessions.
Honor subagent completion settings and templates across the server-driven
web, desktop, and mobile paths, and bring the VS Code webview notification
policy to feature parity.
Use freshly synchronized VS Code settings, retry failed settings syncs,
extract session error messages, and deduplicate authoritative and legacy
completion and error events.
Adds /craft-goal autocomplete and chat handling for starting a Goal crafting session.
Introduces new Magic Prompts content and localized labels/descriptions for Goal crafting.
Migrates desktop draft starters to include Craft a Goal once and persists the migration marker.
Compaction fixes (observed in a real long run):
- the summary message's zeroed tokens froze the goal counter at its
pre-compaction value; segments now close with the previously displayed
total as a continuity floor
- audits and continuations after a summary tail now take execution params
(provider/model/agent/variant) from the newest non-summary assistant
turn instead of inheriting agent 'compaction' and the summarize model
File-backed objectives:
- the objective text lives in <data-dir>/goals/<sessionId>.md, keyed by
session id (one goal per session, a new goal overwrites the file);
metadata carries only an objectiveFile flag so session.updated fanout
stays light, and never a path — ids are pattern-validated before any
filesystem access
- limit raised to 5000 chars, no snapshot field: the UI fetches content
via PUT/GET/DELETE /api/goals/objective/:sessionId (behind the blanket
/api auth gate), writes the file before stamping metadata, and falls
back to an inline objective when the write fails
- the loop reads the file fresh on every tick, so objectives are
live-editable mid-goal; a missing file falls back to the inline text
- scheduled goal tasks write the objective file server-side; VS Code
degrades to the audit note (route unavailable there by design)
Arm the target button in the composer and the next prompt becomes a goal:
the server keeps the session working toward it (idle tick -> small-model
audit -> continuation) until the objective is verifiably complete, blocked,
or out of budget — even with the UI closed.
Server (packages/web/server/lib/session-goal):
- event-driven loop on the global SSE hub; goal state lives in
session.metadata.openchamber.goal (merge-safe patches, stale-write guard
by goal id), so it survives restarts and syncs to every client for free
- the small-model audit (objective + last assistant turn only, language
pinned to the objective) is the sole termination authority; blocked needs
3 consecutive verdicts, audit outages tolerate one unaudited continuation
then stop the goal as resumable-blocked
- hard stops: optional token budget, auto-continuation cap (Resume grants a
fresh allowance), turn errors; user abort pauses the goal instead of
blocking it, and resuming over an aborted tail nudges immediately
- token accounting as a snapshot of the latest turn (input + cache.read +
output), goal-relative via a creation baseline and segmented across
compactions; a compaction summary skips the audit and continues
- continuations reuse the session's own provider/model/agent/variant
UI:
- three-mode target button (arm / disarm / manage dialog), informational
goal strip with inline pause/resume and an Evaluating indicator, sidebar
state glyph, objective length counter (2000-char server clamp),
read-only completed goals
- goal entry points: composer (sessions and drafts), start-new-session-
from-answer dialog, plan implement dialog (plan content becomes the
objective), scheduled tasks (Run as goal + budget)
- Settings -> Chat -> Goal: feature toggle + default token budget with
three-layer parity (web server, client persistence, VS Code bridge);
VS Code renders goal state but hides the entry points (the loop runs in
the web server only)
Notifications: per-turn "ready" notifications are suppressed while a goal
is active; settling sends one final notification (desktop, web-push, APNs
generic titles with the session name as body) honoring the completion
toggle. Error/question/permission notifications are untouched.
Docs: user guide (session-goals) in all 9 locales + sidebar entry,
scheduled-tasks cross-reference, server module DOCUMENTATION.md.
Bohdan Triapitsyn