Fix/resolve different bugs (#964)

* fix: don't overlap thinking banner over conversation

* feat: support message queue and add attention indicator

* fix(shell): use c to copy for claude

* fix: strip timestamp tags from cursor

* fix: select project when loading session from direct URL

* feat: support attached images for all providers

* feat(opencode): support permission options

* fix: resolve source control and chat UX bugs

- make staging real in the git panel: new /stage and /unstage endpoints,
  /status now reports the actual index via porcelain -z (handles renames,
  conflicts, and paths with spaces), and Stage/Unstage All run real git
  commands instead of toggling client-side state
- add branch search to the header dropdown and Branches tab
- persist the chosen permission mode per provider so a brand-new chat
  keeps it once the session id arrives, instead of reverting to default
- replace cmdk's fuzzy model search with strict token matching so
  "chatgpt" no longer surfaces unrelated models
- unit tests for the git status parser

* feat(git): commit graph in history view and cross-spawn everywhere

- render a VSCode-style commit graph in the source control history:
  lane-assignment algorithm, SVG strip with colored rails, merge and
  branch curves, commit dots, and branch/tag badges per commit
- /commits returns parents and ref decorations across all branches
  (--branches --remotes --tags --topo-order) using unit-separator
  fields so pipes in commit subjects can't break parsing
- collect stats via a single --shortstat pass instead of one
  `git show --stat` call per commit; raise the history limit to 50
- replace child_process spawn with cross-spawn in all runtimes,
  routes, and services: it resolves .cmd shims and PATHEXT on
  Windows (fixing taskmaster's npx invocations) and delegates to
  native spawn elsewhere, removing the per-file win32 ternaries
- unit tests for the log parser and lane assignment

* fix: remove gemini support since google discontinued it

* fix: address code review findings

- validate chat image attachments server-side: only files inside the
  ~/.cloudcli/assets upload store may reach provider file reads
- harden asset serving with nosniff and attachment disposition for
  SVGs to prevent stored XSS
- show the timestamp on image-only user messages
- serialize git stage/unstage calls and defer the status re-sync so
  rapid toggles can't interleave or flicker
- use a literal hex fallback for commit ref badges (alpha suffix on a
  var() string produced invalid CSS)
- stop binding a URL session to a guening
  project instead
- add the missing attentionRequiredIndicator key to all sidebar locales
- clean up dangling conjunctions left the
  de/ru/tr/ja/zh-CN/zh-TW READMEs

* fix: address code scanning findings

- enforce a second image trust boundary in the provider builders:
  buildClaudeUserContent/buildCodexInputItems only accept files inside
  the upload store or the run's working directory (CodeQL path
  injection)
- drop an always-true selectedProject check in handleSubmit
- remove the unused resume option from spawnCursor

* fix: use CodeQL-recognized containment check for image reads

Replace the path.relative guard in isPathInsideDirectory with the
resolve + startsWith(root + sep) idiom so the path-injection barrier
is visible to code scanning; behavior is unchanged.

* fix(security): canonicalize Claude image attachment reads

Resolve image attachment paths with realpath before reading so symlinks
inside allowed roots cannot escape to arbitrary files. Preserve support for
symlinked project/upload roots and add focused coverage for both cases.

* fix: show agent subtask

* fix(sessions): title app-created sessions from the first user message

App-created sessions (started by sending a message from cloudcli) were
titled with placeholder names — "Untitled Codex Session" from the disk
indexer and, briefly, "New session" from the empty canonical upsert —
before a later sync finally settled on the right text, causing the
sidebar title to flicker.

- Codex/OpenCode synchronizers now title app-created sessions (distinct
  app id mapped to a provider id) from the first user message, while
  sessions found purely by indexing keep their existing setup. Claude
  keeps its AI-generated titles; Cursor already used the first message.
- Decode OpenCode's JSON-string-literal prompts so titles no longer
  surface wrapped in quotes; hoist unwrapJsonStringLiteral into
  shared/utils since it's now used by both the reader and synchronizer.
- Guard the sidebar upsert merge so an empty summary can never blank out
  a title that is already set.
- Add codex/opencode synchronizer tests for the app-created vs indexed
  naming paths.

* fix(chat): make message queuing reliable across sessions and turn boundaries

Queued messages had four related defects:

- A queued message flashed and then vanished at flush time: concurrent
  transcript refreshes (the `complete` handler racing the watcher-triggered
  update) could resolve out of order, letting a stale response overwrite
  newer server messages after the optimistic row was pruned. Session slots
  now carry a monotonic fetch ticket and discard stale fetch/refresh/
  fetchMore responses.

- Switching sessions flushed the previous session's queued draft into the
  newly viewed one (sending it with the wrong provider's settings, e.g. a
  Claude model into a Codex session). The composer flush is now scoped to
  its session, and a draft restored into an idle session sends after a
  short grace period so a live-run ack can cancel it.

- The thinking banner never appeared for a queued turn: the chat handler's
  session-keyed completeRun safety net could fire after a queued message
  had already started the session's next run, emitting a spurious
  `complete` that killed it. The safety net is now scoped to its own run
  via completeRunIfCurrent (with a regression test).

- Queued messages only sent while their session was being viewed. Drafts
  now persist their send options (model, effort, permissions) at queue
  time, and a new app-level useQueuedMessageAutoSend hook dispatches a
  non-viewed session's queued message as soon as its run completes, using
  the storage key as the claim ticket to prevent double sends.

server/opencode-cli.js

@@ -1,19 +1,51 @@
-import { spawn } from 'child_process';
 import fsSync from 'node:fs';
 
 import crossSpawn from 'cross-spawn';
 import Database from 'better-sqlite3';
 
+import { appendImagesInputTag } from './shared/image-attachments.js';
 import { sessionsService } from './modules/providers/services/sessions.service.js';
 import { providerAuthService } from './modules/providers/services/provider-auth.service.js';
 import { providerModelsService } from './modules/providers/services/provider-models.service.js';
 import { notifyRunFailed, notifyRunStopped } from './services/notification-orchestrator.js';
-import { createCompleteMessage, createNormalizedMessage, getOpenCodeDatabasePath } from './shared/utils.js';
+import { createCompleteMessage, createNormalizedMessage, flattenPromptForWindowsShell, getOpenCodeDatabasePath } from './shared/utils.js';
 
-const spawnFunction = process.platform === 'win32' ? crossSpawn : spawn;
+// cross-spawn resolves .cmd shims/PATHEXT on Windows and delegates to
+// child_process.spawn everywhere else.
+const spawnFunction = crossSpawn;
 
 const activeOpenCodeProcesses = new Map();
 
+/**
+ * Maps the UI permission mode onto OpenCode's non-interactive controls.
+ *
+ * OpenCode has no single "permission mode" flag; each mode uses a different
+ * lever of the `opencode run` CLI (verified against v1.17.13):
+ * - plan              → the built-in read-only `plan` agent (`--agent plan`).
+ * - bypassPermissions → `--auto`, which auto-approves every permission that
+ *                       is not explicitly denied in the user's config.
+ * - acceptEdits       → the OPENCODE_PERMISSION env var, whose JSON body the
+ *                       CLI merges into its permission config. Forcing
+ *                       `edit: allow` guarantees file edits go through while
+ *                       every other rule stays under the user's own config.
+ * - default           → nothing; the user's opencode.json governs. In
+ *                       non-interactive `run` mode any `ask` rule is denied.
+ *
+ * Exported for tests only.
+ */
+export function resolveOpenCodePermissionOptions(permissionMode) {
+  switch (permissionMode) {
+    case 'plan':
+      return { args: ['--agent', 'plan'], env: {} };
+    case 'bypassPermissions':
+      return { args: ['--auto'], env: {} };
+    case 'acceptEdits':
+      return { args: [], env: { OPENCODE_PERMISSION: JSON.stringify({ edit: 'allow' }) } };
+    default:
+      return { args: [], env: {} };
+  }
+}
+
 function resolveOpenCodeEffort(model, effort, modelsDefinition) {
   const selectedModel = modelsDefinition?.OPTIONS?.find((option) => option.value === model);
   const allowedEfforts = selectedModel?.effort?.values?.map((value) => value.value) || [];
@@ -92,7 +124,7 @@ function readOpenCodeTokenUsage(sessionId) {
 
 async function spawnOpenCode(command, options = {}, ws) {
   return new Promise((resolve, reject) => {
-    const { sessionId, projectPath, cwd, model, effort, sessionSummary } = options;
+    const { sessionId, projectPath, cwd, model, effort, sessionSummary, images, permissionMode } = options;
     const workingDir = cwd || projectPath || process.cwd();
     const processKey = sessionId || Date.now().toString();
     let capturedSessionId = sessionId || null;
@@ -223,14 +255,20 @@ async function spawnOpenCode(command, options = {}, ws) {
       if (resolvedEffort) {
         args.push('--variant', resolvedEffort);
       }
+      const permissionOptions = resolveOpenCodePermissionOptions(permissionMode);
+      args.push(...permissionOptions.args);
       if (command && command.trim()) {
-        args.push(command.trim());
+        // Image attachments ride along as an <images_input> path list appended
+        // to the prompt; the session history reader strips the tag back out.
+        // opencode is a .cmd shim on Windows, so the whole argument must be
+        // newline-free or cmd.exe silently truncates it at the first newline.
+        args.push(flattenPromptForWindowsShell(appendImagesInputTag(command.trim(), images)));
       }
 
       opencodeProcess = spawnFunction('opencode', args, {
         cwd: workingDir,
         stdio: ['pipe', 'pipe', 'pipe'],
-        env: { ...process.env },
+        env: { ...process.env, ...permissionOptions.env },
       });
 
       activeOpenCodeProcesses.set(processKey, opencodeProcess);