Qwen Code sessions and token history, live

Run qwen in any terminal and the session appears in the notch with its project, live state, current activity, tool calls, files, commands, and diffs. Expand it for the transcript Qwen records locally, with subagents grouped under their parent instead of flattened into separate runs.

Qwen Code PermissionRequest prompts surface in the notch: approve with ⌘ A, deny with ⌘ N, or open the waiting terminal with ⌥ T. Always Allow appears only when Qwen includes a concrete allow suggestion. Qwen’s ask_user_question interaction stays owned by its native terminal; AgentPeek shows the session waiting for input and Open Terminal takes you there instead of pretending it can answer through the permission channel.

The session composer sends follow-up prompts into resumable Qwen sessions through qwen --acp. AgentPeek uses Qwen ACP session loading, prompt streaming, and cancellation, so replies and tool activity update in the session card and a prompt in flight can be stopped. ACP PermissionRequest approvals stay native in the notch; ask_user_question remains owned by Qwen’s terminal.

AgentPeek reads the token counts in Qwen’s local JSONL session records and shows the monthly total plus daily history in the usage panel. Qwen does not report a local rate-limit quota through those records, so AgentPeek shows the counts it can prove and no fabricated percentage or reset window.

AgentPeek manages hooks in ~/.qwen/settings.json, or the equivalent under $QWEN_HOME, and Doctor checks and repairs the entries. Historical usage scans follow $QWEN_RUNTIME_DIR or Qwen’s default runtime root. Live session lookup follows the exact transcript path Qwen reports, wherever Qwen wrote it, and otherwise scans the projects tree under that runtime root. Quick Routes opens the active Qwen skills, config, project logs, and root folder directly.