@deepseek-harness-tui/dsh-tui 0.10.0 → 0.10.2

package/lib/types/dsh-adapter/channel.d.ts

@@ -1,1380 +1,8 @@
-import { type Agent, type AgentHandle, type AgentStatus } from '@deepseek-ai/dsh-agent';
-import type { LlmModelInfo, LlmProviderInfo } from '@deepseek-ai/dsh-llm';
-import { type ContentBlock } from '@deepseek-ai/dsh-llm';
-import { type RecapOutcome } from './recap.js';
-import { type BalanceResult } from '../deepseekBalance.js';
-import { type SessionEvent } from '@deepseek-ai/dsh-session';
+import { type ChannelLaunchOptions } from './channel/state.js';
 import type { Context } from '@deepseek-ai/cordis';
-import { type CommandCompletion, type LocalCommand } from '../commands.js';
-import { type SessionTreeData } from './sessionTree.js';
-import { type PreviewEntry, type SessionSummary } from './sessions/index.js';
-import { type FileCandidate } from '../utils/fileSuggestions.js';
-import type { OAuthProviderStatus, ProviderSetupHost } from './providerWizard.js';
-import { type SessionModeSpec } from '../sessionModes.js';
-import { type PageMarginSetting, type ScrollGutterMode, type StatusBarConfig, type ToolBackground } from '../tuiDisplayPrefs.js';
-import { type SubagentState } from './subagents.js';
+import { type Agent } from '@deepseek-ai/dsh-agent';
+import type { ChannelState } from './channel/types.js';
 export type { SubagentState } from './subagents.js';
-import { type BackgroundJobState, type BackgroundJobStatus } from './jobs.js';
-import type { SpinnerMode } from '../components/Spinner/spinnerMode.js';
-import { type ActivityState } from 'dsh-working-activity/status';
-import { type TuiWorkspaceCommand, type TuiWorkspaceCommandResult, type TuiWorkspaceTarget } from './workspaces.js';
-import { type TuiSettingsSection } from './settings-sections.js';
-import type { SettingsHost } from './settingsEditor.js';
-import { type TuiSceneDescriptor } from './scenes.js';
-import type { TuiRewindMode } from './extension-events.js';
-import { type TranscriptImage } from './transcript-images.js';
-type ChannelImageBlock = Extract<ContentBlock, {
-    type: 'image';
-}>;
-type ChannelImageMediaType = ChannelImageBlock['attachment']['mediaType'];
-export interface StagedImageInput {
-    data: Uint8Array;
-    mediaType: ChannelImageMediaType;
-    name?: string;
-    /** Absolute local path the bytes were read from, for the preview card's
-     *  path row. Not handed to the attachment store and never persisted. */
-    path?: string;
-}
-/** Opaque capability returned for one staged composer image. The visible
- * `[Image #N]` label is deliberately absent: PromptInput owns presentation
- * numbering while this id is the non-reusable attachment identity. */
-export interface StagedImageHandle {
-    readonly stageId: string;
-}
-/** One visible composer token bound to its opaque staged-image capability. */
-export interface ComposerImageRef {
-    readonly token: string;
-    readonly stageId: string;
-}
-/** Text plus the image capabilities that belong to that exact draft. */
-export interface ComposerSubmission {
-    readonly text: string;
-    readonly images?: readonly ComposerImageRef[];
-}
-/** UI-safe projection of one settled DSH registry command. Keeping the
- * result kind across the adapter boundary lets the composer retain a
- * rejected image draft instead of treating the error text as success. */
-export type ExternalCommandOutcome = {
-    readonly kind: 'success';
-    readonly text: string;
-    readonly consumeDraft: true;
-} | {
-    readonly kind: 'error';
-    readonly text: string;
-    readonly consumeDraft: boolean;
-};
-/** Tool-call card state used by the transcript renderer. */
-export interface ToolRow {
-    readonly callId: string;
-    readonly name: string;
-    /** Raw JSON arguments as the model produced them (displayed truncated). */
-    readonly argsText: string;
-    /** Full arguments, shown when Ctrl+O verbose mode is on; dropped when the
-     *  row is folded (session log retains it). */
-    argsFull?: string;
-    status: 'running' | 'ok' | 'error';
-    resultText?: string;
-    /** Full result text, shown when Ctrl+O verbose mode is on. */
-    resultFull?: string;
-    errorText?: string;
-    /** Tool-owned render intent from dsh-tools `presentCall` (diff/terminal/
-     *  generic). Drives the structured card body instead of the raw text. */
-    callView?: ToolCallView;
-    /** Tool-owned completed-state view from `presentResult` (applied diff
-     *  hunks, terminal output, read content…). Wins over callView once set. */
-    resultView?: ToolResultView;
-    /** Wall-clock start of the call (live elapsed while running). */
-    startedAt: number;
-    /** Settled wall-clock duration, written by tool/result. */
-    durationMs?: number;
-}
-/** One file change in a tool presentation (dsh-tools FileDiff). */
-export interface ToolFileDiff {
-    readonly path: string;
-    /** Prior content, or null for a new file / no before-image. */
-    readonly oldText: string | null;
-    readonly newText: string;
-}
-/** Pending-call render intent (structural subset of dsh-tools ToolCallView). */
-export type ToolCallView = {
-    readonly card: 'generic';
-    readonly title: string;
-    readonly kind?: string;
-} | {
-    readonly card: 'terminal';
-    readonly title: string;
-    readonly description?: string;
-    readonly cwd?: string;
-} | {
-    readonly card: 'diff';
-    readonly title: string;
-    readonly diffs: readonly ToolFileDiff[];
-};
-/** Completed-call render intent (structural subset of dsh-tools
- *  ToolResultView). `web` results and unknown shapes fall back to raw text. */
-export type ToolResultView = {
-    readonly card: 'generic';
-    readonly title?: string;
-    readonly content?: ReadonlyArray<{
-        readonly type: string;
-        readonly text?: string;
-    }>;
-} | {
-    readonly card: 'terminal';
-    readonly title?: string;
-    readonly output?: string;
-    readonly exitCode?: number;
-    readonly signal?: string;
-} | {
-    readonly card: 'diff';
-    readonly title?: string;
-    readonly diffs: readonly ToolFileDiff[];
-} | {
-    readonly card: 'read';
-    readonly title?: string;
-    readonly path?: string;
-    readonly content?: ReadonlyArray<{
-        readonly type: string;
-        readonly text?: string;
-    }>;
-} | {
-    readonly card: 'search';
-    readonly shape: 'matches';
-    readonly title?: string;
-    readonly files: ReadonlyArray<{
-        readonly path: string;
-        readonly matches: ReadonlyArray<{
-            readonly lineNumber: number;
-            readonly line: string;
-        }>;
-    }>;
-    readonly truncated: boolean;
-    readonly total: number;
-} | {
-    readonly card: 'search';
-    readonly shape: 'paths';
-    readonly title?: string;
-    readonly paths: readonly string[];
-    readonly truncated: boolean;
-    readonly total: number;
-};
-/** Re-derives the presentation views foldRows dropped, threaded into
- *  foldBack (module-level, no ctx access) by the channel. */
-export interface ToolViewPresenter {
-    call(name: string, rawArgs: string): ToolCallView | undefined;
-    result(name: string, rawArgs: string, data: SessionEvent<'tool/result'>['data']): ToolResultView | undefined;
-}
-/**
- * Subagent row: displays a subagent's lifecycle (started → running → completed/failed).
- * Derived from agent.task events and history events.
- */
-export interface SubagentControl {
-    interrupt(agentId: string): boolean;
-}
-/**
- * Background-job row control (`/jobs` panel): cancellation with the same
- * authority the owning agent itself would use (`job_kill`). Returns false
- * when the jobs service is absent or the job is unknown/foreign.
- */
-export interface JobControl {
-    kill(id: string): boolean;
-}
-export interface SubagentRow {
-    agentId: string;
-    runId?: string;
-    description: string;
-    provider?: string;
-    model?: string;
-    effort?: string;
-    status: SubagentState['status'];
-    startedAt: number;
-    completedAt?: number;
-    durationMs?: number;
-    outputLines: string[];
-    toolCalls: SubagentState['toolCalls'];
-    tokens?: SubagentState['tokens'];
-    summary?: string;
-    stopReason?: string;
-    error?: string;
-}
-/** One background job as a live transcript card (see `kind: 'job'`). */
-export interface JobRow {
-    id: string;
-    kind: string;
-    label: string;
-    status: BackgroundJobStatus;
-    detail?: string;
-    startedAt: number;
-    finishedAt?: number;
-    /** Mirrored `job_output` tail feeding the card's three-line waterfall. */
-    outputLines: readonly string[];
-}
-/**
- * One rendered transcript row. The DSH session log is the source of truth:
- * rows are derived from `session/event` records (and the initial
- * `snapshotLiveSessionEvents(agent.session)` replay), never from optimistic local state.
- */
-export interface ChatRow {
-    id: number;
-    kind: 'user' | 'assistant' | 'tool' | 'notice' | 'reasoning' | 'interrupt' | 'local' | 'local-output' | 'compact' | 'subagent' | 'job';
-    /** Extra label for non-human user rows (e.g. `steering`). */
-    label?: string;
-    /** Actual execution location for `!command` rows. */
-    executionTarget?: string;
-    text: string;
-    /** Durable session image blocks, loaded lazily through the attachment store. */
-    images?: readonly TranscriptImage[];
-    /** True while an assistant step is still streaming chunks. */
-    streaming?: boolean;
-    /** Present on `tool` rows; the card model. */
-    tool?: ToolRow;
-    /** Present on `subagent` rows; the subagent state snapshot. */
-    subagent?: SubagentRow;
-    /** Present on `job` rows; the background-job state snapshot. */
-    job?: JobRow;
-    /** Event wall-clock time (transcript-mode metadata, assistant rows). */
-    time?: number;
-    /** Present on `reasoning` rows once settled: thinking wall-clock duration. */
-    durationMs?: number;
-    /** Source session event seq — present on every log-derived row (rewind
-     *  fork anchor on user rows; window-floor bookkeeping for the rest). */
-    seq?: number;
-    /** True when the row's full text was folded to keep the transcript window
-     *  bounded (see MAX_ROWS); the session log still holds the full content
-     *  and loadOlder() restores it. */
-    folded?: boolean;
-    /** True when loadOlder() restored this row from the log; restored rows are
-     *  exempt from the next fold pass so a restore is not instantly undone. */
-    restored?: boolean;
-    /** True on rows created by LIVE event handling (not replay/resume/fold
-     *  restore) — the smooth-streaming reveal animates freshly-arrived
-     *  content only; replayed history must paint complete. Set once at
-     *  creation; never mutated afterwards. */
-    fresh?: boolean;
-}
-/** One 计费时段（高峰/空闲）的 token 累计。 */
-export interface TokenBucket {
-    input: number;
-    output: number;
-    cacheRead: number;
-    cacheWrite: number;
-}
-/** Running token totals across the session's assistant messages. */
-export interface TokenUsage {
-    input: number;
-    output: number;
-    /** Prompt-cache hit tokens across the session (priced at the hit rate). */
-    cacheRead: number;
-    /** Prompt-cache write tokens across the session (priced with uncached input). */
-    cacheWrite: number;
-    /** Peak-hour tokens (billed at peak rates) — each usage lands in a bucket
-     *  by its event time, so a session spanning both windows is priced per
-     *  window instead of all at the current rate. */
-    peak: TokenBucket;
-    /** Off-peak-hour tokens (billed at idle rates). */
-    idle: TokenBucket;
-}
-/** 全零 token 累计（新会话 / 复位用）。 */
-export declare function emptyTokenUsage(): TokenUsage;
-/** In-process working-line snapshot derived from the base session stream. */
-export type ActivityStatus = ActivityState;
-/** A transient status message shown above the prompt input. */
-export interface NotificationItem {
-    id: number;
-    text: string;
-    /** Theme color key; defaults to dim. */
-    color?: 'error' | 'warning' | 'success';
-    /** Auto-dismiss after this many ms (default 4000); 0 = sticky, removed
-     *  only through the early-dismiss handle. */
-    timeoutMs: number;
-}
-/**
- * Durable same-session goal projection surfaced on the channel (see
- * {@link Channel['goal']}). Mirrors the goal domain's `GoalSnapshot` +
- * replay counters; declared locally so the UI needs no dsh-goal dependency.
- */
-export interface ChannelGoal {
-    id: string;
-    revision: number;
-    objective: string;
-    phase: 'active' | 'paused' | 'blocked' | 'complete';
-    /** Total admitted goal-round cap. */
-    maxGoalRounds: number;
-    /** Highest admitted continuation round so far. */
-    roundsStarted: number;
-    /** Present exactly while `phase` is `blocked`. */
-    blockedReason?: {
-        code: string;
-        message: string;
-    };
-}
-/** The observable outcome of adopting a persisted session. */
-export type ResumeResult = {
-    readonly ok: true;
-} | {
-    readonly ok: false;
-    readonly reason: 'working';
-} | {
-    readonly ok: false;
-    readonly reason: 'unavailable';
-} | {
-    readonly ok: false;
-    readonly reason: 'cancelled';
-} | {
-    readonly ok: false;
-    readonly reason: 'failed';
-    readonly error: string;
-};
-/**
- * One session's state in the agent view.
- * States describe the session's current lifecycle:
- * `working` — a turn is running; `needs-input` — an approval request is
- * parked for this agent; `idle` — live and waiting for the next prompt;
- * `completed` — a live agent whose last turn ended (task finished, waiting);
- * `failed` — the last turn ended with an error; `stopped` — the session's
- * process is gone (persisted only).
- */
-export type AgentViewStatus = 'working' | 'needs-input' | 'idle' | 'completed' | 'failed' | 'stopped';
-/** One row in the agent view list. */
-export interface AgentViewRow {
-    /** Session id — the attach/dispatch target. */
-    readonly id: string;
-    /** Display title (session title, or a fallback from the prompt/cwd). */
-    readonly title: string;
-    /** Absolute working directory the session runs in. */
-    readonly cwd: string;
-    /** One-line activity summary derived from the session's recent output. */
-    readonly summary: string;
-    readonly status: AgentViewStatus;
-    /** True when an agent for this session is alive in THIS process (✻ vs ∙). */
-    readonly live: boolean;
-    /** True when this is the session the TUI terminal is attached to. */
-    readonly current: boolean;
-    /** Unix epoch milliseconds when the session was created. */
-    readonly createdAt: number;
-    /** Unix epoch milliseconds of the session's latest activity. */
-    readonly updatedAt: number;
-}
-/** The observable outcome of dispatching a new background session. */
-export type AgentViewDispatchResult = {
-    readonly ok: true;
-    readonly sessionId: string;
-} | {
-    readonly ok: false;
-    readonly reason: 'unavailable';
-} | {
-    readonly ok: false;
-    readonly reason: 'failed';
-    readonly error: string;
-};
-/** The observable outcome of backgrounding the attached session. */
-export type BackgroundResult = {
-    readonly ok: true;
-    readonly backgroundedSessionId: string;
-} | {
-    readonly ok: false;
-};
-/** Secret-free credential metadata for configuration and status surfaces. */
-export interface CredentialStatus {
-    configured: boolean;
-    source?: string;
-    writable: boolean;
-}
-/** One entry of the latest todo-list snapshot (mirrors dsh-tool-todo's
- *  `TodoItem`; declared locally so the adapter needn't depend on that plugin). */
-export interface TodoPanelItem {
-    content: string;
-    status: 'pending' | 'in_progress' | 'completed';
-}
-/** One named prompt contribution with its model-visible text. */
-export interface LoadedContextEntry {
-    /** Provider-declared name (e.g. `harness:identity`, `deployment:persona`). */
-    readonly name: string;
-    /** The interpolated text the model receives for this entry. */
-    readonly text: string;
-}
-/** One discovered workspace instruction file (AGENTS.md-family). */
-export interface LoadedContextFile {
-    /** Model-facing path (e.g. `./AGENTS.md`). */
-    readonly displayPath: string;
-}
-/** One model-invocable skill from the skill registry. */
-export interface LoadedContextSkill {
-    readonly name: string;
-    readonly description: string;
-}
-/** One skill in the live agent's catalog, for the `/skills` picker (issue #204). */
-export interface SkillInfo {
-    readonly name: string;
-    readonly description: string;
-    /** True when `/name` invokes it (it appears in the `/` menu, issue #86). */
-    readonly userInvocable: boolean;
-    /** Discovery source bucket (bundled / user-* / project-* / runtime / …). */
-    readonly source: string;
-}
-/** One model-visible tool from the prompt assembly. */
-export interface LoadedContextTool {
-    readonly name: string;
-    readonly description: string;
-}
-/**
- * Snapshot of everything a fresh conversation for the current agent will
- * load: the assembled system prompt (ordered sections, dynamic context,
- * tools), the workspace instruction files baseline discovery would inject,
- * and the skill catalog. Declared locally so screens and helpers consume a
- * self-contained contract instead of the dsh-system-prompt/dsh-skill types.
- */
-export interface LoadedContext {
-    /** Ordered system-prompt sections after strict variable interpolation. */
-    readonly sections: readonly LoadedContextEntry[];
-    /** Dynamic context contributions (runtime snapshot parts). */
-    readonly contexts: readonly LoadedContextEntry[];
-    /** Workspace instruction files (AGENTS.md-family) discovered for the cwd. */
-    readonly files: readonly LoadedContextFile[];
-    /** Model-invocable skills, when the skill registry is mounted. */
-    readonly skills: readonly LoadedContextSkill[];
-    /** Model-visible tools in assembly order. */
-    readonly tools: readonly LoadedContextTool[];
-}
-/**
- * The public channel surface a screen renders: the full transcript and live
- * status snapshot (tokens, spinner, working activity, goals, todos, loaded
- * context) plus every action the TUI can take (submit, steer, cancel,
- * rewind, resume, model switching, …). Implementations mutate internal state
- * and bump `version` so subscribed screens re-render.
- */
-export interface Channel {
-    /** Monotonic version — bump on every mutation so screens can re-render. */
-    readonly version: number;
-    readonly rows: readonly ChatRow[];
-    readonly status: AgentStatus | 'starting' | 'disposed';
-    readonly sessionTitle: string;
-    /** Per-session accent color name (`/color`), '' when unset — persisted via
-     *  a `session/color` log event so it survives resume/rewind. Renders as
-     *  the prompt-input border + session label chip accent. */
-    readonly sessionColor: string;
-    readonly agentId: string;
-    /** TUI-owned generation that changes on every live Agent rebind. */
-    readonly agentBindingGeneration: number;
-    /** `dsh-tui.recapOnOpen` (default on): auto-summarize the session tail
-     *  into the dim AutoRecapRow when the session opens/resumes. Read live
-     *  (settings service), so a `/settings` change applies on the next
-     *  session switch; absent settings service → on. */
-    readonly autoRecapOnOpen: boolean;
-    /** Resolved model id (from the plugin config). */
-    readonly model: string;
-    /** Provider route of the live agent. */
-    readonly provider: string;
-    /** Raw cordis.yml `provider` key (undefined when unset) — the boot-time
-     *  pin `/reload` must never override. */
-    readonly configuredProvider: string | undefined;
-    /** Raw cordis.yml `model` key (undefined when unset). */
-    readonly configuredModel: string | undefined;
-    /** Explicit cordis.yml `preset` (undefined = roster default wins) — `/reload`
-     *  must not override a static deployment choice. */
-    readonly configuredPreset: string | undefined;
-    /** Explicit cordis.yml `activityFrames` (undefined = pref/default wins). */
-    readonly configuredActivityFrames: string | undefined;
-    /** Explicit cordis.yml `lang` (undefined = settings/lang.json wins). */
-    readonly configuredLang: string | undefined;
-    /** Running token totals across the session's assistant messages. */
-    readonly tokens: TokenUsage;
-    /** Working directory of the session. */
-    readonly cwd: string;
-    /** Human-facing cwd (remote POSIX path/URI instead of a host alias). */
-    readonly displayCwd: string;
-    /** Current git branch, when the cwd is inside a git worktree. */
-    readonly gitBranch: string | undefined;
-    /** True between turn/start and turn/end — drives the working spinner. */
-    readonly working: boolean;
-    /** True while a user-requested abort (Ctrl+C/Esc interrupt) has not yet
-     *  converged — no turn/start or turn/end has retired the aborted turn.
-     *  Chat uses it so a repeated Ctrl+C during a stuck abort force-exits. */
-    readonly cancelPending: boolean;
-    /** Which phase the spinner should present while working. */
-    readonly spinnerMode: SpinnerMode;
-    /** Chars streamed as text this turn (feeds the spinner token counter). */
-    readonly responseChars: number;
-    /** Number of tool calls still in flight this turn. */
-    readonly activeToolCount: number;
-    /** Wall-clock ms of turn/start (spinner elapsed timer). */
-    readonly turnStart: number;
-    /** Last user prompt text (sticky header + statusline). */
-    readonly lastUserText: string;
-    /** Transient notifications, newest last. */
-    readonly notifications: readonly NotificationItem[];
-    /** Adapter-advertised context capacity for the model route, when known. */
-    readonly contextWindow: number | undefined;
-    /** Reasoning effort of the latest request header, when the adapter sets one. */
-    readonly reasoningEffort: string | undefined;
-    /** The live route's reasoning-effort level ids, low → high (the last entry
-     *  is the top tier). Consumed by top-tier-triggered UI (effort ignition). */
-    readonly effortLevels: readonly string[] | undefined;
-    /** Usage of the most recent request (context share + cache hits come from
-     *  this, not the running totals — each request's input IS the context). */
-    readonly lastUsage: {
-        input: number;
-        output: number;
-        cacheRead: number;
-        cacheWrite: number;
-    } | undefined;
-    /** Output tokens per second of the current/last turn's response, when known. */
-    readonly tps: number | undefined;
-    /** Per-turn tps samples (sparkline history), oldest first. */
-    readonly tpsSamples: readonly {
-        tps: number;
-        at: number;
-    }[];
-    /** Latest in-process working-activity snapshot. */
-    readonly workingActivity: ActivityStatus | undefined;
-    /** Working-activity indicator preset name (`moon8`/`moon`/…/`random`). */
-    readonly activityFrames: string | undefined;
-    /** Edit/Write diff presentation preference (`auto`/`split`/`unified`). */
-    readonly diffLayout: 'auto' | 'split' | 'unified';
-    /** Thinking-block display (`preview` = 2-3 line live stream + fold per
-     *  step; `full` = expanded until turn end). */
-    readonly thinkingFold: 'preview' | 'full';
-    /** Live tool-card background treatment. */
-    readonly toolBackground: ToolBackground;
-    /** What the fullscreen transcript's right gutter shows (settings
-     *  `dsh-tui.scrollGutter`: turn timeline / proportional scrollbar /
-     *  nothing). */
-    readonly scrollGutter: ScrollGutterMode;
-    /** Root page inset (settings `dsh-tui.pageMargin`): a preset name
-     *  (`none` / `slim` / `normal` (default) / `roomy`) or a custom `NxM`
-     *  spec (columns per side × rows top/bottom) inset the whole UI from the
-     *  terminal edges — terminals without their own viewport padding (bare
-     *  WSL, tmux, SSH) otherwise hug the screen border. */
-    readonly pageMargin: PageMarginSetting;
-    /** Terminal-card header folding (settings `dsh-tui.foldTerminalCommand`):
-     *  collapse a multi-line command title to its first line + count hint. */
-    readonly foldTerminalCommand: boolean;
-    /** Whether the session-name chip shows on the prompt top border's right
-     *  side (settings `dsh-tui.promptSessionLabel`; off by default). */
-    readonly promptSessionLabel: boolean;
-    /** Whether the fullscreen draft editor is enabled (settings
-     *  `dsh-tui.expandEditor`; on by default) — gates the ⛶ affordance and
-     *  the expandEditor shortcut. */
-    readonly expandEditor: boolean;
-    /** Smooth streaming reveal (settings `dsh-tui.smoothStreaming`; on by
-     *  default): live-arriving assistant text, expanded thinking, and tool
-     *  call bodies paint through a ~30fps reveal instead of jumping per
-     *  provider burst. */
-    readonly smoothStreaming: boolean;
-    /** Live status-footer visibility and compactness preferences. */
-    readonly statusBar: Readonly<StatusBarConfig>;
-    /** Whether the header's pixel whale art shows (settings `dsh-tui.whale`). */
-    readonly whale: boolean;
-    /** Whether the settled header whale keeps behaving during the welcome
-     * phase — fin flutters, tail thumps, sleep after inactivity (settings
-     * `dsh-tui.whaleIdle`; on by default — an explicit false keeps the
-     * settled header timer-free). The first agent turn freezes the whale to
-     * the static standard frame regardless of this flag. */
-    readonly whaleIdle: boolean;
-    /** Minimal mode (settings `dsh-tui.minimal`): no header splash, no emoji
-     *  glyphs, no decorative colors; code highlight and tool colors stay. */
-    readonly minimal: boolean;
-    /** Whether the in-process working-activity line is shown (config.activity). */
-    readonly activityEnabled: boolean;
-    /** Whether the segmented context bar row shows in the status footer
-     *  (config.contextBar; the status/mode lines are unaffected). */
-    readonly contextBarEnabled: boolean;
-    /**
-     * Current same-session goal projection, when a goal exists. Derived live
-     * from the durable goal events in the session log — top-level
-     * `goal/change` snapshots (every goal mutation appends one) plus the
-     * goal-sourced continuation rounds that advance the counter — so this
-     * snapshot tracks create/edit/pause/resume/complete/block/clear in real
-     * time and replays correctly on resume/rewind.
-     */
-    readonly goal: ChannelGoal | undefined;
-    /**
-     * Latest todo-list snapshot (`todo/write` whole-list event, last write
-     * wins). Log-only UI state, updated live and on replay.
-     */
-    readonly todos: readonly TodoPanelItem[];
-    /**
-     * Snapshot of the context a fresh conversation for this agent will load
-     * (system prompt sections, dynamic context, workspace instructions, skill
-     * catalog, tools), computed at boot and on every agent swap. `undefined`
-     * while loading or when the snapshot could not be assembled — the startup
-     * panel stays hidden until it lands.
-     */
-    readonly loadedContext: LoadedContext | undefined;
-    /**
-     * Messages submitted while the model was working and not yet claimed by a
-     * turn (`steer` → next step boundary of the running turn, `followup` →
-     * after the turn ends). Driven by agent inbox events.
-     */
-    readonly pending: readonly PendingMessage[];
-    /**
-     * Effective slash commands: built-in locals plus plugin-registered
-     * commands (plan/goal/…) merged from the DSH command registry. The
-     * registry is the source of truth for external names — a plugin shadows
-     * nothing here; locals win on name collisions.
-     */
-    readonly commandList: readonly LocalCommand[];
-    /** Context-aware slash completions, including plugin subcommands. */
-    commandCompletions(input: string): readonly CommandCompletion[];
-    /**
-     * Run a plugin-registered slash command against the live agent (DSH
-     * `dsh-commands` registry): logs `command/run`/`command/done` and returns
-     * the handler's result text. Kept stable for public scene consumers.
-     */
-    runExternalCommand(name: string, rawInput: string, images?: readonly ComposerImageRef[]): Promise<string | undefined>;
-    /** Detailed companion used by draft-owning composers. `undefined` means
-     * the registry no longer has the command, so the draft stays untouched. */
-    runExternalCommandOutcome(name: string, rawInput: string, images?: readonly ComposerImageRef[]): Promise<ExternalCommandOutcome | undefined>;
-    /**
-     * TUI-side permission preset switch: the official `/permission` command
-     * when registered, otherwise the permission-presets service's own write
-     * path (the same handler the command drives). Resolves true when the
-     * durable identity confirms the target. Never falls through to the model.
-     */
-    runPermissionPreset(name: string): Promise<boolean>;
-    /**
-     * Plugin-registered full-screen scene currently replacing the conversation
-     * (the `dsh-tui-scenes` runtime), if any. The chat screen renders its
-     * component INSTEAD of the transcript — the same whole-terminal treatment
-     * the trajectory scene gets — and hands it the keyboard; `undefined`
-     * renders the conversation normally.
-     */
-    readonly pluginScene: TuiSceneDescriptor | undefined;
-    /**
-     * Open a registered plugin scene by id. Plugin command handlers usually
-     * call the runtime directly (`ctx.tuiScenes.open`); this passthrough lets
-     * host-side UI code do the same without touching cordis services.
-     */
-    openPluginScene(id: string): boolean;
-    /** Close the open plugin scene, if any (a no-op otherwise). */
-    closePluginScene(): void;
-    /** 侧问（/btw）：无工具单轮 LLM 调用，复用当前会话上下文；结果不落 session log。 */
-    sideQuestion(question: string, options?: {
-        signal?: AbortSignal;
-        onText?: (delta: string) => void;
-    }): Promise<{
-        answer: string | null;
-        error?: string;
-    }>;
-    /** Estimated context segments by content type (pi-nano-context style bar). */
-    readonly contextSegments: {
-        system: number;
-        prompt: number;
-        assistant: number;
-        thinking: number;
-        tools: number;
-    };
-    /** Active subagents spawned by the current session. */
-    readonly subagents: readonly SubagentState[];
-    /** Native control operations; unavailable providers safely return false. */
-    readonly subagentControl: SubagentControl;
-    /**
-     * Background jobs of the current session (`run_in_background` tool work),
-     * live-tracked from the harness job registry. Empty when the composition
-     * has no jobs service. Drives the `/jobs` panel, transcript job cards and
-     * the status-line chip.
-     */
-    readonly backgroundJobs: readonly BackgroundJobState[];
-    /** Cancellation of a background job with the owning agent's authority. */
-    readonly jobControl: JobControl;
-    subscribe: (listener: () => void) => () => void;
-    /** Current composer generation. Async paste continuations capture this
-     *  before I/O and must not mutate a different session's draft. */
-    stagedImageGeneration(): number;
-    /** Validate and persist an image, returning the historical scene-facing
-     * `[Image #N]` token accepted by submit/steer/registry commands. */
-    stageImage(input: StagedImageInput): Promise<string>;
-    /** Draft-safe composer companion: bind persistence to one session epoch
-     * and return an opaque capability whose visible label belongs to Prompt. */
-    stageComposerImage(input: StagedImageInput, generation: number): Promise<StagedImageHandle>;
-    /** Whether a capability is still live in the current composer session. */
-    hasStagedImage(stageId: string): boolean;
-    /** Revoke a capability that was staged for a draft which no longer exists.
-     * Durable attachment storage remains content-addressed; this only releases
-     * the editable-composer lookup and its preview facade. */
-    discardStagedImage(stageId: string): void;
-    /** The staged image behind one opaque capability, as the same lazily-read
-     *  facade transcript rows use; undefined once evicted or cleared. */
-    stagedImage(stageId: string): TranscriptImage | undefined;
-    /** The profile's image-paste limits, for callers that must bound work
-     *  BEFORE reading bytes (a Finder path is untrusted input). Undefined
-     *  when the composition has no attachment service. */
-    stagedImageLimits(): {
-        readonly maxImageBytes: number;
-        readonly maxImagesPerMessage: number;
-    } | undefined;
-    submit(text: string, images?: readonly ComposerImageRef[]): void;
-    /**
-     * Steer a message into the running turn (Codex/pi semantics): injected at
-     * the next step boundary, the agent continues without aborting.
-     */
-    steer(text: string, images?: readonly ComposerImageRef[]): void;
-    /** Pull a pending message back out of the inbox (Alt+Up) for re-editing. */
-    removePending(id: string): boolean;
-    /** Abort the in-flight turn (`Ctrl+C` while working). While `cancelPending`
-     *  stays true the abort has not converged; Chat force-exits on the next
-     *  Ctrl+C press in that window. */
-    cancel(): void;
-    /** Abort the in-flight turn and process `inputs` right away (Esc/Ctrl+Enter
-     *  with queued input): each message is re-queued as a followup once the abort
-     *  settles, so the new turn starts immediately. Plain strings remain the
-     *  compatibility form for non-composer callers. Returns the count queued. */
-    interruptAndDeliver(inputs: readonly (string | ComposerSubmission)[]): number;
-    /** Rewind the conversation to a past user message (double-Esc rewind):
-     *  forks the session through that message, swaps in a fresh agent, and
-     *  returns the message text for re-editing — or `null` when unwritable.
-     *  `mode` is the plugin-offered rewind mode the user picked (the
-     *  tui/rewind-prompt seam), null for the plain conversation rewind. */
-    rewindTo(row: ChatRow, mode?: string | null): Promise<string | null>;
-    /**
-     * The rewind decision prompt (tui/rewind-prompt event): asked when the
-     * picker confirms a message, before the confirm pane renders. 'cancel'
-     * vetoes the rewind (reason already toasted), `{ modes }` adds plugin
-     * choices to the confirm pane, null means no opinion (plain confirm).
-     */
-    promptRewind(row: ChatRow): Promise<{
-        modes: readonly TuiRewindMode[];
-    } | 'cancel' | null>;
-    /**
-     * The session family tree for the /tree screen (pi's Session Tree): the
-     * live session's whole lineage — ancestors, siblings, descendants —
-     * stitched across fork sessions into one message-level tree. `null` (with
-     * a notify) when session persistence is unavailable or the live session
-     * swapped while the family loaded.
-     */
-    buildSessionTree(): Promise<SessionTreeData | null>;
-    /**
-     * Session-tree fork: `rewind` drops the picked user turn (its prompt comes
-     * back as the returned text), `fork` keeps the picked entry. `seq` is the
-     * tree entry's source event seq inside `sessionId`'s log; `sessionId` may
-     * be any family member (adopting a dead branch forks IT at the picked
-     * point). Null = refused (the channel notified why).
-     */
-    rewindToNode(sessionId: string, seq: number, mode?: 'rewind' | 'fork'): Promise<string | null>;
-    /** `/fork`: fork the current session at its tip into a persisted copy the
-     *  user enters via `/resume` — the live session keeps running untouched. */
-    forkSession(): Promise<boolean>;
-    /** Switch the live agent to a persisted session, replaying its history. */
-    resumeTo(sessionId: string): Promise<ResumeResult>;
-    /** Start a fresh conversation (`/new`): a brand-new agent + session, the
-     *  transcript cleared, the resume marker forgotten. */
-    newSession(): Promise<boolean>;
-    /** Workspace targets contributed by the TUI and optional providers. */
-    listWorkspaces(): Promise<readonly TuiWorkspaceTarget[]>;
-    /** Resolve an absolute path, file URL, or provider URI. */
-    resolveWorkspace(reference: string): Promise<TuiWorkspaceTarget | undefined>;
-    /** Start a fresh session in the selected workspace. */
-    switchWorkspace(target: TuiWorkspaceTarget): Promise<boolean>;
-    /** Rename the current durable workspace. */
-    renameWorkspace(title: string): Promise<boolean>;
-    /** Provider-owned workspace subcommands. */
-    workspaceCommands(): readonly Pick<TuiWorkspaceCommand, 'name' | 'aliases' | 'description'>[];
-    runWorkspaceCommand(name: string, input: string): Promise<TuiWorkspaceCommandResult | undefined>;
-    /** Switch the live model (`/model` picker): forks the conversation at its
-     *  current end and continues it with a new agent routed to `provider`/`model`.
-     *  The history replays unchanged; only the request route changes. */
-    switchModel(provider: string, model: string): Promise<boolean>;
-    /** The live route's effort levels + adapter default for the `/effort`
-     *  slider; empty `efforts` after notifying when unsupported/unavailable. */
-    listEfforts(): Promise<{
-        efforts: readonly EffortOption[];
-        defaultEffort: string | undefined;
-    }>;
-    /** Set one effort level by id (validated against the adapter list);
-     *  false + a notify when the id is not offered. Persists like the old
-     *  Shift+Tab cycle (~/.dsh-tui/effort.json). */
-    setEffort(id: string): Promise<boolean>;
-    /** Re-seat the default reasoning effort new sessions start on (the
-     *  /settings 默认推理强度 field). Unlike {@link Channel.setEffort} it
-     *  does not write effort.json — the settings user layer outranks that
-     *  file — and `undefined` (the field's auto option) re-derives the boot
-     *  chain (cordis.yml `effort` → persisted /effort choice → adapter
-     *  default). The live agent is re-pinned too when its route offers the
-     *  level, so a change lands on its next request. */
-    setDefaultEffort(id: string | undefined): void;
-    /** The session mode currently in force (matched from the session log, or
-     *  the last one Shift+Tab applied). */
-    readonly mode: SessionModeSpec;
-    /** Index of `mode` in the configured cycle; 0 is the unmarked base mode. */
-    readonly modeIndex: number;
-    /** Shift+Tab: advance to the next configured session mode. */
-    cycleMode(): Promise<void>;
-    /** Read the official permission preset roster and current identity. */
-    permissionPresets(): PermissionPresetSnapshot;
-    /** The preset the CURRENT session runs under (issue #8), resolved from its
-     *  log at create/resume time; undefined when no roster is mounted. */
-    readonly agentPreset: string | undefined;
-    /** The roster's presets for the `/preset` picker (empty without a roster). */
-    listPresets(): Promise<readonly PresetOption[]>;
-    /** Switch the agent preset (`/preset`): a blank session swaps composition
-     *  in place (official `recompose` + logged `agent-preset/selected`); a
-     *  started session is locked, so the choice persists as the default for
-     *  future sessions instead. False when the roster is absent, the id is
-     *  unknown/broken, or a turn is running. */
-    switchPreset(presetId: string): Promise<boolean>;
-    /** Reset the visible transcript (`/clear`). */
-    clear(): void;
-    /**
-     * Re-render rows older than the current in-memory window from the session
-     * log (rows beyond {@link ChannelState.rows}' cap are folded away; this
-     * restores them for review). Returns the number of rows restored, 0 when
-     * the whole log is already materialized.
-     */
-    loadOlder(): number;
-    /** Push a transient notification above the prompt input. Returns an
-     *  early-dismiss handle (the auto-timeout still runs as the backstop). */
-    notify(text: string, options?: {
-        color?: NotificationItem['color'];
-        timeoutMs?: number;
-    }): () => void;
-    /** Switch the working-activity indicator preset (`/activity`): validates
-     *  the name, persists it to `~/.dsh-tui/working-activity.json`, and
-     *  re-renders the indicator immediately; false when the name is unknown
-     *  or the preference cannot be written. */
-    setActivityFrames(name: string): boolean;
-    /** Advertised models across every registered provider route (empty when the LLM service is absent). */
-    listModels(): Promise<readonly LlmModelInfo[]>;
-    /** Provider display identities for the same routes (picker group labels). */
-    listProviders(): Promise<readonly LlmProviderInfo[]>;
-    /** Drop the `/model <provider/id>` completion cache so the next `/model `
-     *  refetch reflects a provider-catalog change (`/provider` add/edit/delete,
-     *  OAuth sign-in/out) — the same consistency the picker's per-open refetch
-     *  already provides. */
-    invalidateModelCompletion(): void;
-    /** The live agent's full skill catalog for `/skills` (issue #204) — name,
-     *  description, invocation flags and source bucket. Undefined on a failed
-     *  or incomplete registry read (the picker shows an error); empty only
-     *  when no registry is mounted or it genuinely holds nothing. */
-    listSkills(): Promise<readonly SkillInfo[] | undefined>;
-    /** Safe credential metadata for `/login`; undefined without the service. */
-    describeCredential(ref: string): Promise<CredentialStatus | undefined>;
-    /** DeepSeek official account balance for `/balance`: resolves
-     *  `DEEPSEEK_API_KEY` through the credentials seam (env fallback) and
-     *  queries the official balance endpoint. The key is used only for the
-     *  request header — never logged, printed or persisted. */
-    balanceInfo(): Promise<BalanceResult>;
-    /** Runtime capabilities for the `/provider` wizard, over the settings /
-     *  credentials / llm seams; undefined when the composition lacks them
-     *  (bare cordis.yml start without the dsh-base services). */
-    providerSetup(): ProviderSetupHost | undefined;
-    /** OAuth sign-in states from a mounted dsh-auth-style plugin; undefined
-     *  without the plugin, so `/login` renders exactly what it did before. */
-    oauthProviderStatuses(): Promise<readonly OAuthProviderStatus[] | undefined>;
-    /**
-     * Runtime capabilities for the `/settings` screen, over the settings /
-     * credentials seams; undefined when the composition lacks the settings
-     * service (the screen then renders plugin sections as unavailable and
-     * namespaces read-only).
-     */
-    settingsHost(): SettingsHost | undefined;
-    /** Plugin-declared settings sections from the `tuiSettingsSections` seam
-     *  (empty when the seam or every provider is absent). */
-    settingsSections(): readonly TuiSettingsSection[];
-    /** Subscribe to settings-section register/unregister events. */
-    subscribeSettingsSections(listener: () => void): () => void;
-    /** Structured `@` file completion, using the session's remote fs service. */
-    listFileCandidates(query: string, options?: {
-        signal?: AbortSignal;
-        topK?: number;
-    }): Promise<readonly FileCandidate[]>;
-    /** Backward-compatible top-level/recursive listing. */
-    listFiles(): Promise<readonly string[]>;
-    /** Every session the persistence backend stores, classified and unfiltered
-     *  — the browser (`/resume`) decides which of them a given view shows. */
-    listSessions(): Promise<readonly SessionSummary[]>;
-    /** Trailing exchanges of a persisted session, for the browser's preview. */
-    previewSession(sessionId: string): Promise<readonly PreviewEntry[]>;
-    /** Mark a session for `dsh-tui --resume` on the next launch. */
-    setResumeTarget(sessionId: string): void;
-    /** Rename the current session (/rename): appends a `session/title`
-     *  event, which the status line and the /resume picker both read. */
-    renameSession(title: string): void;
-    /** Set the current session's accent color (`/color <name>`): appends a
-     *  `session/color` event; '' clears it back to the theme default. */
-    setSessionColor(color: string): void;
-    /** Generate a recap of the session's recent activity (`/recap`): one
-     *  tool-less LLM call over the tail exchanges, returning a one-line
-     *  summary plus an optional proposed title. The answer is pure UI state
-     *  and never enters the session log. */
-    recapRecent(options?: {
-        signal?: AbortSignal;
-        onText?: (delta: string) => void;
-    }): Promise<RecapOutcome>;
-    /** Delete a persisted session (`/resume` picker ctrl+d): removes its log
-     *  directory, its last-used entry, and the resume marker when it points
-     *  here. False for the live session or a missing/unwritable log. */
-    deleteSession(sessionId: string): Promise<boolean>;
-    /** Rename any persisted session (`/resume` picker ctrl+r): appends a
-     *  `session/title` event to its log (live sessions go through the normal
-     *  rename path). False when the log is absent or undecodable. */
-    renameSessionTo(sessionId: string, title: string): Promise<boolean>;
-    /** Manually compact the session history (/compact); no-op notify when the leaf lacks a compaction service. */
-    compact(): void;
-    /** Render a multi-line local report in the transcript (`/status`,
-     *  `/doctor`, …): a `local` row plus one `local-output` row per line. */
-    pushLocal(title: string, lines: readonly string[]): void;
-    /** MCP server/tool status for /mcp: one line per server, or setup guidance. */
-    mcpStatus(): string[];
-    /** Write the conversation transcript to `dsh-tui-export-<ts>.md` in the
-     *  session cwd; returns the written path, or null on failure. */
-    exportSession(): string | null;
-    /** Create `AGENTS.md` in the session cwd (DSH workspace-context file);
-     *  returns the path, `'exists'` when already present, or null on failure. */
-    initWorkspace(): string | null;
-    /** Environment diagnostics for `/doctor`. */
-    doctorInfo(): string[];
-    /** Plugin contract/grant/ledger diagnostics for `/plugins` (C-070 trust
-     *  banner first line; `check <path>` runs validatePlugin + negotiate). */
-    pluginsInfo(args: string): string[];
-    /** Subagent rows for `/agents` (DSH subagent service; empty message when
-     *  the service is absent). */
-    listSubagents(): Promise<string[]>;
-    /**
-     * The agent view row snapshot: every live agent in
-     * this process plus every persisted session that no live agent owns,
-     * ordered needs-input/working first, then most recently active. Reading it
-     * is cheap; subscribe for changes.
-     */
-    agentViewRows(): readonly AgentViewRow[];
-    /** Change feed for {@link agentViewRows}: fired on agent lifecycle/status
-     *  changes and — throttled — on session events of background agents. */
-    subscribeAgentView(listener: () => void): () => void;
-    /**
-     * Dispatch a new background session (`agent view` input): creates an agent
-     * in this process, delivers the prompt as a user message, and keeps the
-     * TUI attached to its current session. The new session keeps running until
-     * it finishes its turn or is stopped — it lives only while this process
-     * does.
-     */
-    dispatchBackgroundAgent(prompt: string): Promise<AgentViewDispatchResult>;
-    /** Stop a background session (Ctrl+X): abort its turn and dispose its
-     *  agent; the persisted log survives for resume. False for the attached
-     *  session or one this TUI does not own. */
-    stopBackgroundAgent(sessionId: string): Promise<boolean>;
-    /**
-     * Attach the TUI terminal to a session (`agent view` Enter/→): a live
-     * agent is adopted in place (its handle becomes the channel's), a
-     * persisted one resumes through the persistence seam. The previously
-     * attached agent is NOT disposed — it keeps running as a background
-     * session unless it was already idle with no history.
-     */
-    attachToAgent(sessionId: string): Promise<ResumeResult>;
-    /** Trailing exchanges of any session — the live agent's in-memory log when
-     *  it is alive in this process, the persisted artifact otherwise. */
-    peekAgentSession(sessionId: string): Promise<readonly PreviewEntry[]>;
-    /** `/bg` — background the attached session: swap the TUI to a fresh agent
-     *  while the current one keeps running. The agent view lists it as a
-     *  background session; `backgroundedSessionId` is the move's return target
-     *  (Esc returns to that conversation). */
-    backgroundCurrent(): Promise<BackgroundResult>;
-    /** Send a follow-up user message to a session from the agent view's peek
-     *  panel. Live sessions receive it directly; a session no live agent owns
-     *  cannot take a reply (false + a notify to attach instead). */
-    replyToAgent(sessionId: string, text: string): Promise<boolean>;
-    /**
-     * Dispose the host-registry entries this channel registered (skill slash
-     * commands).
-     *
-     * `commandService.register` binds the registration to ITS own context, not
-     * the caller's, so the entries outlive this channel unless released: after a
-     * launcher recompose the stale registrations would still answer, but the
-     * fresh channel would see the names taken and stop managing them, freezing
-     * the menu. The plugin calls this from its teardown effect, where the real
-     * cordis context lives.
-     */
-    releaseContributions(): void;
-    /**
-     * The live agent's session event log (immutable snapshot, replaced on
-     * every append — dsh-session caches the frozen array) — the `/trace`
-     * trajectory view's data source. Screens already re-render on `version`
-     * bumps, so a view reading this per render follows live events in real
-     * time; agent swaps (/resume /rewind /new) are reflected immediately.
-     */
-    traceEvents(): readonly SessionEvent[];
-}
-/** @internal */
-/** One roster entry in the `/preset` picker (see {@link Channel.listPresets}). */
-export interface PresetOption {
-    id: string;
-    name?: string;
-    description?: string;
-    /** Present when the roster marked this preset unloadable (shown verbatim). */
-    broken?: string;
-    isDefault: boolean;
-}
-export type PermissionPresetAvailability = 'runtime' | 'legacy' | 'unavailable';
-export interface PermissionPresetOption {
-    readonly value: string;
-    readonly name: string;
-    readonly description?: string;
-}
-export interface PermissionPresetCurrent {
-    readonly value: string;
-    readonly name: string;
-    readonly description?: string;
-    readonly kind: 'preset' | 'custom';
-}
-/**
- * Adapter-owned permission roster snapshot. `options` never contains the
- * official `custom` sentinel; it is represented only by `current`.
- */
-export interface PermissionPresetSnapshot {
-    readonly availability: PermissionPresetAvailability;
-    readonly options: readonly PermissionPresetOption[];
-    readonly current?: PermissionPresetCurrent;
-}
-/** @internal */
-/** One user message submitted while the model was working, not yet claimed
- *  by a turn. `steer` lands at the next step boundary of the running turn;
- *  `followup` waits for the turn to end. */
-export interface PendingMessage {
-    id: string;
-    text: string;
-    images: readonly ComposerImageRef[];
-    placement: 'steer' | 'followup';
-}
-/**
- * Mutable channel state owned by {@link createChannel}: the screen's
- * reactive store. Screens subscribe and re-render on `version` bumps; the
- * fields mirror the public {@link Channel} contract, and the `@internal`
- * emit hooks belong to the implementation.
- */
-/** One adapter-owned reasoning-effort level for the `/effort` slider. */
-export interface EffortOption {
-    id: string;
-    name: string;
-    description?: string;
-}
-export interface ChannelState {
-    version: number;
-    rows: ChatRow[];
-    status: AgentStatus | 'starting' | 'disposed';
-    sessionTitle: string;
-    sessionColor: string;
-    autoRecapOnOpen: boolean;
-    agentId: string;
-    /** TUI-owned generation that changes on every live Agent rebind. */
-    agentBindingGeneration: number;
-    model: string;
-    provider: string;
-    tokens: TokenUsage;
-    cwd: string;
-    displayCwd: string;
-    gitBranch: string | undefined;
-    working: boolean;
-    /** Whether a requested abort is still converging (see the public Channel type). */
-    cancelPending: boolean;
-    spinnerMode: SpinnerMode;
-    responseChars: number;
-    activeToolCount: number;
-    turnStart: number;
-    lastUserText: string;
-    notifications: NotificationItem[];
-    /** Adapter-advertised context capacity for the model route, when known. */
-    contextWindow: number | undefined;
-    /** Reasoning effort of the latest request header, when the adapter sets one. */
-    reasoningEffort: string | undefined;
-    /** The live route's reasoning-effort level ids, low → high. */
-    effortLevels: readonly string[] | undefined;
-    /** Usage of the most recent request (context share + cache hits). */
-    lastUsage: {
-        input: number;
-        output: number;
-        cacheRead: number;
-        cacheWrite: number;
-    } | undefined;
-    /** Output tokens per second of the current/last turn's response, when known. */
-    tps: number | undefined;
-    /** Per-turn tps samples (sparkline history), oldest first. */
-    tpsSamples: {
-        tps: number;
-        at: number;
-    }[];
-    /** Latest working-activity snapshot (see the public Channel type). */
-    workingActivity: ActivityStatus | undefined;
-    /** Working-activity indicator preset (see the public Channel type). */
-    activityFrames: string | undefined;
-    /** Raw cordis.yml pins `/reload` must respect (see the public Channel type). */
-    configuredProvider: string | undefined;
-    configuredModel: string | undefined;
-    configuredPreset: string | undefined;
-    configuredActivityFrames: string | undefined;
-    configuredLang: string | undefined;
-    /** Diff presentation preference (see the public Channel type). */
-    diffLayout: 'auto' | 'split' | 'unified';
-    /** Thinking-block display (see the public Channel type). */
-    thinkingFold: 'preview' | 'full';
-    /** Tool-card background treatment (see the public Channel type). */
-    toolBackground: ToolBackground;
-    /** Transcript gutter mode (see the public Channel type). */
-    scrollGutter: ScrollGutterMode;
-    /** Root page inset setting (see the public Channel type). */
-    pageMargin: PageMarginSetting;
-    /** Terminal-card header folding (see the public Channel type). */
-    foldTerminalCommand: boolean;
-    /** Session-name chip on the prompt border (see the public Channel type). */
-    promptSessionLabel: boolean;
-    /** Fullscreen draft editor gate (see the public Channel type). */
-    expandEditor: boolean;
-    /** Smooth streaming reveal (see the public Channel type). */
-    smoothStreaming: boolean;
-    /** Status-footer preferences (see the public Channel type). */
-    statusBar: StatusBarConfig;
-    /** Apply a diff-layout change (see the public Channel type). */
-    setDiffLayout(layout: 'auto' | 'split' | 'unified'): void;
-    /** Apply a thinking-display change (see the public Channel type). */
-    setThinkingFold(mode: 'preview' | 'full'): void;
-    /** Apply a tool-card background change. */
-    setToolBackground(background: ToolBackground): void;
-    /** Apply a transcript gutter mode change. */
-    setScrollGutter(mode: ScrollGutterMode): void;
-    /** Apply a root page-inset setting change (drives the PageMargin box). */
-    setPageMargin(setting: PageMarginSetting): void;
-    /** Apply a terminal-card header folding change. */
-    setFoldTerminalCommand(enabled: boolean): void;
-    /** Apply a prompt session-name chip change. */
-    setPromptSessionLabel(enabled: boolean): void;
-    /** Apply a fullscreen-editor gate change. */
-    setExpandEditor(enabled: boolean): void;
-    /** Apply a smooth-streaming reveal change. */
-    setSmoothStreaming(enabled: boolean): void;
-    /** Apply status-footer preference changes. */
-    setStatusBar(config: Partial<StatusBarConfig>): void;
-    /** Whale header art switch (see the public Channel type). */
-    whale: boolean;
-    /** Idle whale behaviors switch (see the public Channel type). */
-    whaleIdle: boolean;
-    /** Apply a whale-visibility change (see the public Channel type). */
-    setWhale(visible: boolean): void;
-    /** Apply an idle-whale-behavior change (see the public Channel type). */
-    setWhaleIdle(enabled: boolean): void;
-    minimal: boolean;
-    /** Apply a minimal-mode change (see the public Channel type). */
-    setMinimal(enabled: boolean): void;
-    /** Working-activity display switch (see the public Channel type). */
-    activityEnabled: boolean;
-    /** Context bar row switch (see the public Channel type). */
-    contextBarEnabled: boolean;
-    /** Current same-session goal projection (see the public Channel type). */
-    goal: ChannelGoal | undefined;
-    /** Latest todo-list snapshot (see the public Channel type). */
-    todos: TodoPanelItem[];
-    /** Loaded-context snapshot (see the public Channel type). */
-    loadedContext: LoadedContext | undefined;
-    /** Messages submitted while working, awaiting their turn/step boundary.
-     *  Driven by agent inbox events (inserted/claimed/discarded). */
-    pending: PendingMessage[];
-    /** 侧问（见 public Channel.sideQuestion）。 */
-    sideQuestion(question: string, options?: {
-        signal?: AbortSignal;
-        onText?: (delta: string) => void;
-    }): Promise<{
-        answer: string | null;
-        error?: string;
-    }>;
-    /** 会话 recap（见 public Channel.recapRecent）。 */
-    recapRecent(options?: {
-        signal?: AbortSignal;
-        onText?: (delta: string) => void;
-    }): Promise<RecapOutcome>;
-    /** 会话强调色（见 public Channel.setSessionColor）。 */
-    setSessionColor(color: string): void;
-    /** Effective slash commands (see the public Channel type). */
-    commandList: readonly LocalCommand[];
-    /** Context-aware slash completions (see the public Channel type). */
-    commandCompletions(input: string): readonly CommandCompletion[];
-    /** Run a plugin-registered command (see the public Channel type). */
-    runExternalCommand(name: string, rawInput: string, images?: readonly ComposerImageRef[]): Promise<string | undefined>;
-    /** Run a command while retaining its draft-consumption outcome. */
-    runExternalCommandOutcome(name: string, rawInput: string, images?: readonly ComposerImageRef[]): Promise<ExternalCommandOutcome | undefined>;
-    /** TUI-side permission preset switch (see the public Channel type). */
-    runPermissionPreset(name: string): Promise<boolean>;
-    /** Open plugin scene mirrored from the scenes runtime (see the public Channel type). */
-    pluginScene: TuiSceneDescriptor | undefined;
-    /** Open a plugin scene by id (see the public Channel type). */
-    openPluginScene(id: string): boolean;
-    /** Close the open plugin scene (see the public Channel type). */
-    closePluginScene(): void;
-    /** Estimated context segments by content type (pi-nano-context style bar). */
-    contextSegments: {
-        system: number;
-        prompt: number;
-        assistant: number;
-        thinking: number;
-        tools: number;
-    };
-    /** Active subagents roster (see the public Channel type). */
-    subagents: readonly SubagentState[];
-    subagentControl: SubagentControl;
-    /** Background jobs of the current session (see the public Channel type). */
-    backgroundJobs: readonly BackgroundJobState[];
-    jobControl: JobControl;
-    subscribe: (listener: () => void) => () => void;
-    stagedImageGeneration(): number;
-    stageImage(input: StagedImageInput): Promise<string>;
-    stageComposerImage(input: StagedImageInput, generation: number): Promise<StagedImageHandle>;
-    hasStagedImage(stageId: string): boolean;
-    discardStagedImage(stageId: string): void;
-    stagedImage(stageId: string): TranscriptImage | undefined;
-    stagedImageLimits(): {
-        readonly maxImageBytes: number;
-        readonly maxImagesPerMessage: number;
-    } | undefined;
-    /** @internal event bump (the public `notify(text)` posts a notification). */
-    emit(): void;
-    /** @internal frame-aligned emit for high-frequency streaming deltas:
-     *  version bumps synchronously but listeners fire at most once per 16ms
-     *  window (trailing edge). */
-    emitStream(): void;
-    submit(text: string, images?: readonly ComposerImageRef[]): void;
-    steer(text: string, images?: readonly ComposerImageRef[]): void;
-    removePending(id: string): boolean;
-    cancel(): void;
-    /** @internal interrupt-and-deliver (see the public Channel type). */
-    interruptAndDeliver(inputs: readonly (string | ComposerSubmission)[]): number;
-    rewindTo(row: ChatRow, mode?: string | null): Promise<string | null>;
-    /** @internal rewind decision prompt (see the public Channel.promptRewind). */
-    promptRewind(row: ChatRow): Promise<{
-        modes: readonly TuiRewindMode[];
-    } | 'cancel' | null>;
-    /** @internal session-tree assembly (see the public Channel.buildSessionTree). */
-    buildSessionTree(): Promise<SessionTreeData | null>;
-    /** @internal tree-entry rewind/fork (see the public Channel.rewindToNode). */
-    rewindToNode(sessionId: string, seq: number, mode?: 'rewind' | 'fork'): Promise<string | null>;
-    /** @internal tip fork (see the public Channel.forkSession). */
-    forkSession(): Promise<boolean>;
-    /** Switch the live agent to a persisted session, replaying its history. */
-    resumeTo(sessionId: string): Promise<ResumeResult>;
-    /** Start a fresh conversation (`/new`). */
-    newSession(): Promise<boolean>;
-    listWorkspaces(): Promise<readonly TuiWorkspaceTarget[]>;
-    resolveWorkspace(uri: string): Promise<TuiWorkspaceTarget | undefined>;
-    switchWorkspace(target: TuiWorkspaceTarget): Promise<boolean>;
-    renameWorkspace(title: string): Promise<boolean>;
-    workspaceCommands(): readonly Pick<TuiWorkspaceCommand, 'name' | 'aliases' | 'description'>[];
-    runWorkspaceCommand(name: string, input: string): Promise<TuiWorkspaceCommandResult | undefined>;
-    /** Switch the live model (`/model` picker). */
-    switchModel(provider: string, model: string): Promise<boolean>;
-    /** The route's effort levels for `/effort` (see the public Channel type). */
-    listEfforts(): Promise<{
-        efforts: readonly EffortOption[];
-        defaultEffort: string | undefined;
-    }>;
-    /** Set one effort level by id (see the public Channel type). */
-    setEffort(id: string): Promise<boolean>;
-    /** Re-seat the future-sessions default reasoning effort (see the public
-     *  Channel.setDefaultEffort). */
-    setDefaultEffort(id: string | undefined): void;
-    /** The session mode currently in force (see the public Channel type). */
-    mode: SessionModeSpec;
-    /** Index of `mode` in the configured cycle (see the public Channel type). */
-    modeIndex: number;
-    /** Shift+Tab session-mode advance (see the public Channel type). */
-    cycleMode(): Promise<void>;
-    /** Read the official permission preset roster and current identity. */
-    permissionPresets(): PermissionPresetSnapshot;
-    /** The preset the current session runs under (see the public Channel type). */
-    agentPreset: string | undefined;
-    /** The roster's presets for the `/preset` picker (see the public Channel type). */
-    listPresets(): Promise<readonly PresetOption[]>;
-    /** Switch the agent preset (see the public Channel type). */
-    switchPreset(presetId: string): Promise<boolean>;
-    clear(): void;
-    /** @internal older-row restoration (see the public Channel.loadOlder). */
-    loadOlder(): number;
-    notify(text: string, options?: {
-        color?: NotificationItem['color'];
-        timeoutMs?: number;
-    }): () => void;
-    /** Switch the working-activity indicator preset (see the public Channel). */
-    setActivityFrames(name: string): boolean;
-    listModels(): Promise<readonly LlmModelInfo[]>;
-    /** Provider display identities (see the public Channel type). */
-    listProviders(): Promise<readonly LlmProviderInfo[]>;
-    /** Drop the `/model` completion cache (see the public Channel type). */
-    invalidateModelCompletion(): void;
-    /** The live agent's skill catalog for `/skills` (see the public Channel type). */
-    listSkills(): Promise<readonly SkillInfo[] | undefined>;
-    /** Safe credential metadata for `/login` (see the public Channel type). */
-    describeCredential(ref: string): Promise<CredentialStatus | undefined>;
-    /** DeepSeek official balance for `/balance` (see the public Channel type). */
-    balanceInfo(): Promise<BalanceResult>;
-    /** `/provider` wizard capabilities (see the public Channel type). */
-    providerSetup(): ProviderSetupHost | undefined;
-    /** OAuth sign-in states (see the public Channel type). */
-    oauthProviderStatuses(): Promise<readonly OAuthProviderStatus[] | undefined>;
-    /** `/settings` screen capabilities (see the public Channel type). */
-    settingsHost(): SettingsHost | undefined;
-    /** Plugin-declared settings sections (see the public Channel type). */
-    settingsSections(): readonly TuiSettingsSection[];
-    /** Subscribe to settings-section register/unregister events. */
-    subscribeSettingsSections(listener: () => void): () => void;
-    listFileCandidates(query: string, options?: {
-        signal?: AbortSignal;
-        topK?: number;
-    }): Promise<readonly FileCandidate[]>;
-    listFiles(): Promise<readonly string[]>;
-    listSessions(): Promise<readonly SessionSummary[]>;
-    /** Trailing exchanges of a persisted session (see the public Channel type). */
-    previewSession(sessionId: string): Promise<readonly PreviewEntry[]>;
-    setResumeTarget(sessionId: string): void;
-    /** Rename the current session (see the public Channel type). */
-    renameSession(title: string): void;
-    /** Delete a persisted session (see the public Channel type). */
-    deleteSession(sessionId: string): Promise<boolean>;
-    /** Rename any persisted session (see the public Channel type). */
-    renameSessionTo(sessionId: string, title: string): Promise<boolean>;
-    /** Manually compact the session history (/compact). */
-    compact(): void;
-    /** Multi-line local report (`/status`, `/doctor`, …). */
-    pushLocal(title: string, lines: readonly string[]): void;
-    /** MCP server/tool status for /mcp: one line per server, or setup guidance. */
-    mcpStatus(): string[];
-    /** Export the transcript to a markdown file (/export). */
-    exportSession(): string | null;
-    /** Create `AGENTS.md` in the session cwd (/init). */
-    initWorkspace(): string | null;
-    /** Environment diagnostics (/doctor). */
-    doctorInfo(): string[];
-    /** Plugin diagnostics (/plugins); see the public Channel type. */
-    pluginsInfo(args: string): string[];
-    /** Subagent rows (/agents). */
-    listSubagents(): Promise<string[]>;
-    /** See {@link Channel.agentViewRows}. */
-    agentViewRows(): readonly AgentViewRow[];
-    /** See {@link Channel.subscribeAgentView}. */
-    subscribeAgentView(listener: () => void): () => void;
-    /** See {@link Channel.dispatchBackgroundAgent}. */
-    dispatchBackgroundAgent(prompt: string): Promise<AgentViewDispatchResult>;
-    /** See {@link Channel.stopBackgroundAgent}. */
-    stopBackgroundAgent(sessionId: string): Promise<boolean>;
-    /** See {@link Channel.attachToAgent}. */
-    attachToAgent(sessionId: string): Promise<ResumeResult>;
-    /** See {@link Channel.peekAgentSession}. */
-    peekAgentSession(sessionId: string): Promise<readonly PreviewEntry[]>;
-    /** See {@link Channel.backgroundCurrent}. */
-    backgroundCurrent(): Promise<BackgroundResult>;
-    /** See {@link Channel.replyToAgent}. */
-    replyToAgent(sessionId: string, text: string): Promise<boolean>;
-    /**
-     * Bind the plugin's approval store (post-construction): row derivation
-     * reads its parked ask ids for the "needs input" state, and its emits
-     * re-publish as agent-view changes.
-     */
-    bindApprovalStore(store: {
-        pendingAgentIds(): readonly string[];
-        pendingAgentDetail(agentId: string): {
-            toolName: string;
-            reason?: string;
-            command?: string;
-        } | undefined;
-        subscribe(listener: () => void): () => void;
-    }): void;
-    /** See {@link Channel.releaseContributions}. */
-    releaseContributions(): void;
-    /** Live session event log (see the public Channel type, `/trace`). */
-    traceEvents(): readonly SessionEvent[];
-}
 /**
  * Create the live channel state for one agent session: replay the durable
  * transcript, subscribe to the agent's events, and expose every TUI action.
@@ -1386,154 +14,10 @@ export interface ChannelState {
  *   reasoning-effort / working-activity / agent-handle preferences.
  * @returns The live channel state, subscribed and ready to render.
  */
-export declare function createChannel(ctx: Context, initialAgent: Agent, options: {
-    model: string;
-    cwd: string;
-    provider: string;
-    /** Configured reasoning effort: applied to the agent's requests when the
-     *  live route offers it (silently ignored otherwise), and shown from
-     *  startup until the first request/header event reports the adapter's
-     *  live value. */
-    effort?: string;
-    /** Derive the working line from base session events; default on. */
-    activity?: boolean;
-    /** Indicator preset for the working-activity line (`moon8`/`moon`/
-     *  `comet`/`dots`/… or `random`); default `moon8`. */
-    activityFrames?: string;
-    /** Edit/Write diff presentation; default `auto` (side-by-side ≥110
-     *  columns, unified below). */
-    diffLayout?: 'auto' | 'split' | 'unified';
-    /** Thinking-block display; default `preview` (2-3 line live preview,
-     *  fold per step) — `full` keeps thinking expanded until turn end. */
-    thinkingFold?: 'preview' | 'full';
-    /** Tool-card background treatment; default `none`. */
-    toolBackground?: ToolBackground;
-    /** Transcript gutter mode; default `timeline` (settings `dsh-tui.scrollGutter`). */
-    scrollGutter?: ScrollGutterMode;
-    /** Root page inset setting; default `normal` (settings `dsh-tui.pageMargin`). */
-    pageMargin?: PageMarginSetting;
-    /** Terminal-card header folding; default off (settings
-     *  `dsh-tui.foldTerminalCommand`). */
-    foldTerminalCommand?: boolean;
-    /** Session-name chip on the prompt top border; default off (settings
-     *  `dsh-tui.promptSessionLabel`). */
-    promptSessionLabel?: boolean;
-    /** Fullscreen draft editor entry points; default on (settings
-     *  `dsh-tui.expandEditor`). */
-    expandEditor?: boolean;
-    /** Smooth streaming reveal; default on (settings
-     *  `dsh-tui.smoothStreaming`). */
-    smoothStreaming?: boolean;
-    /** Status-footer field visibility and compactness. */
-    statusBar?: Partial<StatusBarConfig>;
-    /** Show the header's pixel whale art; default on. */
-    whale?: boolean;
-    /** Idle whale behaviors (fin/tail/sleep) after the intro settles;
-     * default off (click-hearts are always available). */
-    whaleIdle?: boolean;
-    /** Minimal mode; default off (settings `dsh-tui.minimal`). */
-    minimal?: boolean;
-    /** Show the segmented context bar row in the status footer; default on
-     *  (cordis.yml `contextBar: false` hides it, issue #29). */
-    contextBar?: boolean;
-    /** cordis.yml's static preset choice (`preset` key): wins over the
-     *  persisted `/preset` preference for NEW sessions this channel starts. */
-    configuredPreset?: string;
-    /** cordis.yml's static route (`provider`/`model` keys), undefined when
-     *  unset: wins over the persisted `/model` preference for NEW sessions
-     *  only when BOTH halves are pinned (atomic rule, issue #67), and is the
-     *  only route a resume overrides the target's own record with. */
-    configuredProvider?: string;
-    configuredModel?: string;
-    /** cordis.yml's raw `lang` key, undefined when unset: `/reload` consults
-     *  it so a static deployment choice is never overridden by lang.json. */
-    configuredLang?: string;
-    /** cordis.yml's raw `activityFrames` key, undefined when unset: the
-     *  static choice `/reload` must not override. */
-    configuredActivityFrames?: string;
-    /** The preset the initial agent's session runs under (from resolveAgent). */
-    agentPreset?: string;
-    /** Shift+Tab session-mode cycle from cordis.yml `modes`; undefined →
-     *  the built-in default/plan/full cycle (sessionModes.ts). */
-    modes?: readonly SessionModeSpec[];
-    /** Handle of the initial agent; disposed when a rewind replaces it. */
-    handle?: AgentHandle;
-}): ChannelState;
-/**
- * `/resume` project filter (issue #96): exact cwd match, PLUS sessions
- * recorded in a subdirectory — pre-upgrade launches recorded the launch
- * subdirectory as the header cwd, and with the cwd default now resolving to
- * the git worktree root an exact match would hide those sessions forever.
- * They belong to the same workspace, so they stay listed. Comparison follows
- * the platform's filesystem semantics (case-insensitive on Windows — a
- * pre-upgrade header may record `C:\Repo` where the current launch resolves
- * `c:\repo`). `caseInsensitive` is a parameter (not a platform read) so the
- * verifier can exercise both modes on any host. Exported for
- * scripts/verify-session-cwd.mjs.
- *
- * Boundary rule (issue #153): container directories are nobody's workspace.
- * $HOME and the Windows root forms — plain drive roots (`C:`), UNC share
- * roots (`//server/share`), and extended-length roots (`//?/C:`,
- * `//?/UNC/server/share`) — are ancestors of unrelated projects, so the
- * descendant rules below would list every session on the machine from `~`
- * (and every session on the drive/share from those roots). At these
- * boundaries, in either direction, only an exact match passes.
- */
-export declare function sessionCwdMatches(stateCwd: string, headerCwd: string, caseInsensitive?: boolean): boolean;
-/** The fs-service surface `@`-mention expansion consumes (dsh-fs-local). */
-export interface MentionFs {
-    resolve(path: string): Promise<{
-        displayPath: string;
-    }>;
-    stat(target: {
-        displayPath: string;
-    }): Promise<{
-        type: 'file' | 'directory' | 'other';
-    } | undefined>;
-    readText(target: {
-        displayPath: string;
-    }): Promise<string>;
-    readBytes?(target: {
-        displayPath: string;
-    }, signal: AbortSignal | undefined, maxBytes: number): Promise<Uint8Array>;
-    listDir(target: {
-        displayPath: string;
-    }): Promise<Array<{
-        name: string;
-        type: 'file' | 'directory' | 'other';
-    }>>;
-}
-type MentionImageBlock = ChannelImageBlock;
-type MentionImageMediaType = ChannelImageMediaType;
-/** Attachment subset used to turn an image path into a durable user block. */
-export interface MentionAttachments {
-    readonly imageLimits: {
-        readonly maxImageBytes: number;
-        readonly maxImagesPerMessage: number;
-        readonly maxMessageImageBytes: number;
-        readonly mediaTypes: readonly MentionImageMediaType[];
-    };
-    saveImage(input: {
-        data: Uint8Array;
-        mediaType: MentionImageMediaType;
-        name?: string;
-    }): Promise<MentionImageBlock['attachment']>;
-}
-export interface MentionExpansion {
-    /** Model-facing blocks: the typed text first, one block per attachment. */
-    blocks: ContentBlock[];
-    /** Paths that resolved and were attached (for the confirmation notice). */
-    attached: string[];
-    /** Mention tokens that failed to resolve (kept literal, warned about). */
-    missing: string[];
-}
-/**
- * Expand a submitted text's `@` mentions (issue #15) into model-facing
- * attachment blocks: supported image files become durable image blocks,
- * other files contribute capped text, and directories contribute a shallow
- * listing. Reads always go through the active fs service, so provider-owned
- * workspaces keep their routing semantics. The typed text stays first and
- * verbatim. Best-effort failures degrade to `missing`, never a failed send.
- */
-export declare function expandMentions(fs: MentionFs | undefined, cwd: string, text: string, attachments?: MentionAttachments, stagedImages?: ReadonlyMap<string, MentionImageBlock['attachment']>): Promise<MentionExpansion>;
+export declare function createChannel(ctx: Context, initialAgent: Agent, options: ChannelLaunchOptions): ChannelState;
+export type { ChannelLaunchOptions } from './channel/state.js';
+export { expandMentions } from './channel/mentions.js';
+export { sessionCwdMatches } from './channel/paths.js';
+export type { ActivityStatus, AgentViewDispatchResult, AgentViewRow, AgentViewStatus, BackgroundResult, Channel, ChannelGoal, ChannelState, ChatRow, ComposerImageRef, ComposerSubmission, CredentialStatus, EffortOption, ExternalCommandOutcome, JobControl, JobRow, LoadedContext, LoadedContextEntry, LoadedContextFile, LoadedContextSkill, LoadedContextTool, MentionAttachments, MentionExpansion, MentionFs, NotificationItem, PendingMessage, PermissionPresetAvailability, PermissionPresetCurrent, PermissionPresetOption, PermissionPresetSnapshot, PresetOption, ResumeResult, SkillInfo, StagedImageHandle, StagedImageInput, SubagentControl, SubagentRow, TodoPanelItem, TokenBucket, TokenUsage, ToolCallView, ToolFileDiff, ToolResultView, ToolRow, ToolViewPresenter, TranscriptImage } from './channel/types.js';
+export { emptyTokenUsage } from './channel/usage.js';
 //# sourceMappingURL=channel.d.ts.map