@anthropic-ai/claude-agent-sdk 0.3.210 → 0.3.212

package/sdk.d.ts

@@ -1463,7 +1463,7 @@ export declare type Options = {
      * Enable beta features. Currently supported:
      * - `'context-1m-2025-08-07'` - Enable 1M token context window (Sonnet 4/4.5 only)
      *
-     * @see https://docs.anthropic.com/en/api/beta-headers
+     * @see https://platform.claude.com/docs/en/api/beta-headers
      */
     betas?: SdkBeta[];
     /**
@@ -1607,7 +1607,7 @@ export declare type Options = {
      *
      * When set, takes precedence over the deprecated `maxThinkingTokens`.
      *
-     * @see https://docs.anthropic.com/en/docs/build-with-claude/adaptive-thinking
+     * @see https://platform.claude.com/docs/en/build-with-claude/adaptive-thinking
      */
     thinking?: ThinkingConfig;
     /**
@@ -1620,7 +1620,7 @@ export declare type Options = {
      * - `'xhigh'` — Deeper than high (Fable 5, Opus 4.7+, Sonnet 5)
      * - `'max'` — Maximum effort (Fable 5, Opus 4.6+, Sonnet 4.6+)
      *
-     * @see https://docs.anthropic.com/en/docs/build-with-claude/effort
+     * @see https://platform.claude.com/docs/en/build-with-claude/effort
      */
     effort?: EffortLevel;
     /**
@@ -1813,7 +1813,7 @@ export declare type Options = {
      * }
      * ```
      *
-     * @see https://docs.anthropic.com/en/docs/claude-code/settings#sandbox-settings
+     * @see https://code.claude.com/docs/en/settings#sandbox-settings
      */
     sandbox?: SandboxSettings;
     /**
@@ -2014,6 +2014,15 @@ export declare type Options = {
     spawnClaudeCodeProcess?: (options: SpawnOptions) => SpawnedProcess;
 };
 
+/**
+ * Emitted on the same severity:'error' path as the usage-limit bucket, but
+ * the condition is an org policy, not an exhausted limit — consumers route
+ * these to org-disabled presentation, never the usage-limit card.
+ *
+ * @alpha
+ */
+export declare const ORG_POLICY_LIMIT_PREFIXES: readonly ["This service is disabled for your org"];
+
 export declare type OutputFormat = JsonSchemaOutputFormat;
 
 export declare type OutputFormatType = 'json_schema';
@@ -2799,6 +2808,10 @@ export declare type SDKAssistantMessage = {
     uuid: UUID;
     session_id: string;
     request_id?: string;
+    /**
+     * This turn continued the preceding truncated assistant turn inside its trailing signed thinking block (max-output-tokens recovery). Its thinking signatures are cumulative over that preceding thinking-only turn, so a history replayed through the bridge must carry this flag back for the normalizer to keep the run's prefix on the wire. Wrapper-level sibling — never inside `message.content` — so it is not replayed to the model.
+     */
+    resumed_from_incomplete_thinking?: true;
     /**
      * Wire uuids of previously-delivered messages that this message replaces (refusal-fallback supersede). The list can include tombstoned tool_result frames from the refused leg, not only assistant frames. Evict the named messages on arrival and treat this frame as their canonical replacement. Idempotent with the end-of-turn model_refusal_fallback notice, whose retracted_message_uuids remains the complete audit record for the turn.
      */
@@ -2812,7 +2825,10 @@ export declare type SDKAssistantMessage = {
      */
     task_description?: string;
 
-
+    /**
+     * ISO timestamp of when this content block finished on the originating process. One API assistant turn may produce several assistant messages sharing a message.id, each with its own timestamp. Uses the originating host's clock, so it's for display only; do not order messages by this field. Older emitters omit it; consumers should fall back to receive time.
+     */
+    timestamp?: string;
 
 
 
@@ -3421,7 +3437,7 @@ declare type SDKControlListModelsRequest = {
 };
 
 /**
- * Invokes an MCP tool via the subprocess MCP client without a model turn. No permission check (control channel is trusted, same as other subtypes). SDK-type MCP servers (config.type === "sdk") are rejected — they are caller-provided, so the caller can invoke them directly without the subprocess round-trip. Result content passes through the same processing as model-turn MCP calls. Session expiry is not retried automatically; callers can mcp_reconnect and retry. UrlElicitationRequired (-32042) tries Elicitation hooks; if no hook resolves, the call errors with the URL in the message — open it out-of-band, then retry mcp_call. STAGED calls (input_files/output_files declared) additionally stage lane rows in/out around the call — see the input_files describe. Staged failures come back as a success-subtype response whose staging field carries a typed error_code; subtype:error is emitted only when the call could not be attempted at all (server not connected, kill switch, dispatch failure) and means nothing ran. Standard RPC semantics: a redelivered request_id supersedes the in-flight run (it is aborted and its response suppressed — exactly one response per request_id); conversion is idempotent, so re-running is safe. Cancellable via control_cancel_request.
+ * Invokes an MCP tool via the subprocess MCP client without a model turn. No permission check (control channel is trusted, same as other subtypes). SDK-type MCP servers (config.type === "sdk") are rejected — they are caller-provided, so the caller can invoke them directly without the subprocess round-trip. Result content passes through the same processing as model-turn MCP calls. Session expiry is not retried automatically; callers can mcp_reconnect and retry. UrlElicitationRequired (-32042) tries Elicitation hooks; if no hook resolves, the call errors with the URL in the message — open it out-of-band, then retry mcp_call. STAGED calls (input_files/output_files declared) additionally stage lane rows in/out around the call — see the input_files describe. Staged failures come back as a success-subtype response whose staging field carries a typed error_code; subtype:error is emitted only when the call could not be attempted at all (server not connected, kill switch, dispatch failure) and means nothing ran. A target server that is not yet connected is brought up on demand: dispatch runs the deferred plugin/MCP startup resolution (the work a first model turn would have done) and waits up to 30s — shortened by expires_at when that is sooner — for the server to connect before answering "MCP server not connected", so a dispatch that races plugin startup (e.g. after an idle-wake reattach) succeeds instead of failing until a turn runs. Standard RPC semantics: a redelivered request_id supersedes the in-flight run (it is aborted and its response suppressed — exactly one response per request_id); conversion is idempotent, so re-running is safe. Cancellable via control_cancel_request.
  */
 declare type SDKControlMcpCallRequest = {
     subtype: 'mcp_call';
@@ -6126,6 +6142,10 @@ export declare interface Settings {
      * Probability (0–1) that the session quality survey appears when eligible. 0.05 is a reasonable starting point.
      */
     feedbackSurveyRate?: number;
+    /**
+     * Model-drafted feedback (the SendFeedback tool). "notify" (default) shows a one-line notice when a draft is queued; "quiet" shows only the footer counter; "off" disables the tool entirely so drafts are never queued.
+     */
+    feedbackDrafts?: 'notify' | 'quiet' | 'off';
     /**
      * Whether to show tips in the spinner
      */
@@ -6505,6 +6525,12 @@ export declare interface SpawnedProcess {
     readonly killed: boolean;
     /** Exit code if the process has exited, null otherwise */
     readonly exitCode: number | null;
+    /**
+     * Signal that terminated the process, if any. Optional: ChildProcess
+     * provides it; custom spawners may omit it (signal exits then read as
+     * still-running until their 'exit' event delivers the signal).
+     */
+    readonly signalCode?: NodeJS.Signals | null;
     /**
      * Kill the process with the given signal
      * @param signal - The signal to send (e.g., 'SIGTERM', 'SIGKILL')
@@ -6514,6 +6540,11 @@ export declare interface SpawnedProcess {
      * Register a callback for when the process exits
      * @param event - Must be 'exit'
      * @param listener - Callback receiving exit code and signal
+     *
+     * ProcessTransport's built-in local spawn delivers this only after the
+     * child's stderr has also closed (bounded by a short grace), so exit
+     * consumers see a complete stderr tail in exit errors. Custom
+     * `spawnClaudeCodeProcess` implementations emit plain process exit.
      */
     on(event: 'exit', listener: (code: number | null, signal: NodeJS.Signals | null) => void): void;
     /**
@@ -6838,6 +6869,33 @@ export declare interface Transport {
     [Symbol.dispose]?(): void;
 }
 
+/**
+ * Messages meaning "a usage limit was genuinely reached" — the error-path
+ * outputs of getLimitReachedText (rateLimitMessages.ts) and
+ * getFableCreditsRequiredContent (api/errors.ts).
+ *
+ * @alpha
+ */
+export declare const USAGE_LIMIT_ERROR_PREFIXES: readonly ["You've hit your", "You've reached your", "You're out of usage credits", "Your org is out of usage · add funds to continue", "Your org is out of usage · contact your admin", "Your seat type doesn't include usage credits", "Your seat type doesn't include usage", "Your usage allocation has been disabled by your admin", "Your group's usage limit is set to $0", "Fable 5 requires usage credits", "You're out of extra usage", "Your seat type doesn't include extra usage"];
+
+/**
+ * Overage-transition notifications ("now drawing from credits"). Toast only;
+ * these never arrive as API errors.
+ *
+ * @alpha
+ */
+export declare const USAGE_TRANSITION_PREFIXES: readonly ["You're now using usage credits", "You're now using your usage allocation", "Now using your usage allocation", "Now using usage credits", "You're now using extra usage", "Now using extra usage"];
+
+/**
+ * Approaching-limit warnings (severity:'warning'). Footer/toast only; these
+ * never arrive as API errors. ("Approaching …" early warnings are
+ * deliberately unregistered — they render in the footer without
+ * <RateLimitMessage> styling; see the generator-coverage tests.)
+ *
+ * @alpha
+ */
+export declare const USAGE_WARNING_PREFIXES: readonly ["You've used", "You're close to"];
+
 /**
  * A `request_user_dialog` control request from the CLI, asking the SDK
  * consumer to render a blocking dialog and return the user's choice.