@oneharness/sdk 0.10.2 → 0.11.0

package/dist/generated/contracts.d.ts

@@ -119,6 +119,26 @@ export type ExecutionTelemetry = {
     tool_ms: number;
     [k: string]: unknown;
 };
+/**
+ * Whether a candidate's normalized result carries **evidence it did the task's
+ * work** — the first thing [`startup_failure_reason`] consults.
+ *
+ * Two independent witnesses, either of which is decisive:
+ *
+ * - **Tool events.** A recorded tool call is the harness acting on the task.
+ * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
+ *   `signals::record_reports_work` classifies a raw harness record with, so the
+ *   two readings of "billed" are one contract with one implementation.
+ *
+ * It is also a **published reading**, not only an internal one: a run that
+ * failed with nothing to classify carries it as [`RunResult::work`] and in its
+ * history record, because there the question "did this candidate do anything?"
+ * is the only one left. One type for both, so the value a reader sees is the
+ * same value the fall-through verdict consulted.
+ *
+ * [billed]: crate::domain::signals::Usage::reports_billed_work
+ */
+export type RunWork = "done" | "none";
 /**
  * The top-level `run` report written to stdout.
  */
@@ -297,6 +317,24 @@ export interface FallbackReport {
      * when no candidate could run at all — every one was a startup failure.
      */
     ran: string | null;
+    /**
+     * Whether the candidate named by `ran` stopped the chain **having shown no
+     * evidence it did the task's work, and with nothing to say why** — its
+     * [`RunResult::work`] read [`RunWork::None`] on a failure no classifier
+     * recognized.
+     *
+     * The chain still stops there, deliberately: re-running a task that may
+     * genuinely have failed for free would burn the next identity's quota on
+     * the same failure, which is worse than stopping (see
+     * [`crate::domain::fallback::startup_failure_reason`]). What this flag adds
+     * is the *attribution* — without it a candidate that never got started reads
+     * in the report exactly like one that tried the task and failed it, and the
+     * remaining candidates look untried for a reason nobody can name.
+     *
+     * `false` whenever `ran` is `null` (no candidate ran at all — every one is
+     * in `fell_through`, each with its reason).
+     */
+    stopped_without_work: boolean;
     [k: string]: unknown;
 }
 /**
@@ -481,6 +519,22 @@ export interface RunResult {
      * Named preset, when this result came from a composed harness id.
      */
     variant: string | null;
+    /**
+     * What this run has to show for itself when `failure_kind` has nothing to
+     * say: [`RunWork::Done`] if the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] if nothing says it got that far. `null` on every
+     * other run — a success needs no such reading, and a *classified* failure
+     * already names its cause.
+     *
+     * This is the reading the fallback verdict itself consults
+     * ([`crate::domain::fallback::startup_failure_reason`]), published rather
+     * than left to be re-derived: an unclassified failure that did nothing is
+     * otherwise indistinguishable from one that ran the task and failed it, and
+     * the two mean opposite things to whoever reads the run. Publishing it
+     * changes no verdict — a candidate reading `none` still stops a chain, and
+     * one reading `done` still never falls through it.
+     */
+    work: RunWork | null;
     [k: string]: unknown;
 }
 /**

package/dist/generated/history-line.d.ts

@@ -6,7 +6,7 @@ export type HistoryLine = ({
     harness: string;
     harness_id?: string | null | undefined;
     run_id: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     type: "event";
     variant?: string | null | undefined;
     [k: string]: unknown;
@@ -27,7 +27,7 @@ export type HistoryLine = ({
     [k: string]: unknown;
 } | ({
     error: string;
-    schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | undefined;
+    schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 } & ({
     status?: "nonzero" | "timeout" | "cancelled" | "spawn-error" | "skipped" | undefined;
@@ -36,22 +36,30 @@ export type HistoryLine = ({
     failure_kind: "tool_deferred";
     [k: string]: unknown;
 }))) & ({
+    work?: null | undefined;
+    [k: string]: unknown;
+} | {
+    schema_version?: "1.7" | undefined;
+    status?: "nonzero" | "timeout" | "cancelled" | undefined;
+    work: "done" | "none";
+    [k: string]: unknown;
+}) & ({
     status?: "ok" | "nonzero" | "timeout" | "spawn-error" | "skipped" | "planned" | undefined;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.4" | "1.5" | "1.6" | undefined;
+    schema_version?: "1.4" | "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 }) & (({
     failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "tool_deferred" | "untrusted_directory" | "input_too_large" | null;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.5" | "1.6" | undefined;
+    schema_version?: "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 }) & ({
     failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | null;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.6" | undefined;
+    schema_version?: "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 })) & ({
     duration_ms: number;
@@ -74,7 +82,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at: string;
@@ -87,6 +95,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -109,7 +122,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at: string;
@@ -122,6 +135,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -144,7 +162,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at?: never | undefined;
@@ -157,6 +175,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -179,7 +202,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at?: never | undefined;
@@ -192,6 +215,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -214,7 +242,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at: string;
@@ -227,6 +255,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -249,7 +282,7 @@ export type HistoryLine = ({
     permission_mode: PermissionMode;
     project: string;
     prompt: string;
-    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.0" | "1.1" | "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     session: string;
     session_id: string | null;
     started_at?: never | undefined;
@@ -262,6 +295,11 @@ export type HistoryLine = ({
     type: "run";
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * Work evidence for a failure nothing classified (see
+     * [`HistoryRecord::work`]). Omitted on the wire when absent.
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 }));
 export type ToolCallStatus = "completed" | "failed" | "timeout" | "interrupted";
@@ -288,6 +326,26 @@ export type FailureKind = "auth" | "rate_limit" | "model_not_found" | "quota" |
  * downgrading it.
  */
 export type PermissionMode = "read-only" | "plan" | "default" | "edit" | "auto" | "bypass";
+/**
+ * Whether a candidate's normalized result carries **evidence it did the task's
+ * work** — the first thing [`startup_failure_reason`] consults.
+ *
+ * Two independent witnesses, either of which is decisive:
+ *
+ * - **Tool events.** A recorded tool call is the harness acting on the task.
+ * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
+ *   `signals::record_reports_work` classifies a raw harness record with, so the
+ *   two readings of "billed" are one contract with one implementation.
+ *
+ * It is also a **published reading**, not only an internal one: a run that
+ * failed with nothing to classify carries it as [`RunResult::work`] and in its
+ * history record, because there the question "did this candidate do anything?"
+ * is the only one left. One type for both, so the value a reader sees is the
+ * same value the fall-through verdict consulted.
+ *
+ * [billed]: crate::domain::signals::Usage::reports_billed_work
+ */
+export type RunWork = "done" | "none";
 /**
  * The outcome of attempting to run one harness.
  */

package/dist/generated/history-records.d.ts

@@ -93,7 +93,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -120,6 +120,22 @@ export type HistoryRecord = {
     tool_ms: number;
     usage: Usage;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -246,6 +262,22 @@ export type HistoryRecord = {
     tool_ms: number;
     usage: Usage1;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -323,7 +355,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -350,6 +382,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage2;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -415,7 +463,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -442,6 +490,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage3;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -507,7 +571,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -534,6 +598,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage4;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number;
@@ -599,7 +679,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.3" | "1.4" | "1.5" | "1.6";
+    schema_version: "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -626,6 +706,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage5;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -721,6 +817,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage6;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -828,6 +940,22 @@ export type HistoryRecord = {
     tool_ms?: never | undefined;
     usage: Usage7;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 } | {
     duration_ms: number | null;
@@ -933,6 +1061,22 @@ export type HistoryRecord = {
     tool_ms?: number | null | undefined;
     usage: Usage8;
     variant?: string | null | undefined;
+    /**
+     * What a run that failed with **nothing to classify** has to show for
+     * itself: [`RunWork::Done`] where the harness recorded a tool call or billed
+     * usage, [`RunWork::None`] where nothing says it got that far. Copied from
+     * [`RunResult::work`] — one value with one source, like every signal here —
+     * and omitted on the wire on every other run, since a success raises the
+     * question and a classified failure has already answered it.
+     *
+     * This is the field that tells apart, for a reader of the record, the two
+     * things an unexplained failure can be: a candidate that never got started,
+     * and one that ran the task and lost. Without it both are a `nonzero` line with a
+     * null `failure_kind` and empty accounting, and a chain that stopped at the
+     * first reads exactly like a chain that stopped at the second. Gated to
+     * [`FIRST_WORK_EVIDENCE_SCHEMA_VERSION`].
+     */
+    work?: RunWork | null | undefined;
     [k: string]: unknown;
 };
 export type ToolCallStatus = "completed" | "failed" | "timeout" | "interrupted";
@@ -956,12 +1100,32 @@ export type FailureKind = "auth" | "rate_limit" | "model_not_found" | "quota" |
  * The outcome of attempting to run one harness.
  */
 export type Status = "ok" | "nonzero" | "timeout" | "cancelled" | "spawn-error" | "skipped" | "planned";
+/**
+ * Whether a candidate's normalized result carries **evidence it did the task's
+ * work** — the first thing [`startup_failure_reason`] consults.
+ *
+ * Two independent witnesses, either of which is decisive:
+ *
+ * - **Tool events.** A recorded tool call is the harness acting on the task.
+ * - **Usage accounting.** [`Usage::reports_billed_work`][billed] — the same definition
+ *   `signals::record_reports_work` classifies a raw harness record with, so the
+ *   two readings of "billed" are one contract with one implementation.
+ *
+ * It is also a **published reading**, not only an internal one: a run that
+ * failed with nothing to classify carries it as [`RunResult::work`] and in its
+ * history record, because there the question "did this candidate do anything?"
+ * is the only one left. One type for both, so the value a reader sees is the
+ * same value the fall-through verdict consulted.
+ *
+ * [billed]: crate::domain::signals::Usage::reports_billed_work
+ */
+export type RunWork = "done" | "none";
 export type HistoryRecords = (({
     error?: null | undefined;
     [k: string]: unknown;
 } | ({
     error: string;
-    schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | undefined;
+    schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 } & ({
     status?: "nonzero" | "timeout" | "cancelled" | "spawn-error" | "skipped" | undefined;
@@ -970,22 +1134,30 @@ export type HistoryRecords = (({
     failure_kind: "tool_deferred";
     [k: string]: unknown;
 }))) & ({
+    work?: null | undefined;
+    [k: string]: unknown;
+} | {
+    schema_version?: "1.7" | undefined;
+    status?: "nonzero" | "timeout" | "cancelled" | undefined;
+    work: "done" | "none";
+    [k: string]: unknown;
+}) & ({
     status?: "ok" | "nonzero" | "timeout" | "spawn-error" | "skipped" | "planned" | undefined;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.4" | "1.5" | "1.6" | undefined;
+    schema_version?: "1.4" | "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 }) & (({
     failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "tool_deferred" | "untrusted_directory" | "input_too_large" | null;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.5" | "1.6" | undefined;
+    schema_version?: "1.5" | "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 }) & ({
     failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | null;
     [k: string]: unknown;
 } | {
-    schema_version?: "1.6" | undefined;
+    schema_version?: "1.6" | "1.7" | undefined;
     [k: string]: unknown;
 })) & HistoryRecord)[];
 /**