@oneharness/sdk 0.11.3 → 0.12.0

package/dist/generated/history-stream-envelope.d.ts

@@ -14,7 +14,7 @@ export type HistoryStreamEnvelope = {
         [k: string]: unknown;
     } | ({
         error: string;
-        schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | undefined;
+        schema_version?: "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8" | undefined;
         [k: string]: unknown;
     } & ({
         status?: "nonzero" | "timeout" | "cancelled" | "spawn-error" | "skipped" | undefined;
@@ -26,27 +26,40 @@ export type HistoryStreamEnvelope = {
         work?: null | undefined;
         [k: string]: unknown;
     } | {
-        schema_version?: "1.7" | undefined;
+        schema_version?: "1.7" | "1.8" | undefined;
         status?: "nonzero" | "timeout" | "cancelled" | undefined;
         work: "done" | "none";
         [k: string]: unknown;
+    }) & ({
+        observed_model?: null | undefined;
+        [k: string]: unknown;
+    } | {
+        observed_model: string;
+        schema_version?: "1.8" | undefined;
+        [k: string]: unknown;
     }) & ({
         status?: "ok" | "nonzero" | "timeout" | "spawn-error" | "skipped" | "planned" | undefined;
         [k: string]: unknown;
     } | {
-        schema_version?: "1.4" | "1.5" | "1.6" | "1.7" | undefined;
+        schema_version?: "1.4" | "1.5" | "1.6" | "1.7" | "1.8" | undefined;
         [k: string]: unknown;
     }) & (({
-        failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "tool_deferred" | "untrusted_directory" | "input_too_large" | null;
+        failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "tool_deferred" | "untrusted_directory" | "input_too_large" | "model_mismatch" | null;
+        [k: string]: unknown;
+    } | {
+        schema_version?: "1.5" | "1.6" | "1.7" | "1.8" | undefined;
+        [k: string]: unknown;
+    }) & ({
+        failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | "model_mismatch" | null;
         [k: string]: unknown;
     } | {
-        schema_version?: "1.5" | "1.6" | "1.7" | undefined;
+        schema_version?: "1.6" | "1.7" | "1.8" | undefined;
         [k: string]: unknown;
     }) & ({
-        failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | null;
+        failure_kind?: "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | "untrusted_directory" | "input_too_large" | null;
         [k: string]: unknown;
     } | {
-        schema_version?: "1.6" | "1.7" | undefined;
+        schema_version?: "1.8" | undefined;
         [k: string]: unknown;
     })) & HistoryRecord;
     type: "record";
@@ -127,7 +140,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms: number;
@@ -136,6 +152,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -151,7 +183,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -269,7 +301,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels1 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms: number;
@@ -278,6 +313,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -389,7 +440,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels2 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -398,6 +452,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -413,7 +483,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -497,7 +567,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels3 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -506,6 +579,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms: number;
     /**
      * The normalized approval mode requested for the run.
@@ -521,7 +610,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -605,7 +694,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels4 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -614,6 +706,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -629,7 +737,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -713,7 +821,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels5 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -722,6 +833,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -737,7 +864,7 @@ export type HistoryRecord = {
      * run's single prompt).
      */
     prompt: string;
-    schema_version: "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     /**
      * The oneharness session id this run belongs to (the history file's stem).
      */
@@ -824,7 +951,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels6 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -833,6 +963,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -947,7 +1093,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels7 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: never | undefined;
@@ -956,6 +1105,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     observed_tool_ms?: never | undefined;
     /**
      * The normalized approval mode requested for the run.
@@ -1058,7 +1223,10 @@ export type HistoryRecord = {
     history_id: string;
     labels?: HistoryLabels8 | undefined;
     /**
-     * The effective top-level model for the run, if any.
+     * The model this run **requested** — the value oneharness put on the
+     * harness's model flag or protocol frame — if any. Copied from
+     * [`RunResult::model`], so it keeps meaning the requested model even where
+     * `observed_model` below is present.
      */
     model: string | null;
     model_ms?: number | null | undefined;
@@ -1067,6 +1235,22 @@ export type HistoryRecord = {
      * every record so a reader can resolve a session by name from any line.
      */
     name: string;
+    /**
+     * The model the harness **itself** reported the conversation would run
+     * under, read off its own protocol before any token was spent — for codex
+     * over `app-server`, `result.model` of the `thread/start` /
+     * `thread/resume` response. Copied from [`RunResult::observed_model`]:
+     * never inferred and never copied from `model`, and omitted on the wire on
+     * every path that reports none.
+     *
+     * On a completed app-server turn the two agree, because a turn whose
+     * reported model differed from the requested one was refused before it
+     * ran (`model_mismatch`) — so a completed record's `model` is what codex
+     * ran, and the record can no longer say one model while another was
+     * billed. They differ only on such a refused record, and where no model
+     * was requested at all. Gated to [`FIRST_MODEL_OBSERVATION_SCHEMA_VERSION`].
+     */
+    observed_model?: string | null | undefined;
     /**
      * Union of tool intervals observed at the stdout pipe. Unlike `tool_ms`,
      * this is not provider-measured and has no model-latency counterpart.
@@ -1153,7 +1337,7 @@ export type TimingSource = "provider_measured" | "stdout_observed";
  * the report, history) one definition to share instead of scattered string
  * literals.
  */
-export type FailureKind = "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | "untrusted_directory" | "input_too_large";
+export type FailureKind = "auth" | "rate_limit" | "model_not_found" | "quota" | "session_not_found" | "tool_deferred" | "untrusted_directory" | "input_too_large" | "model_mismatch";
 /**
  * The outcome of attempting to run one harness.
  */
@@ -1183,7 +1367,7 @@ export type HistoryEventLine = {
     harness: string;
     harness_id?: string | null | undefined;
     run_id: string;
-    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7";
+    schema_version: "1.2" | "1.3" | "1.4" | "1.5" | "1.6" | "1.7" | "1.8";
     variant?: string | null | undefined;
     [k: string]: unknown;
 } | {