@evolvingmachines/evolve 0.0.60 → 0.0.61
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/chunk-N2LMMVL4.js +427 -0
- package/dist/cli/index.cjs +38 -37
- package/dist/cli/index.d.cts +3 -1
- package/dist/cli/index.d.ts +3 -1
- package/dist/cli/index.js +29 -28
- package/dist/index.cjs +54 -54
- package/dist/index.d.cts +111 -7
- package/dist/index.d.ts +111 -7
- package/dist/index.js +38 -38
- package/dist/{types-DlpTxdR_.d.cts → types-CMEpx9QI.d.cts} +313 -5
- package/dist/{types-DlpTxdR_.d.ts → types-CMEpx9QI.d.ts} +313 -5
- package/hosted-error-codes.json +6 -0
- package/package.json +5 -5
- package/skills/evolve-evals/references/cli-reference/analysis.mdx +69 -0
- package/skills/evolve-evals/references/cli-reference/check.mdx +86 -0
- package/skills/evolve-evals/references/cli-reference/dataset.mdx +13 -0
- package/skills/evolve-evals/references/cli-reference/run.mdx +6 -0
- package/skills/evolve-evals/references/cli-reference/trial.mdx +71 -2
- package/skills/evolve-evals/references/core-concepts/trial-outputs.mdx +14 -0
- package/skills/evolve-evals/references/sdk/python.mdx +22 -0
- package/skills/evolve-evals/references/sdk/typescript.mdx +22 -0
- package/skills/evolve-evals/references/sdk-reference/analyses.mdx +70 -0
- package/skills/evolve-evals/references/sdk-reference/checks.mdx +105 -4
- package/skills/evolve-evals/references/sdk-reference/datasets.mdx +40 -0
- package/skills/evolve-evals/references/sdk-reference/errors.mdx +11 -0
- package/skills/evolve-evals/references/sdk-reference/jobs.mdx +1 -1
- package/skills/evolve-evals/references/sdk-reference/trials.mdx +70 -0
- package/spec/openapi.yaml +3021 -360
- package/dist/chunk-JS2UTK2I.js +0 -427
|
@@ -636,6 +636,8 @@ interface JobCreate {
|
|
|
636
636
|
* means no embedded analysis — `jobs().analyze()` remains the manual door.
|
|
637
637
|
*/
|
|
638
638
|
analyze?: AnalyzeConfigInput;
|
|
639
|
+
/** Record the box's own system log stream (read back with `logs({ stream: "system" })`); off by default. */
|
|
640
|
+
system_log?: boolean;
|
|
639
641
|
/**
|
|
640
642
|
* Multiplier for task timeouts — Harbor's `--timeout-multiplier`, all five
|
|
641
643
|
* fields flat on this body exactly as Harbor's JobConfig carries them. The
|
|
@@ -1220,6 +1222,8 @@ interface Job {
|
|
|
1220
1222
|
* on a job this platform ran.
|
|
1221
1223
|
*/
|
|
1222
1224
|
sandbox_provider: EvalSandboxProvider | null;
|
|
1225
|
+
/** The create's `system_log`; derived jobs inherit it, a regrade and every pre-switch job answer false. */
|
|
1226
|
+
system_log: boolean;
|
|
1223
1227
|
/** Entity cardinality only — things with no status of their own. */
|
|
1224
1228
|
counts: {
|
|
1225
1229
|
agents: number;
|
|
@@ -3051,6 +3055,8 @@ interface ListChecksOptions extends PageOptions {
|
|
|
3051
3055
|
scope?: JobListScope;
|
|
3052
3056
|
/** Only checks in these statuses (the check's own ladder, CHECK_STATUSES). */
|
|
3053
3057
|
status?: CheckStatus[];
|
|
3058
|
+
/** Only checks on this dataset: "name" (every version) or "name@version". */
|
|
3059
|
+
dataset?: string;
|
|
3054
3060
|
}
|
|
3055
3061
|
/** Options for jobs().tasks() (default page 50, max 200) */
|
|
3056
3062
|
interface ListJobTasksOptions extends PageOptions {
|
|
@@ -3442,6 +3448,8 @@ interface DatasetsClient {
|
|
|
3442
3448
|
* — answers 404 `task_not_found`.
|
|
3443
3449
|
*/
|
|
3444
3450
|
getTaskBuild(ref: string, taskName: string): Promise<TaskBuild>;
|
|
3451
|
+
/** A task's files from the retained package (the bytes every trial ran against); `ref` must pin the version. */
|
|
3452
|
+
taskFiles(ref: string, taskName: string): TaskPackageFiles;
|
|
3445
3453
|
/**
|
|
3446
3454
|
* Pre-flight a local corpus BEFORE publishing (dry run): collect only the
|
|
3447
3455
|
* metadata files (each task's task.toml + the optional dataset.toml —
|
|
@@ -3850,7 +3858,7 @@ interface JobsClient {
|
|
|
3850
3858
|
* can hold it to the spec's enum, and the CLI can build its `--stream`
|
|
3851
3859
|
* validation from the same list instead of a second copy.
|
|
3852
3860
|
*/
|
|
3853
|
-
declare const TRIAL_ARTIFACT_STREAMS: readonly ["trace-parsed", "verifier", "trace-stdout", "trace-stderr", "trace-atif", "trajectory", "agent-home"];
|
|
3861
|
+
declare const TRIAL_ARTIFACT_STREAMS: readonly ["trace-parsed", "verifier", "trace-stdout", "trace-stderr", "trace-atif", "trajectory", "agent-home", "filesystem"];
|
|
3854
3862
|
/** One `?stream=` selector on the trace route. */
|
|
3855
3863
|
type TrialArtifactStream = (typeof TRIAL_ARTIFACT_STREAMS)[number];
|
|
3856
3864
|
/** Client for globally addressable trials — no job id in any signature */
|
|
@@ -3887,7 +3895,7 @@ interface TrialsClient {
|
|
|
3887
3895
|
* (a normal answer, not an error). "trace-parsed" is not an
|
|
3888
3896
|
* artifact — the parsed event trace rides trace()/traceEvents().
|
|
3889
3897
|
*/
|
|
3890
|
-
artifact(trialId: string, stream: Exclude<TrialArtifactStream, "trace-parsed" | "agent-home">): Promise<string | null>;
|
|
3898
|
+
artifact(trialId: string, stream: Exclude<TrialArtifactStream, "trace-parsed" | "agent-home" | "filesystem">): Promise<string | null>;
|
|
3891
3899
|
artifact(trialId: string, stream: "agent-home"): Promise<Record<string, string> | null>;
|
|
3892
3900
|
/**
|
|
3893
3901
|
* Regrade one settled trial: re-run its verifier against its recorded
|
|
@@ -3933,6 +3941,281 @@ interface TrialsClient {
|
|
|
3933
3941
|
* the API's typed 404.
|
|
3934
3942
|
*/
|
|
3935
3943
|
file(trialId: string, path: string, range?: TrialFileRange): Promise<Buffer>;
|
|
3944
|
+
/** The trial's file system, sandbox logs and process list: the running box while it lives, the kept tree after. */
|
|
3945
|
+
filesystem(trialId: string): RunFilesystem;
|
|
3946
|
+
}
|
|
3947
|
+
/** `live` = the box runs; `captured` = the tree was kept; `capturing` = still being written; `none` = nothing yet. */
|
|
3948
|
+
type FilesystemState = "live" | "captured" | "none" | "capturing";
|
|
3949
|
+
/** The running box a live file system reads. */
|
|
3950
|
+
interface FilesystemBox {
|
|
3951
|
+
provider: EvalSandboxProvider;
|
|
3952
|
+
id: string;
|
|
3953
|
+
role: "agent" | "verifier" | "analyzer" | "checker";
|
|
3954
|
+
/** ISO instant the box started. */
|
|
3955
|
+
since: string;
|
|
3956
|
+
}
|
|
3957
|
+
/** The kept tree's record. */
|
|
3958
|
+
interface FilesystemCapture {
|
|
3959
|
+
id: string;
|
|
3960
|
+
/** When the capture settled; null on one abandoned mid-way. */
|
|
3961
|
+
at: string | null;
|
|
3962
|
+
/** After a shared verifier ran in the box, after the seal (a separate verifier runs elsewhere), or after a run that failed. */
|
|
3963
|
+
phase: "after_verifier" | "after_seal" | "after_run";
|
|
3964
|
+
entries: number;
|
|
3965
|
+
changed_files: number;
|
|
3966
|
+
changed_bytes: number;
|
|
3967
|
+
/** `incomplete` = `left_out` names every path the run touched whose bytes are not stored. */
|
|
3968
|
+
status: "ready" | "incomplete" | "failed";
|
|
3969
|
+
left_out: string[];
|
|
3970
|
+
/** Why a `failed` capture stored nothing; null otherwise. */
|
|
3971
|
+
failure_reason: string | null;
|
|
3972
|
+
}
|
|
3973
|
+
/** `GET {owner}/filesystem` — what the other file system reads will answer from. */
|
|
3974
|
+
interface FilesystemStatus {
|
|
3975
|
+
state: FilesystemState;
|
|
3976
|
+
box: FilesystemBox | null;
|
|
3977
|
+
/** How live change events arrive: from the provider, or from polling the folders you declared open. */
|
|
3978
|
+
watcher: "native" | "poll" | null;
|
|
3979
|
+
root: string;
|
|
3980
|
+
/** Where the run works (`/app`). */
|
|
3981
|
+
work_dir: string;
|
|
3982
|
+
capture: FilesystemCapture | null;
|
|
3983
|
+
}
|
|
3984
|
+
/** One entry of a folder listing. */
|
|
3985
|
+
interface FilesystemEntry {
|
|
3986
|
+
name: string;
|
|
3987
|
+
type: "dir" | "file" | "symlink" | "other";
|
|
3988
|
+
size: number;
|
|
3989
|
+
mtime: string;
|
|
3990
|
+
/** Octal, e.g. `0644`. */
|
|
3991
|
+
mode: string;
|
|
3992
|
+
owner: string;
|
|
3993
|
+
/** What the run did to it since the box started; null when not known for the source. */
|
|
3994
|
+
changed: "created" | "modified" | null;
|
|
3995
|
+
/** Which part of the run changed it — the platform's own setup, the agent, or the verifier. */
|
|
3996
|
+
phase: "setup" | "agent" | "verifier" | null;
|
|
3997
|
+
/** Captured source: false = only the image holds it, it lists but does not open. Live: null. Package: true. */
|
|
3998
|
+
captured: boolean | null;
|
|
3999
|
+
/** Captured source, a file the run touched whose bytes are not stored — why, in the capture's own words. */
|
|
4000
|
+
left_out?: string;
|
|
4001
|
+
/** A symlink's target. */
|
|
4002
|
+
target?: string;
|
|
4003
|
+
}
|
|
4004
|
+
/** One page of a folder (`next_cursor` null = no next page). */
|
|
4005
|
+
interface FilesystemListing {
|
|
4006
|
+
path: string;
|
|
4007
|
+
source: "live" | "capture" | "package";
|
|
4008
|
+
entries: FilesystemEntry[];
|
|
4009
|
+
next_cursor: string | null;
|
|
4010
|
+
/** Server time spent, milliseconds. */
|
|
4011
|
+
ms: number;
|
|
4012
|
+
}
|
|
4013
|
+
interface FilesystemSearchHit {
|
|
4014
|
+
path: string;
|
|
4015
|
+
line: number;
|
|
4016
|
+
snippet: string;
|
|
4017
|
+
}
|
|
4018
|
+
interface FilesystemSearchResult {
|
|
4019
|
+
hits: FilesystemSearchHit[];
|
|
4020
|
+
/** More hits exist, or the search budget ran out — narrow the path or the text. */
|
|
4021
|
+
truncated: boolean;
|
|
4022
|
+
/** `box` = the whole box was searched (seconds); `path` = one folder. */
|
|
4023
|
+
scope: "path" | "box";
|
|
4024
|
+
source: "live" | "capture";
|
|
4025
|
+
/** Captured source only: files the run never touched were not searchable. */
|
|
4026
|
+
image_files_excluded?: boolean;
|
|
4027
|
+
ms: number;
|
|
4028
|
+
}
|
|
4029
|
+
/** One row of the changed-files list. */
|
|
4030
|
+
interface FilesystemChange {
|
|
4031
|
+
path: string;
|
|
4032
|
+
type: "dir" | "file" | "symlink" | "other";
|
|
4033
|
+
changed: "created" | "modified" | "removed";
|
|
4034
|
+
phase: "setup" | "agent" | "verifier";
|
|
4035
|
+
size: number;
|
|
4036
|
+
mtime: string;
|
|
4037
|
+
/** Captured source, a file whose bytes are not stored — why, in the capture's own words. */
|
|
4038
|
+
left_out?: string;
|
|
4039
|
+
}
|
|
4040
|
+
interface FilesystemChanges {
|
|
4041
|
+
source: "live" | "capture";
|
|
4042
|
+
/** The whole list's count and bytes; `items` is one page of it. */
|
|
4043
|
+
total: number;
|
|
4044
|
+
changed_bytes: number;
|
|
4045
|
+
items: FilesystemChange[];
|
|
4046
|
+
next_cursor: string | null;
|
|
4047
|
+
}
|
|
4048
|
+
interface FilesystemWatchResult {
|
|
4049
|
+
watcher: "native" | "poll";
|
|
4050
|
+
paths: string[];
|
|
4051
|
+
}
|
|
4052
|
+
/** The source a read may force; omitted = whichever the run has. */
|
|
4053
|
+
type FilesystemSource = "live" | "capture";
|
|
4054
|
+
interface FilesystemListOptions {
|
|
4055
|
+
/** Absolute box path (default `/`). */
|
|
4056
|
+
path?: string;
|
|
4057
|
+
source?: FilesystemSource;
|
|
4058
|
+
/** The previous page's `next_cursor`. */
|
|
4059
|
+
cursor?: string;
|
|
4060
|
+
/** Default 500, max 1000. */
|
|
4061
|
+
limit?: number;
|
|
4062
|
+
}
|
|
4063
|
+
interface FilesystemReadOptions {
|
|
4064
|
+
source?: FilesystemSource;
|
|
4065
|
+
/** A byte range (the wire's single `Range` grammar): `{ start, end }`, `{ start }`, or `{ suffix }`. */
|
|
4066
|
+
range?: TrialFileRange;
|
|
4067
|
+
}
|
|
4068
|
+
interface FilesystemSearchOptions {
|
|
4069
|
+
/** The text to find; a regular expression with `regex: true`. */
|
|
4070
|
+
q: string;
|
|
4071
|
+
/** The folder to search under (default `/`, the whole box — seconds). */
|
|
4072
|
+
path?: string;
|
|
4073
|
+
regex?: boolean;
|
|
4074
|
+
/** Hit cap (default 200, max 1000). */
|
|
4075
|
+
limit?: number;
|
|
4076
|
+
source?: FilesystemSource;
|
|
4077
|
+
}
|
|
4078
|
+
interface FilesystemChangesOptions {
|
|
4079
|
+
source?: FilesystemSource;
|
|
4080
|
+
/** Only one phase's changes (default all). */
|
|
4081
|
+
phase?: "setup" | "agent" | "verifier" | "all";
|
|
4082
|
+
cursor?: string;
|
|
4083
|
+
/** Default 500, max 1000. */
|
|
4084
|
+
limit?: number;
|
|
4085
|
+
}
|
|
4086
|
+
interface FilesystemArchiveOptions {
|
|
4087
|
+
/** The subtree to archive (default `/`, the whole tree). */
|
|
4088
|
+
path?: string;
|
|
4089
|
+
source?: FilesystemSource;
|
|
4090
|
+
}
|
|
4091
|
+
/** Resume + cancellation for the two live streams. */
|
|
4092
|
+
interface FilesystemStreamOptions {
|
|
4093
|
+
/** Resume after this event id (`fs` events: the seq; log lines: `<stream>:<seq>`). */
|
|
4094
|
+
lastEventId?: string;
|
|
4095
|
+
signal?: AbortSignal;
|
|
4096
|
+
}
|
|
4097
|
+
/** One frame of `GET {owner}/filesystem/events`. */
|
|
4098
|
+
type FilesystemStreamEvent = {
|
|
4099
|
+
event: "state";
|
|
4100
|
+
id?: string;
|
|
4101
|
+
data: {
|
|
4102
|
+
state: FilesystemState;
|
|
4103
|
+
box: FilesystemBox | null;
|
|
4104
|
+
};
|
|
4105
|
+
} | {
|
|
4106
|
+
event: "fs";
|
|
4107
|
+
id: string;
|
|
4108
|
+
data: {
|
|
4109
|
+
seq: number;
|
|
4110
|
+
t: string;
|
|
4111
|
+
path: string;
|
|
4112
|
+
type: "create" | "write" | "remove" | "rename";
|
|
4113
|
+
source: "watch" | "poll";
|
|
4114
|
+
};
|
|
4115
|
+
} | {
|
|
4116
|
+
event: "ping";
|
|
4117
|
+
id?: string;
|
|
4118
|
+
data: Record<string, never>;
|
|
4119
|
+
};
|
|
4120
|
+
/** The named sandbox streams a run records. Runtime list so the CLI validates `--stream` against it. */
|
|
4121
|
+
declare const SANDBOX_LOG_STREAMS: readonly ["agent", "verifier", "setup", "system", "metrics"];
|
|
4122
|
+
type SandboxLogStream = (typeof SANDBOX_LOG_STREAMS)[number];
|
|
4123
|
+
interface SandboxLogLine {
|
|
4124
|
+
seq: number;
|
|
4125
|
+
/** When the line was recorded; null when the record holds no time for it. */
|
|
4126
|
+
t: string | null;
|
|
4127
|
+
fd: "out" | "err";
|
|
4128
|
+
line: string;
|
|
4129
|
+
}
|
|
4130
|
+
/** One page of a stream. */
|
|
4131
|
+
interface SandboxLogLines {
|
|
4132
|
+
stream: SandboxLogStream;
|
|
4133
|
+
lines: SandboxLogLine[];
|
|
4134
|
+
next_cursor: string | null;
|
|
4135
|
+
/** Why the page is empty when the platform holds nothing for this stream (never an error). */
|
|
4136
|
+
reason?: string;
|
|
4137
|
+
}
|
|
4138
|
+
interface SandboxLogOptions {
|
|
4139
|
+
stream: SandboxLogStream;
|
|
4140
|
+
/** The last `seq` you hold. */
|
|
4141
|
+
cursor?: string;
|
|
4142
|
+
/** Default 1000, max 1000. */
|
|
4143
|
+
limit?: number;
|
|
4144
|
+
}
|
|
4145
|
+
/** One frame of `GET {owner}/logs/events`. */
|
|
4146
|
+
type SandboxLogEvent = {
|
|
4147
|
+
event: "line";
|
|
4148
|
+
id: string;
|
|
4149
|
+
data: SandboxLogLine & {
|
|
4150
|
+
stream: SandboxLogStream;
|
|
4151
|
+
};
|
|
4152
|
+
} | {
|
|
4153
|
+
event: "state";
|
|
4154
|
+
id?: string;
|
|
4155
|
+
data: {
|
|
4156
|
+
state: FilesystemState;
|
|
4157
|
+
box: FilesystemBox | null;
|
|
4158
|
+
};
|
|
4159
|
+
} | {
|
|
4160
|
+
event: "ping";
|
|
4161
|
+
id?: string;
|
|
4162
|
+
data: Record<string, never>;
|
|
4163
|
+
};
|
|
4164
|
+
interface SandboxProcs {
|
|
4165
|
+
/** The process listing as text. */
|
|
4166
|
+
text: string;
|
|
4167
|
+
ms: number;
|
|
4168
|
+
}
|
|
4169
|
+
/** Reads answer from the running box while it lives and the kept tree after; `source` forces one (else `filesystem_state`). */
|
|
4170
|
+
interface RunFilesystem {
|
|
4171
|
+
/** Which source the run has, its box, and its capture record. */
|
|
4172
|
+
status(): Promise<FilesystemStatus>;
|
|
4173
|
+
/** One page of a folder, sorted by name. A folder that is not there is 404 `not_found`. */
|
|
4174
|
+
list(options?: FilesystemListOptions): Promise<FilesystemListing>;
|
|
4175
|
+
/** Raw bytes, byte-exact; an unranged read over the server's ceiling is 413 — read it in slices. */
|
|
4176
|
+
read(path: string, options?: FilesystemReadOptions): Promise<Buffer>;
|
|
4177
|
+
/** Content search under a folder (fast) or over the whole box (seconds); `truncated` says when there was more. */
|
|
4178
|
+
search(options: FilesystemSearchOptions): Promise<FilesystemSearchResult>;
|
|
4179
|
+
/** The files the run created, modified or removed, with the phase; served from the kept tree. */
|
|
4180
|
+
changes(options?: FilesystemChangesOptions): Promise<FilesystemChanges>;
|
|
4181
|
+
/**
|
|
4182
|
+
* A `.tar.gz` of one subtree — in memory, saved under `to`, or as a stream (jobs().download()'s three shapes).
|
|
4183
|
+
* While the box runs, a subtree too large to read out in time is refused (`feature_unsupported`).
|
|
4184
|
+
*/
|
|
4185
|
+
archive(options?: FilesystemArchiveOptions): Promise<Buffer>;
|
|
4186
|
+
archive(options: FilesystemArchiveOptions & {
|
|
4187
|
+
to: string;
|
|
4188
|
+
}): Promise<string>;
|
|
4189
|
+
archive(options: FilesystemArchiveOptions & {
|
|
4190
|
+
stream: true;
|
|
4191
|
+
}): Promise<ReadableStream<Uint8Array>>;
|
|
4192
|
+
/** The folders you have open, so change events cover them where the watcher is `poll`; replaces the set, at most 8. */
|
|
4193
|
+
watch(paths: string[]): Promise<FilesystemWatchResult>;
|
|
4194
|
+
/** `state`, then one `fs` frame per change, `ping` every 15 s; a `state` reply to `lastEventId` means relist. */
|
|
4195
|
+
events(options?: FilesystemStreamOptions): AsyncIterableIterator<FilesystemStreamEvent>;
|
|
4196
|
+
/** One page of a named sandbox stream; a stream the platform holds nothing for is an empty page with its `reason`. */
|
|
4197
|
+
logs(options: SandboxLogOptions): Promise<SandboxLogLines>;
|
|
4198
|
+
/** Every stream at once; ends once the box is gone and every recorded line was sent. `lastEventId` = `<stream>:<seq>`. */
|
|
4199
|
+
logEvents(options?: FilesystemStreamOptions): AsyncIterableIterator<SandboxLogEvent>;
|
|
4200
|
+
/** The box's process list, live only (409 `filesystem_state` once it is gone). */
|
|
4201
|
+
procs(): Promise<SandboxProcs>;
|
|
4202
|
+
}
|
|
4203
|
+
/** `GET …/tasks/{task}/filesystem` — the task package owner's status: `state` always `none`. */
|
|
4204
|
+
interface TaskPackageFilesystemStatus extends FilesystemStatus {
|
|
4205
|
+
state: "none";
|
|
4206
|
+
source: "package";
|
|
4207
|
+
/** False when the version keeps no package — its files cannot be served (409 `task_package_not_retained`). */
|
|
4208
|
+
package_retained: boolean;
|
|
4209
|
+
}
|
|
4210
|
+
/** A task's files from the retained package — a read-only owner: no live source, no events, no logs. */
|
|
4211
|
+
interface TaskPackageFiles {
|
|
4212
|
+
status(): Promise<TaskPackageFilesystemStatus>;
|
|
4213
|
+
/** One page of a folder of the task directory (`/` holds instruction.md, task.toml, environment/, tests/). */
|
|
4214
|
+
list(options?: Omit<FilesystemListOptions, "source">): Promise<FilesystemListing>;
|
|
4215
|
+
/** RAW BYTES of one file of the task directory; `range` reads a slice. */
|
|
4216
|
+
read(path: string, options?: {
|
|
4217
|
+
range?: TrialFileRange;
|
|
4218
|
+
}): Promise<Buffer>;
|
|
3936
4219
|
}
|
|
3937
4220
|
/**
|
|
3938
4221
|
* The stored artifacts an analysis run OWNS under its own id — the analyzer's
|
|
@@ -4091,6 +4374,8 @@ interface AnalysesClient {
|
|
|
4091
4374
|
download(analysisId: string, options: {
|
|
4092
4375
|
stream: true;
|
|
4093
4376
|
}): Promise<ReadableStream<Uint8Array>>;
|
|
4377
|
+
/** The analysis run's FILE SYSTEM, sandbox logs and process list (RunFilesystem): the analyzer's box while it runs, the kept tree after. */
|
|
4378
|
+
filesystem(analysisId: string): RunFilesystem;
|
|
4094
4379
|
download(analysisId: string, options?: DownloadJobOptions): Promise<Buffer | string | ReadableStream<Uint8Array>>;
|
|
4095
4380
|
}
|
|
4096
4381
|
/**
|
|
@@ -4120,6 +4405,8 @@ interface AnalysesClient {
|
|
|
4120
4405
|
* each with its reason.
|
|
4121
4406
|
*/
|
|
4122
4407
|
interface CheckConfigInput {
|
|
4408
|
+
/** A name for the check (Harbor's `--job-name`); omitted, the accept timestamp `YYYY-MM-DD__HH-MM-SS`. 1-120 characters. */
|
|
4409
|
+
name?: string;
|
|
4123
4410
|
/** Model the checker agent runs (Harbor's `-m/--model`); must be on the claude roster (`GET /api/meta`). */
|
|
4124
4411
|
model_name?: string;
|
|
4125
4412
|
/** The rubric (Harbor's `-r/--rubric` file as its `{criteria}` object); default: the platform's check rubric (eleven criteria). */
|
|
@@ -4233,6 +4520,8 @@ interface TaskCheck {
|
|
|
4233
4520
|
*/
|
|
4234
4521
|
interface Check {
|
|
4235
4522
|
id: string;
|
|
4523
|
+
/** The caller's name, or the accept timestamp in Harbor's `YYYY-MM-DD__HH-MM-SS` shape. */
|
|
4524
|
+
name: string;
|
|
4236
4525
|
status: CheckStatus;
|
|
4237
4526
|
source: CheckSource;
|
|
4238
4527
|
model_name: string;
|
|
@@ -4257,6 +4546,21 @@ interface Check {
|
|
|
4257
4546
|
/** When the last task settled; null until every task has. */
|
|
4258
4547
|
finished_at: string | null;
|
|
4259
4548
|
}
|
|
4549
|
+
/**
|
|
4550
|
+
* The policy an empty check config resolves to (GET /api/checks/defaults):
|
|
4551
|
+
* each key the value `Check` echoes for a check created with no config,
|
|
4552
|
+
* except `prompt`, which `Check` serves as null and this serves as the
|
|
4553
|
+
* template text.
|
|
4554
|
+
*/
|
|
4555
|
+
interface CheckDefaults {
|
|
4556
|
+
model_name: string;
|
|
4557
|
+
rubric: Rubric;
|
|
4558
|
+
/** The built-in check prompt template, unrendered — pass it as `prompt` to run the default body explicitly, or edit it from here. */
|
|
4559
|
+
prompt: string;
|
|
4560
|
+
/** The effort the default model runs at when the config names none. */
|
|
4561
|
+
reasoning_effort: string;
|
|
4562
|
+
sandbox_provider: EvalSandboxProvider;
|
|
4563
|
+
}
|
|
4260
4564
|
/** Options for checks().watch() */
|
|
4261
4565
|
interface WatchCheckOptions {
|
|
4262
4566
|
/** Called on every observed change of the check's per-task statuses, with the check body the observation came from. */
|
|
@@ -4323,8 +4627,10 @@ interface ChecksClient {
|
|
|
4323
4627
|
create(input: CreateCheckInput): Promise<Check>;
|
|
4324
4628
|
/** The check with its per-task results — for every status. 404 `check_not_found` for an id you cannot read. */
|
|
4325
4629
|
get(checkId: string): Promise<Check>;
|
|
4326
|
-
/** Every check you may read, newest first (cursor-paged); `{ scope, status }` narrow it. */
|
|
4630
|
+
/** Every check you may read, newest first (cursor-paged); `{ scope, status, dataset }` narrow it. */
|
|
4327
4631
|
list(options?: ListChecksOptions): CheckList;
|
|
4632
|
+
/** The defaults a check runs under when its config names nothing (GET /api/checks/defaults): model, effort, provider, rubric and the unrendered prompt template. */
|
|
4633
|
+
defaults(): Promise<CheckDefaults>;
|
|
4328
4634
|
/** Poll a check until every task settled; resolves with the final Check. */
|
|
4329
4635
|
watch(checkId: string, options?: WatchCheckOptions): Promise<Check>;
|
|
4330
4636
|
/**
|
|
@@ -4376,6 +4682,8 @@ interface ChecksClient {
|
|
|
4376
4682
|
download(id: string, options: {
|
|
4377
4683
|
stream: true;
|
|
4378
4684
|
}): Promise<ReadableStream<Uint8Array>>;
|
|
4685
|
+
/** The task check's FILE SYSTEM, sandbox logs and process list (RunFilesystem), under the check that owns it. */
|
|
4686
|
+
taskFilesystem(checkId: string, taskCheckId: string): RunFilesystem;
|
|
4379
4687
|
download(id: string, options?: DownloadJobOptions): Promise<Buffer | string | ReadableStream<Uint8Array>>;
|
|
4380
4688
|
}
|
|
4381
4689
|
/** A key descriptor. The secret is never returned. */
|
|
@@ -4491,7 +4799,7 @@ interface OrgsClient {
|
|
|
4491
4799
|
* that file existed. Adding a code means editing the spec, that file, this
|
|
4492
4800
|
* list, and the Python pair.
|
|
4493
4801
|
*/
|
|
4494
|
-
declare const HOSTED_ERROR_CODES: readonly ["missing_authorization", "invalid_api_key", "read_only_key", "credential_service_unavailable", "rate_limited", "insufficient_credits", "quota_exceeded", "invalid_json", "invalid_input", "invalid_limit", "invalid_status", "invalid_visibility", "invalid_cursor", "invalid_after", "invalid_format", "invalid_ids", "invalid_multipart", "idempotency_key_reused", "dataset_not_found", "dataset_version_not_found", "dataset_name_taken", "dataset_in_use", "dataset_not_owned", "dataset_import_in_progress", "upstream_not_watchable", "no_active_version", "version_not_ready", "version_not_activatable", "unknown_task_names", "no_tasks", "task_not_found", "task_failed_to_build", "upload_session_not_found", "upload_offset_mismatch", "upload_chunk_digest_mismatch", "upload_incomplete", "upload_archive_digest_mismatch", "upload_session_failed", "too_many_concurrent_upload_chunks", "agent_not_found", "agent_name_taken", "agent_name_reserved", "agent_invalid_name", "agent_source_required", "agent_source_conflict", "agent_invalid_env", "agent_too_large", "agent_limit_reached", "skill_not_found", "skill_name_not_found", "skill_ref_invalid", "skill_unresolvable", "skill_invalid", "skill_in_use", "skill_too_large", "skill_limit_reached", "too_many_concurrent_skill_uploads", "secret_not_found", "secret_ambiguous", "secret_brokered_unsupported", "secret_exists", "secret_not_attached", "agent_version_not_found", "agent_version_unresolvable", "agent_kwarg_unsupported", "agent_config_unsupported", "agent_config_key_refused", "agent_preset_unsupported", "provider_unsupported", "job_not_found", "job_not_terminal", "no_failed_trials", "trial_not_found", "trial_not_settled", "concurrent_update", "regrade_source_ineligible", "no_regradable_trials", "invalid_rubric", "analysis_already_running", "analysis_not_found", "analysis_not_terminal", "check_not_found", "check_not_terminal", "no_checkable_tasks", "too_many_concurrent_check_uploads", "no_analyzable_trials", "not_a_job_dir", "invalid_trial", "trial_too_large", "upload_too_large", "job_uploaded", "job_already_uploaded", "too_many_concurrent_job_uploads", "job_import_not_found", "import_not_found", "import_too_large", "too_many_concurrent_imports", "invalid_archive", "unpinned_git_ref", "hub_package_not_found", "hub_unreachable", "package_not_retained", "package_corrupt", "package_missing", "too_many_concurrent_package_downloads", "org_not_found", "org_slug_taken", "org_forbidden", "org_personal_immutable", "org_last_owner", "org_in_use", "org_member_not_found", "invite_not_found", "invite_invalid", "internal_error"];
|
|
4802
|
+
declare const HOSTED_ERROR_CODES: readonly ["missing_authorization", "invalid_api_key", "read_only_key", "credential_service_unavailable", "rate_limited", "insufficient_credits", "quota_exceeded", "invalid_json", "invalid_input", "invalid_limit", "invalid_status", "invalid_visibility", "invalid_cursor", "invalid_after", "invalid_format", "invalid_ids", "invalid_multipart", "idempotency_key_reused", "dataset_not_found", "dataset_version_not_found", "dataset_name_taken", "dataset_in_use", "dataset_not_owned", "dataset_import_in_progress", "upstream_not_watchable", "no_active_version", "version_not_ready", "version_not_activatable", "unknown_task_names", "no_tasks", "task_not_found", "task_failed_to_build", "upload_session_not_found", "upload_offset_mismatch", "upload_chunk_digest_mismatch", "upload_incomplete", "upload_archive_digest_mismatch", "upload_session_failed", "too_many_concurrent_upload_chunks", "agent_not_found", "agent_name_taken", "agent_name_reserved", "agent_invalid_name", "agent_source_required", "agent_source_conflict", "agent_invalid_env", "agent_too_large", "agent_limit_reached", "skill_not_found", "skill_name_not_found", "skill_ref_invalid", "skill_unresolvable", "skill_invalid", "skill_in_use", "skill_too_large", "skill_limit_reached", "too_many_concurrent_skill_uploads", "secret_not_found", "secret_ambiguous", "secret_brokered_unsupported", "secret_exists", "secret_not_attached", "agent_version_not_found", "agent_version_unresolvable", "agent_kwarg_unsupported", "agent_config_unsupported", "agent_config_key_refused", "agent_preset_unsupported", "provider_unsupported", "job_not_found", "job_not_terminal", "no_failed_trials", "trial_not_found", "trial_not_settled", "concurrent_update", "regrade_source_ineligible", "no_regradable_trials", "invalid_rubric", "analysis_already_running", "analysis_not_found", "analysis_not_terminal", "check_not_found", "check_not_terminal", "no_checkable_tasks", "too_many_concurrent_check_uploads", "no_analyzable_trials", "not_a_job_dir", "invalid_trial", "trial_too_large", "upload_too_large", "job_uploaded", "job_already_uploaded", "too_many_concurrent_job_uploads", "job_import_not_found", "import_not_found", "import_too_large", "too_many_concurrent_imports", "invalid_archive", "unpinned_git_ref", "hub_package_not_found", "hub_unreachable", "package_not_retained", "package_corrupt", "package_missing", "too_many_concurrent_package_downloads", "org_not_found", "org_slug_taken", "org_forbidden", "org_personal_immutable", "org_last_owner", "org_in_use", "org_member_not_found", "invite_not_found", "invite_invalid", "not_found", "not_captured", "filesystem_state", "feature_unsupported", "provider_unreachable", "task_package_not_retained", "internal_error"];
|
|
4495
4803
|
/** One of the API's stable error codes. */
|
|
4496
4804
|
type HostedErrorCode = (typeof HOSTED_ERROR_CODES)[number];
|
|
4497
4805
|
/** True when `value` is a code this SDK version knows about (narrowing guard). */
|
|
@@ -4774,4 +5082,4 @@ interface CapabilityDocument {
|
|
|
4774
5082
|
error_codes: string[];
|
|
4775
5083
|
}
|
|
4776
5084
|
|
|
4777
|
-
export { type AnalysisList as $, type AgentInput as A, type AgentDatasetStats as B, type Check as C, type DatasetVersion as D, type AgentEffortSupport as E, type AgentInfo as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentList as I, type JobCreate as J, type AgentModelOption as K, type AgentPage as L, type AgentResult as M, type AgentSource as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentSourceInput as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentUpsertInput as V, type AnalysisArtifactStream as W, type AnalysisCheck as X, type AnalysisEvidence as Y, type AnalysisFailure as Z, type AnalysisLabel as _, type TrialAnalysis as a,
|
|
5085
|
+
export { type AnalysisList as $, type AgentInput as A, type AgentDatasetStats as B, type Check as C, type DatasetVersion as D, type AgentEffortSupport as E, type AgentInfo as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentList as I, type JobCreate as J, type AgentModelOption as K, type AgentPage as L, type AgentResult as M, type AgentSource as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentSourceInput as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentUpsertInput as V, type AnalysisArtifactStream as W, type AnalysisCheck as X, type AnalysisEvidence as Y, type AnalysisFailure as Z, type AnalysisLabel as _, type TrialAnalysis as a, GATEWAY_TRACE_SEQ_BASE as a$, type AnalysisPage as a0, type AnalysisStatus as a1, type AnalysisTranscript as a2, type AnalysisTranscriptOptions as a3, type AnalyzeConfig as a4, type AnalyzeConfigInput as a5, type ApiKey as a6, type AttemptPhase as a7, type AuthStatus as a8, type Awaitable as a9, type DownloadDatasetOptions as aA, type DownloadJobOptions as aB, EVAL_SANDBOX_PROVIDERS as aC, type Agent as aD, type ModelInfo as aE, type EvalSandboxProvider as aF, type StepResult as aG, type ExceptionInfo as aH, type FilesystemArchiveOptions as aI, type FilesystemBox as aJ, type FilesystemCapture as aK, type FilesystemChange as aL, type FilesystemChanges as aM, type FilesystemChangesOptions as aN, type FilesystemEntry as aO, type FilesystemListOptions as aP, type FilesystemListing as aQ, type FilesystemReadOptions as aR, type FilesystemSearchHit as aS, type FilesystemSearchOptions as aT, type FilesystemSearchResult as aU, type FilesystemSource as aV, type FilesystemState as aW, type FilesystemStatus as aX, type FilesystemStreamEvent as aY, type FilesystemStreamOptions as aZ, type FilesystemWatchResult as a_, CHECK_STATUSES as aa, type CheckDefaults as ab, type CheckLabel as ac, type CompareCell as ad, type CompareCoverage as ae, type CompareJobAggregate as af, type CompareResponse as ag, type CompareTaskRow as ah, type Dataset as ai, type DatasetImportFailure as aj, type DatasetImportList as ak, type DatasetImportPage as al, type DatasetImportStatus as am, type DatasetList as an, type DatasetPage as ao, type DatasetPatch as ap, type DatasetPreflight as aq, type DatasetRef as ar, type DatasetSelector as as, type DatasetSource as at, type DatasetVersionArchiveSource as au, type DatasetVersionArchiveUrlSource as av, type DatasetVersionGitSource as aw, type DatasetVersionHubSource as ax, type DatasetVersionSource as ay, type DatasetVersionState as az, type DatasetImport as b, type RubricCriterion as b$, type GatewayUsage as b0, type GetDatasetOptions as b1, type GrepJobOptions as b2, HOSTED_ERROR_CODES as b3, type ImportPhase as b4, type ImportPhaseProgress as b5, type ImportWarning as b6, type InfraFailureSignature as b7, JOB_LIST_SCOPES as b8, type JobAnalysisStats as b9, type ListDatasetsOptions as bA, type ListImportsOptions as bB, type ListJobImportsOptions as bC, type ListJobTasksOptions as bD, type ListJobsOptions as bE, type ListSkillsOptions as bF, type ListTrialFilesOptions as bG, type ListTrialsOptions as bH, type ManagedProviderCapability as bI, type OrgQuota as bJ, type OrgRole as bK, type OrgUsage as bL, type Organization as bM, type OrganizationDetail as bN, type Page as bO, type PageOptions as bP, type PassAtKGroup as bQ, type PassAtKPoint as bR, type PreflightDatasetInput as bS, type PreflightDeferredCheck as bT, type PreflightManifestVerdict as bU, type PreflightTaskVerdict as bV, type ProviderCapability as bW, type PublishDatasetOptions as bX, type RegradeRequest as bY, type ResumeRequest as bZ, type RetryRequest as b_, type JobDeleteResult as ba, type JobFailure as bb, type JobGrepGroup as bc, type JobGrepPage as bd, type JobImport as be, type JobImportFailure as bf, type JobImportList as bg, type JobImportPage as bh, type JobImportPhaseName as bi, type JobImportPhaseProgress as bj, type JobImportProgress as bk, type JobImportSkippedTrial as bl, type JobImportSource as bm, type JobList as bn, type JobListScope as bo, type JobPage as bp, type JobStats as bq, type JobStatus as br, type JobTaskLink as bs, type JobTaskRollup as bt, type JobTaskRollupList as bu, type JobTaskRollupPage as bv, type JobWatch as bw, type JudgeResult as bx, type ListAgentsOptions as by, type ListAnalysesOptions as bz, type JobEvent as c, type RunFilesystem as c0, SANDBOX_LOG_STREAMS as c1, type SandboxLogEvent as c2, type SandboxLogLine as c3, type SandboxLogLines as c4, type SandboxLogOptions as c5, type SandboxLogStream as c6, type SandboxProcs as c7, type SkillLock as c8, type SkillUpload as c9, type TrialFileRange as cA, type TrialGpuCost as cB, type TrialList as cC, type TrialPage as cD, type TrialStatus as cE, type TrialStatusTally as cF, type TrialTaskLink as cG, type TrialUploadProvenance as cH, type UploadJobOptions as cI, type UploadProvenance as cJ, type UpstreamStatus as cK, type VerifierEnvironmentMode as cL, type VerifierResult as cM, type WatchAnalysisOptions as cN, type WatchImportOptions as cO, type WatchJobImportOptions as cP, type WatchJobOptions as cQ, gatewayUsageOf as cR, isHostedErrorCode as cS, passAtK as cT, type SkillUploadList as ca, type SkillUploadPage as cb, type SourceJob as cc, type SpendSource as cd, type StartJobOptions as ce, type StatusVocabulary as cf, type StopResponse as cg, TASK_LINKED_BY as ch, TASK_LINK_REASONS as ci, TRIAL_ARTIFACT_STREAMS as cj, TRIAL_STATUSES as ck, type Task as cl, type TaskCheckTranscript as cm, type TaskLinkReason as cn, type TaskLinkedBy as co, type TaskNote as cp, type TaskPackageFiles as cq, type TaskPackageFilesystemStatus as cr, type TaskProviderVerdict as cs, type TimingInfo as ct, type TraceEventPage as cu, type TraceOptions as cv, type TrialArtifactStream as cw, type TrialCounts as cx, type TrialFile as cy, type TrialFilePage as cz, type DatasetFailedTask as d, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type Job as i, type TaskCheck as j, type DatasetsClient as k, type AgentsClient as l, type JobsClient as m, type TrialsClient as n, type AnalysesClient as o, type ChecksClient as p, type CapabilityDocument as q, type HostedClientConfig as r, type AuthClient as s, AGENT_EFFORT_SUPPORT_VALUES as t, ANALYSIS_ARTIFACT_STREAMS as u, ANALYSIS_STATUSES as v, type ActiveDataset as w, type AgentArm as x, type AgentArmInput as y, type AgentCapability as z };
|