@evolvingmachines/evolve 0.0.55-project-sable.20260916.db0bc0bb

package/docs-evals/cli-reference/dataset.mdx

@@ -0,0 +1,133 @@
+---
+title: "evolve dataset"
+description: "Browse and publish the dataset catalog."
+---
+
+import GlobalOptions from '/snippets/global-options.mdx';
+import ListFlags from '/snippets/list-flags.mdx';
+
+```bash
+evolve dataset <command> [options]
+```
+
+## dataset list
+
+```bash
+evolve dataset list [options]
+```
+
+List the dataset catalog.
+
+<ListFlags />
+
+<ParamField path="--search" type="text">
+  Free-text filter over name and description.
+</ParamField>
+
+## dataset show
+
+```bash
+evolve dataset show <name[@version]> [options]
+```
+
+Show one dataset: its versions, its manifest identity, its tasks with a NOTES column, the providers each task can run on, and the failed tasks of the shown version with their reasons. One line reports a git dataset whose upstream ref has moved; it never appears in `--json`, where the same fact is the `upstream` field.
+
+<ParamField path="-l, --limit" type="n">
+  Task-list page size.
+</ParamField>
+
+<ParamField path="--cursor" type="c">
+  Resume the task list from a cursor.
+</ParamField>
+
+```bash
+evolve dataset show terminal-bench-4@4.0
+```
+
+## dataset check
+
+```bash
+evolve dataset check <dir>
+```
+
+Pre-flight a local corpus. A dry run: nothing is uploaded, nothing is written. Prints a verdict per task and exits 1 on any refusal.
+
+## dataset publish
+
+```bash
+evolve dataset publish [options]
+```
+
+Publish a dataset version from a git source, a local directory, or a fetchable source: a public tarball URL, or a public hub package. A name owned by someone else is refused before the bytes move, and an archive over the published ceiling is refused with `import_too_large`.
+
+A directory upload prints `upload <sent>/<total> (P%)` once per 10% of the archive; with `--json --watch` the same steps are `upload.progress` events (`sent_bytes`, `total_bytes`, `elapsed_sec`). A large upload prints `Registered import <id>` early and reads `QUEUED (receiving)` until the corpus has arrived. Once the build settles, `--watch` prints every task's outcome and ends with `built N of M tasks`.
+
+<ParamField path="--git" type="url">
+  Git repository URL. With `--ref`.
+</ParamField>
+
+<ParamField path="--ref" type="ref">
+  Pinned git ref: a full 40-hex commit sha, or a tag, resolved to its commit at publish and verified at import. Branch names are refused as `unpinned_git_ref`. With `--git`.
+</ParamField>
+
+<ParamField path="--path" type="subfolder">
+  Repository subfolder holding the corpus. With `--git`; only that folder is imported.
+</ParamField>
+
+<ParamField path="--dir" type="path">
+  Local corpus directory, tarred and uploaded. Pre-flighted first; see `--skip-preflight`.
+</ParamField>
+
+<ParamField path="--from" type="url | hub:org/name[@ref]">
+  Fetchable source the server pulls itself, with no local bytes: a public https tarball URL, or `hub:org/name[@ref]`, a public Harbor hub package. The ref is the latest tag by default, a revision number, or `sha256:<digest>`; it is resolved and digest-pinned when the publish is accepted.
+</ParamField>
+
+<ParamField path="--name" type="dataset">
+  Catalog dataset name to create or extend. Optional with `--dir` when the corpus carries a `dataset.toml` manifest, and with `--from hub:…`, which defaults to the package's short name. Required with `--git` and `--from <url>`.
+</ParamField>
+
+<ParamField path="--version" type="v">
+  Version label for the published version. Optional with `--dir` when `dataset.toml` declares one, and with `--from hub:…`, which defaults to the resolved revision. Required with `--git` and `--from <url>`.
+</ParamField>
+
+<ParamField path="--watch">
+  Poll until the publish settles: the version `READY` (built and active) or `FAILED`.
+</ParamField>
+
+<ParamField path="--skip-preflight">
+  Upload without the pre-flight check. A refused task then lands `FAILED` at import instead of being caught here.
+</ParamField>
+
+```bash
+evolve dataset publish --from hub:cookbook/hello-world --watch
+```
+
+## dataset watch
+
+```bash
+evolve dataset watch <name | import-id>
+```
+
+Re-attach to a publish and follow it to `READY` or `FAILED`, the same follow `dataset publish --watch` renders. Works after the CLI exited, or from another machine. A name with no live publish is refused, naming the newest settled import. With `--json` the same events stream, opened with `import.attached`; the transfer's own `upload.progress` lines belong to the publishing process and never appear here.
+
+## dataset download
+
+```bash
+evolve dataset download <name[@version]> [options]
+```
+
+Download the original corpus package. Owner only.
+
+<ParamField path="-o, --output-dir" type="dir">
+  Directory to save into. Defaults to the current directory.
+</ParamField>
+
+## dataset activate
+
+```bash
+evolve dataset activate <name> <version>
+```
+
+Make a `READY` version the dataset's active version.
+
+<GlobalOptions />

package/docs-evals/cli-reference/index.mdx

@@ -0,0 +1,108 @@
+---
+title: "CLI reference"
+description: "Every verb of the evolve command, from its own help output."
+---
+
+```bash
+evolve <command> [options]
+```
+
+The `evolve` command starts jobs and reads everything back. Every verb accepts `--json` for machine-readable output, and `-h` or `--help` prints its flags under short headings, with defaults, and one to three examples.
+
+## Verbs
+
+| Verb | What it does |
+| --- | --- |
+| [`run`](/cli-reference/run) | Start a job. Add `--watch` to follow it. The short form of `job start`. |
+| [`analyze`](/cli-reference/analyze) | Analyze a finished job's trial traces against a rubric. |
+| [`check`](/cli-reference/check) | Check task quality against a rubric. |
+| [`upload`](/cli-reference/upload) | Upload a job directory as a finished job. |
+
+## Command groups
+
+| Group | Commands |
+| --- | --- |
+| [`skills`](/cli-reference/skills) | `list`, `get`, `path`, `install` |
+| [`job`](/cli-reference/job) | `start`, `list`, `show`, `trials`, `tasks`, `compare`, `cancel`, `delete`, `stop`, `resume`, `retry`, `regrade`, `imports`, `import`, `download`, `grep` |
+| [`trial`](/cli-reference/trial) | `show`, `trace`, `download`, `retry`, `regrade`, `stop` |
+| [`analysis`](/cli-reference/analysis) | `list`, `show`, `trace`, `download` |
+| [`check`](/cli-reference/check) | `list`, `show`, `trace`, `download` |
+| [`dataset`](/cli-reference/dataset) | `list`, `show`, `check`, `publish`, `watch`, `download`, `activate` |
+| [`skill`](/cli-reference/skill) | `list`, `upload`, `show`, `delete` |
+| [`agent`](/cli-reference/agent) | `list`, `show`, `add`, `remove` |
+| [`auth`](/cli-reference/auth) | `status`, `org list`, `org show` |
+| [`secrets`](/cli-reference/secrets) | `set`, `list`, `delete` |
+
+The singular noun is canonical; `jobs`, `trials`, `analyses` and `datasets` are hidden aliases, as are `secret` for `secrets` and `ls` for `list`. `skills` is not the plural of `skill`: it is the group that serves the bundled skills to your coding agent. `agents` is not an alias: it is reserved for the managed-agents CLI and refuses, pointing at `evolve agent`.
+
+## Global options
+
+These work on every command.
+
+<ParamField path="--json">
+  Machine-readable JSON output. A refusal then prints one `{"error": {...}}` object on stdout with the server's `code`, `message`, `param`, `details`, `retryAfterSec` and `request_id` when present.
+</ParamField>
+
+<ParamField path="--api-key" type="key">
+  API key. Defaults to `$EVOLVE_API_KEY`.
+</ParamField>
+
+<ParamField path="--base-url" type="url">
+  API base URL. Defaults to the Evolve dashboard API.
+</ParamField>
+
+<ParamField path="-v, --version">
+  Print the CLI version.
+</ParamField>
+
+## Ids
+
+Every verb that takes an id (`job`, `trial`, `analysis`, `check` with a check id or a task check id, `skill`) also takes an unambiguous prefix of at least 8 characters. A prefix that matches nothing or more than one row is refused naming the candidates. A full id is never prefix-matched, `name:<skill-name>` is a name, and a job import id is taken as printed.
+
+A trial prefix is resolved against every trial of every job in scope, so on an account with many jobs the lookup can hit the rate limit; the full id, which `evolve job trials <job>` prints, always works.
+
+## Output
+
+Human tables on a TTY, tab-separated rows when piped, `--json` for the machine shape (NDJSON under `--watch`), and `-q` for ids only. `--no-trunc` disables cell truncation and `--no-headers` drops the header row from piped output. A money cell carries its state: `$0.06` is final, `at least $0.06` is still being written, `-` means not measured yet.
+
+`--columns` chooses and orders list columns, comma-separated; `--columns help` prints the keys. For `job list` they are `id`, `name`, `status`, `datasets`, `agents`, `trials`, `spent`, `started`; for `job trials` `task`, `agent`, `attempt`, `status`, `reward`, `spent`, `tokens`, `gpu`, `id`; for `analysis list` `id`, `status`, `task`, `job`, `trial`, `model`, `attempts`, `spent`, `created`, `finished`.
+
+## Listing flags
+
+Every `list` command, and `job trials`, `job tasks` and `job imports`, share one set of paging and formatting flags. Two take only the formatting flags, not `-l` or `--cursor`: `auth org list` and `secrets list`.
+
+<ParamField path="-l, --limit" type="n">
+  Page size.
+</ParamField>
+
+<ParamField path="--cursor" type="c">
+  Resume from a page cursor.
+</ParamField>
+
+<ParamField path="--columns" type="keys | all | help">
+  Choose and order columns, comma-separated. `help` lists the keys.
+</ParamField>
+
+<ParamField path="-q, --quiet">
+  Print only ids, one per line, for piping.
+</ParamField>
+
+<ParamField path="--no-trunc">
+  Full cell content instead of one-line truncation.
+</ParamField>
+
+<ParamField path="--no-headers">
+  Omit the header row in piped (TSV) output.
+</ParamField>
+
+## Scope
+
+`--scope` on `job list`, `analysis list` and `check list` is `my`, what you created and the default, or `shared`, your organizations' rows that teammates created. `all` is refused. Every id printed under either scope resolves on its `show`, `trace` and `download`; an id outside your organizations answers the noun's not-found code, exactly like an id that does not exist.
+
+## Values
+
+A flag's value may begin with `-`, such as `-x '-*'` or a negative number. Only a token that spells another flag of the same command is refused, and the refusal shows the `--flag=value` form. A typo in `--stream`, `--status`, `--scope` or `run`'s `-e` is a usage error naming the legal values, before any request.
+
+## Exit codes
+
+`0` on success (with `--watch`: the job `COMPLETED`, a publish settled `READY`, every analysis completed, no task check errored). `1` on a runtime or API failure (with `--watch`: `FAILED` or `CANCELLED`, a failed analysis, an errored task check). `2` on a usage error, and on a quota refusal, which prints `Launch quota exceeded:` with the server's sentence. A rate limit prints one line with the wait and exits 1; the SDK's watch loops wait it out on their own.

package/docs-evals/cli-reference/job.mdx

@@ -0,0 +1,239 @@
+---
+title: "evolve job"
+description: "Start, follow, and derive jobs."
+---
+
+import GlobalOptions from '/snippets/global-options.mdx';
+import ListFlags from '/snippets/list-flags.mdx';
+
+```bash
+evolve job <command> [options]
+```
+
+Every command below takes the [global options](/cli-reference#global-options).
+
+## job start
+
+```bash
+evolve job start [options]
+```
+
+Start a job; add `--watch` to follow it. Exactly the options of [`evolve run`](/cli-reference/run).
+
+## job list
+
+```bash
+evolve job list [options]
+```
+
+List jobs, newest first: yours, or `--scope shared` for your organizations'.
+
+<ListFlags />
+
+<ParamField path="--search" type="text">
+  Free-text filter over job name and dataset names.
+</ParamField>
+
+<ParamField path="--scope" type="my | shared">
+  Visibility scope: `my`, what you created (the default), or `shared`, your organizations' rows that teammates created.
+</ParamField>
+
+```bash
+evolve job list --limit 20 -q
+```
+
+## job show
+
+```bash
+evolve job show <id> [id...]
+```
+
+Show one or more jobs in full. It ends with a pass@k block, one line per arm and dataset that has numbers; `--json` carries the raw `stats.evals[].pass_at_k`. An `analyze` row holds the embedded policy and an `analysis` row the tally with the analyzer's spend. An uploaded job prints its provider as `ported` and its spend as `reported $X.XX (N/M trials reporting)`.
+
+## job trials
+
+```bash
+evolve job trials <id> [options]
+```
+
+List a job's trials.
+
+<ListFlags />
+
+<ParamField path="--status" type="s1,s2,...">
+  Filter by trial status, for example `INFRASTRUCTURE_ERROR`.
+</ParamField>
+
+<ParamField path="--dataset" type="name">
+  Filter to one dataset's trials.
+</ParamField>
+
+```bash
+evolve job trials 3e1f9a2c-… --status INFRASTRUCTURE_ERROR,SCORING_ERROR
+```
+
+## job tasks
+
+```bash
+evolve job tasks <id> [options]
+```
+
+Per-task rollup of a job: one row per task.
+
+<ListFlags />
+
+## job compare
+
+```bash
+evolve job compare <id> <id> [...]
+```
+
+Compare two to ten jobs side by side.
+
+## job cancel
+
+```bash
+evolve job cancel <id>
+```
+
+Request cancellation of a job.
+
+## job delete
+
+```bash
+evolve job delete <id> [options]
+```
+
+Permanently delete a job you created: its trials, traces, analyses and stored files. It names the job, then asks. Another member of the organization is refused with `org_forbidden`; a live job, a running analysis wave or a live regrade refuses as on [jobs](/core-concepts/jobs#stop-cancel-delete).
+
+<ParamField path="-y, --yes">
+  Delete without a confirmation prompt. On a non-interactive stdin the command refuses without it, rather than guessing. With `--json` the receipt is `{"job_id", "trials_deleted", "analyses_deleted"}`.
+</ParamField>
+
+## job stop
+
+```bash
+evolve job stop <id> --dataset <name>
+```
+
+Stop one dataset's live trials without cancelling the job. Trials already settled are reported as already terminal, and a dataset the job never spanned is refused. If the command fails part way (a rate limit, say), the report prints first, marked `PARTIAL` with the trials no answer came back for (`partial: true` and `unreported` under `--json`), then the error, exit 1; re-running the same command finishes the rest.
+
+<ParamField path="--dataset" type="name" required>
+  The dataset whose live trials to stop.
+</ParamField>
+
+## job resume
+
+```bash
+evolve job resume <id> [options]
+```
+
+A new linked job over a terminal job's failed or stopped trials.
+
+<ParamField path="-f, --filter-error-type" type="type">
+  Failure types to resume, matched on the trial's exception type. Repeatable. Defaults to the standard failure set plus stopped trials.
+</ParamField>
+
+```bash
+evolve job resume 3e1f9a2c-… -f InfrastructureError
+```
+
+## job retry
+
+```bash
+evolve job retry <id> [options]
+```
+
+A new linked job re-running selected trials: all of them, the failed ones only, or named ids.
+
+<ParamField path="--failed-only">
+  Only retry failed trials: `SCORING_ERROR`, `INFRASTRUCTURE_ERROR`, `BUDGET`, `INDETERMINATE`. Stopped and scored trials are not failures.
+</ParamField>
+
+<ParamField path="-t, --trial" type="trial-id">
+  Retry exactly this trial. Repeatable, all or nothing: each must be settled, and the job may still be running. Not combinable with `--failed-only`.
+</ParamField>
+
+## job regrade
+
+```bash
+evolve job regrade <id> [options]
+```
+
+Verifier-only re-run of a terminal job. The result is a job.
+
+<ParamField path="--status" type="s1,s2,...">
+  Only regrade source trials in these statuses.
+</ParamField>
+
+<ParamField path="--task" type="name">
+  Only regrade source trials of this task.
+</ParamField>
+
+## job imports
+
+```bash
+evolve job imports [options]
+```
+
+List your job imports (uploads), newest first. This is how an import id is found again.
+
+<ListFlags />
+
+<ParamField path="--status" type="QUEUED | RUNNING | COMPLETED | FAILED">
+  Filter by import status.
+</ParamField>
+
+## job import
+
+```bash
+evolve job import <import-id> [options]
+```
+
+Show one job import, an upload's record, with a `skipped` count and one line per skipped trial when the ingest left any out.
+
+<ParamField path="--watch">
+  Poll until the import settles: `COMPLETED`, and the job is printed, or `FAILED`.
+</ParamField>
+
+## job download
+
+```bash
+evolve job download <id> [options]
+```
+
+Download the job's results, unpacked as the standard job-directory tree plus `evolve.json` records.
+
+<ParamField path="-o, --output-dir" type="dir">
+  Directory to unpack into. Defaults to the current directory. The tree lands in `<dir>/job-<id>/`.
+</ParamField>
+
+<ParamField path="--overwrite">
+  Replace an existing `<dir>/job-<id>/`.
+</ParamField>
+
+## job grep
+
+```bash
+evolve job grep <id> <pattern> [options]
+```
+
+Search every trial's parsed trace in one server-side pass and print one line per matching trial with its sampled events. The pattern grammar and the page semantics are on [`jobs().grep()`](/sdk-reference/jobs#grep).
+
+<ParamField path="--type" type="event-type">
+  Only search events of exactly this type.
+</ParamField>
+
+<ParamField path="-l, --limit" type="n" default="50">
+  Per-trial match groups per page. Max 200.
+</ParamField>
+
+<ParamField path="--cursor" type="c">
+  Resume after this trial id, the previous page's `nextCursor`.
+</ParamField>
+
+```bash
+evolve job grep 3e1f9a2c-… 'permission denied'
+```
+
+<GlobalOptions />

package/docs-evals/cli-reference/run.mdx

@@ -0,0 +1,196 @@
+---
+title: "evolve run"
+description: "Start a job, and optionally follow it to the end."
+---
+
+import GlobalOptions from '/snippets/global-options.mdx';
+
+```bash
+evolve run [options]
+```
+
+Start a job. `evolve run` is the short form of `evolve job start`, and both take the same options.
+
+```bash
+evolve run \
+  -d terminal-bench-4@4.0 \
+  -a codex \
+  -m gpt-5.5 \
+  -k 2 \
+  -l 5 \
+  --watch
+```
+
+## What to run
+
+<ParamField path="-d, --dataset" type="name[@version]">
+  Dataset to run. Repeatable. A bare name means the active version.
+</ParamField>
+
+<ParamField path="-i, --include-task-name" type="glob">
+  Include filter over task names, applied to every dataset. Repeatable.
+</ParamField>
+
+<ParamField path="-x, --exclude-task-name" type="glob">
+  Exclude filter over task names, applied to every dataset. Repeatable.
+</ParamField>
+
+<ParamField path="-l, --n-tasks" type="n">
+  Cap the task count of each dataset after the filters.
+</ParamField>
+
+<ParamField path="-a, --agent" type="name[@version]">
+  Agent, built-in or registered.
+</ParamField>
+
+<ParamField path="-m, --model" type="name">
+  Model. Repeatable; each model is one arm.
+</ParamField>
+
+<ParamField path="-k, --n-attempts" type="n" default="1">
+  Attempts per task and arm.
+</ParamField>
+
+<ParamField path="-n, --n-concurrent" type="n" default="4">
+  Parallel trials, 1 to 150.
+</ParamField>
+
+<ParamField path="-e, --env" type="provider" default="daytona">
+  Sandbox provider: `e2b`, `daytona` or `modal`.
+</ParamField>
+
+<ParamField path="--job-name" type="name">
+  User-facing label. Generated by the server when omitted.
+</ParamField>
+
+## The arm
+
+Each of these applies to every arm of the job.
+
+<ParamField path="--effort" type="value">
+  Reasoning effort for every arm, applied verbatim. The values each harness accepts are on the [models](/core-concepts/models#reasoning-effort) page and in the [capability document](/sdk-reference/meta). An agent that cannot honor the value is refused, never silently skipped.
+</ParamField>
+
+<ParamField path="--preset" type="name">
+  Named settings preset for every arm: `no-internet` turns the vendor's server-side web tools off, `pinned-context` fixes the context window at 200000 tokens. Only `claude` and `codex` take one; a preset wins over `--ak config` where they disagree. An unknown name is refused with `invalid_input`, a preset the agent cannot guarantee with `agent_preset_unsupported`.
+</ParamField>
+
+<ParamField path="--agent-kwarg, --ak" type="key=value">
+  Agent kwarg for every arm, as `key=value`, repeatable. The delivered key is `config`: `--ak 'config=<path|inline JSON>'` becomes the harness's native settings file, your document as the base and the platform's routing on top; a path is read locally, as JSON or as TOML for a Codex config. Only `claude` and `codex` take a config. A key other than `config` is refused with `agent_kwarg_unsupported`, a config on another harness with `agent_config_unsupported`, and a key touching billing, base URL, routing or env with `agent_config_key_refused`.
+</ParamField>
+
+<ParamField path="--skills, --skill" type="ref | path">
+  Skill for every agent arm. Repeatable. Accepts `skills.sh/<owner>/<repo>[/<skill>]`, `org/repo[@ref]`, an https git URL, `upload:<id>`, `name:<skill-name>` (your moving name pointer, resolved server-side), or a local folder, which is uploaded first and then referenced; `--print-config` still shows the path you typed.
+</ParamField>
+
+<ParamField path="--ae, --agent-env" type="KEY=VALUE">
+  Sent as the job's `agent_env`, which the server refuses on this deployment: the create fails naming the two `--ve` keys it honors instead. Agent environment comes from the task's `[environment.env]` table or an attached `--secret`.
+</ParamField>
+
+<ParamField path="--ve, --verifier-env" type="KEY=VALUE">
+  Environment for every verifier run. Repeatable. The server honors exactly `REWARDKIT_JUDGE` and `REWARDKIT_MODEL`, rewardkit's judge override, and refuses any other key.
+</ParamField>
+
+<ParamField path="--secret" type="NAME[@LABEL][=ENVNAME]">
+  Attach one of your stored env secrets to every agent run. Repeatable. `NAME` is the stored secret's name.
+
+  `@LABEL` picks a labeled row; omitted, the `default` row or the only row, and several labels with no `default` is refused as ambiguous. `=ENVNAME` renames the env var inside the sandbox. References only: the value never rides the command line or the wire.
+</ParamField>
+
+<ParamField path="--secret-inline" type="NAME[@LABEL]:DELIVERY=VALUE">
+  Save `VALUE` into your vault as an env secret and attach it to this job in one step. Repeatable. `DELIVERY` is `brokered` or `direct` and sits before `=`, so everything after the first `=` is the value, byte for byte. `@LABEL` defaults to `default`.
+
+  Restating an existing secret exactly attaches it, so re-running the same command converges; a different value or delivery is refused as `secret_exists`. The job stores only the reference.
+</ParamField>
+
+## Spend and retries
+
+<ParamField path="--max-trial-spend" type="usd">
+  Model-spend cap for each trial. Defaults to the server's, $200.
+</ParamField>
+
+<ParamField path="-r, --max-retries" type="n">
+  Max automatic retries per trial on infrastructure errors. Defaults to 2; `0` turns retries off. Each attempt carries the full trial cap. Over a `-c` file it overrides only its own field of `retry`.
+</ParamField>
+
+<ParamField path="--retry-include" type="exception">
+  Exception types to retry on. Repeatable. The names are `AgentAuthenticationError` (a gateway credential refusal), `ModelNotFoundError` (a model that never served) and `InfrastructureError` (everything else). Defaults to everything `--retry-exclude` admits.
+</ParamField>
+
+<ParamField path="--retry-exclude" type="exception">
+  Exception types not to retry on. Repeatable, and it wins over `--retry-include`. Defaults to the platform's non-retryable set; an explicit `null` in a `-c` file's `retry` turns exclusions off entirely.
+</ParamField>
+
+## Embedded analysis
+
+<ParamField path="--analyze">
+  Analyze each trial's trace server-side as it settles. Cancelled trials are skipped. Bare, it uses the platform's analyze model, rubric and prompt.
+</ParamField>
+
+<ParamField path="--analyze-model" type="name">
+  Model the analyzer runs. Must be on the `claude` roster. Implies `--analyze`.
+</ParamField>
+
+<ParamField path="--analyze-rubric" type="path">
+  Rubric file for the analyzer: TOML, YAML or JSON in the `{criteria}` shape. Implies `--analyze`.
+</ParamField>
+
+<ParamField path="--analyze-prompt" type="path">
+  Prompt file for the analyzer. Replaces the built-in prompt. Implies `--analyze`.
+</ParamField>
+
+<ParamField path="--analyze-provider" type="provider">
+  Sandbox provider the analyzer runs on. Defaults to the platform's analysis default. Implies `--analyze`.
+</ParamField>
+
+<ParamField path="--analyze-effort" type="value">
+  Reasoning effort the analyzer runs at. Defaults to the per-model default: `high` on `openrouter/deepseek/deepseek-v4.1-flash`, `max` on `glm-5.3-flash`. Implies `--analyze`.
+</ParamField>
+
+## Timeouts
+
+<ParamField path="--timeout-multiplier" type="x" default="1.0">
+  Multiplier for task timeouts, any finite number above 0; below 1 shrinks. It multiplies each task's declared timeouts for this job only; the task is never rewritten. A product past 2,147,483,647 ms is refused at create. Resumed and retried jobs inherit the multipliers.
+</ParamField>
+
+<ParamField path="--agent-timeout-multiplier" type="x">
+  Multiplier for the agent execution timeout. Overrides `--timeout-multiplier`.
+</ParamField>
+
+<ParamField path="--verifier-timeout-multiplier" type="x">
+  Multiplier for the verifier timeout. Overrides `--timeout-multiplier`.
+</ParamField>
+
+<ParamField path="--agent-setup-timeout-multiplier" type="x">
+  Multiplier for the agent setup timeout. Overrides `--timeout-multiplier`.
+</ParamField>
+
+<ParamField path="--environment-build-timeout-multiplier" type="x">
+  Multiplier for the environment build timeout. Overrides `--timeout-multiplier`.
+</ParamField>
+
+## Config file and output
+
+<ParamField path="-c, --config" type="path">
+  Job config file, YAML or JSON with the same field names `jobs().start()` takes. Explicit flags override its fields one at a time, inside the `retry` and `analyze` objects too. The file may be partial: `-d` and `-a`/`-m` can supply what it omits, but a selector inside it needs `name` and an arm needs `name` and `model_name`.
+
+  YAML is read with the 1.1 schema (`yes` and `on` are booleans, `012` is octal), and a duplicate key, a second document or an unknown tag refuse with a line number. The body is checked before any request: an unknown key is refused by name, a string must be a string (quote `version: "1.10"`), `n_attempts` and `n_tasks` are at least 1, `n_concurrent_trials` is 1 to 150, `job_name` is at most 200 characters. A bare date or `.inf` is refused rather than rewritten.
+</ParamField>
+
+<ParamField path="--print-config">
+  Print the resolved job body as JSON and exit without running.
+</ParamField>
+
+<ParamField path="--watch">
+  Stream events until the job finishes.
+</ParamField>
+
+<ParamField path="-q, --quiet">
+  With `--watch`: suppress the event log and print the final block only.
+</ParamField>
+
+<ParamField path="-y, --yes">
+  Accepted for compatibility. Hosted runs have no prompt to confirm.
+</ParamField>
+
+<GlobalOptions />