Field catalog

Use this catalog after the conceptual guides. It is generated from the post-resolution Pydantic constructor models, including nested types. Required fields have no usable default. Custom cross-field validators also apply.

These schemas describe resolved object values, not Python references, source materialization, runtime capability checks, or side effects. The imperative public Job constructor is omitted because it is not a Pydantic model; use the Jobs guide and Python SDK guide for its source, Agent, mode, attempt, concurrency, retry, planning, and run arguments. Methods and Client request types belong in the API reference.

For Python-authored Environments, runtime is a required Environment parameter. The resolved schema below lists it as optional because it also describes stored definitions; use the Environments guide when creating an Environment.

Download the complete schemas.

Find a contract

• Environment
• Harness
• Task
• DeterministicVerifier
• AgentVerifier
• HumanVerifier
• Agent
• Benchmark
• AgentCredential
• AppAction
• AppDefinition
• BenchmarkCategory
• BenchmarkScoring
• DeclarativeImage
• Resource
• Runtime
• EvaluationTrack
• ExecutionLimits
• ExecutionTarget
• ExternalAgent
• FileDeclaration
• Guardrail
• HarnessCapability
• HarnessDefinition
• HarnessPolicy
• NetworkMode
• Party
• ResourceRequirements
• RubricCriterion
• RuntimePackages
• RuntimeVariable
• SandboxCapability
• Secret
• VerifierRuntime

Environment

Schema-v2 Environment: Apps composed with one Runtime and one set of limits.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• description — string; optional. Default: "".
• overview — string; optional. Default: "".
• readme — string; optional. Default: "".
• apps — array of AppDefinition; optional. Default: [].
• python — string | null; optional. Default: null.
• guardrails — array of Guardrail; optional. Default: [].
• resources — array of Resource; optional. Default: [].
• runtime — Runtime; optional.
• harness_policy — HarnessPolicy; optional.
• limits — ExecutionLimits; optional.
• party — Party | null; optional. Default: null. How many Agents share each episode, for a game such as poker. Leave it unset for an Environment one Agent works in alone.
• metadata — object; optional.

Harness

Base class for an Agent interaction loop.

• config — object; optional.

Task

One versioned unit of work in a Python-authored Environment.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• instructions — string; required. Constraints: {"minLength": 1}.
• goals — array of string; optional. Default: [].
• info — JSON value; optional. Default: null.
• metadata — object; optional.
• environment — JSON value; required.
• verifiers — array of DeterministicVerifier | AgentVerifier | HumanVerifier; required. Constraints: {"minItems": 1}.
• resources — array of Resource; optional. Default: [].
• initial_state — object; optional.
• seat_instructions — object; optional.
• expected — JSON value; optional. Default: null.

DeterministicVerifier

Code that verifies an episode: it scores the end, rewards actions, or both.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• info — JSON value; optional. Default: null.
• criteria — array of RubricCriterion; optional. Default: [].
• weight — number; optional. Default: 1. Constraints: {"exclusiveMinimum": 0}.
• metadata — object; optional.
• secrets — array of Secret; optional. Default: [].
• kind — "deterministic"; optional. Default: "deterministic".
• check — JSON value; optional. Default: null.
• reward — JSON value; optional. Default: null.
• hindsight — JSON value; optional. Default: null.
• runtime — VerifierRuntime; optional.
• result_path — string; optional. Default: "verifier-result.json".
• evidence_required — boolean; optional. Default: true.

AgentVerifier

Model-judge verifier with its own runtime and connectivity policy.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• info — JSON value; optional. Default: null.
• criteria — array of RubricCriterion; required. Constraints: {"minItems": 1}.
• weight — number; optional. Default: 1. Constraints: {"exclusiveMinimum": 0}.
• metadata — object; optional.
• secrets — array of Secret; optional. Default: [].
• kind — "agent"; optional. Default: "agent".
• model — string; required. Constraints: {"minLength": 1}.
• instructions — string; required. Constraints: {"minLength": 1}.
• provider — string | null; optional. Default: null.
• fallback_models — array of string; optional. Default: [].
• runtime — VerifierRuntime; optional.

HumanVerifier

Human review verifier.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• info — JSON value; optional. Default: null.
• criteria — array of RubricCriterion; required. Constraints: {"minItems": 1}.
• weight — number; optional. Default: 1. Constraints: {"exclusiveMinimum": 0}.
• metadata — object; optional.
• secrets — array of Secret; optional. Default: [].
• kind — "human"; optional. Default: "human".
• instructions — string; optional. Default: "".

Agent

A catalog-backed model and its optional execution harness.

• model — string; required. Constraints: {"minLength": 1}.
• name — string; optional. Default: "".
• version — string; optional. Default: "0.1.0".
• provider — string | null; optional. Default: null.
• instructions — string; optional. Default: "".
• fallback_models — array of string; optional. Default: [].
• temperature — number | null; optional. Default: null.
• max_tokens — integer | null; optional. Default: null.
• harness — HarnessDefinition | Harness | string | null; optional. Default: null.
• harness_config — object; optional. Options for a built-in Harness, such as reasoning_effort.
• auth — "environment" | "api_key" | "oauth" | "none"; optional. Default: "environment". How the Harness authenticates to the model; none when it calls no model.
• secrets — array of string; optional. Default: []. Names of the Harness's declared secrets this Agent grants it.
• endpoint — string | null; optional. Default: null.
• endpoint_key — string | null; optional. Default: null.
• external — ExternalAgent | null; optional. Default: null.
• credentials — object; optional.
• metadata — object; optional.

Benchmark

A semantic version that pins an ordered set of Tasks.

• categories — array of BenchmarkCategory; optional. Default: [].
• scoring — BenchmarkScoring; optional.
• tracks — array of EvaluationTrack; optional. Default: [].
• default_view — "models" | "agents"; optional. Default: "agents".
• purpose — string; optional. Default: "".
• success — string; optional. Default: "".
• limitations — array of string; optional. Default: [].
• license — string; optional. Default: "".
• forked_from — string | null; optional. Default: null.
• name — string; required. Constraints: {"minLength": 1}.
• version — string; required.
• tasks — array of Task; required. Constraints: {"minItems": 1}.
• primary_metric — "score"; optional. Default: "score".
• description — string; optional. Default: "".
• metadata — object; optional.

AgentCredential

A credential this Agent supplies to the Apps it works with.

• scopes — array of string; optional. Default: [].
• description — string; optional. Default: "".

AppAction

One move an App offers the Agent.

• name — string; required. Constraints: {"maxLength": 128, "minLength": 1}.
• description — string; optional. Default: "".
• parameters — object; optional.
• mutates_state — boolean; optional. Default: true.
• timeout_seconds — number; optional. Default: 30. Constraints: {"exclusiveMinimum": 0}.
• binding — object; optional.

AppDefinition

One App inside an Environment.

• name — string; required.
• kind — "python" | "mcp" | "openapi" | "replay"; optional. Default: "python".
• description — string; optional. Default: "".
• python — string | null; optional. Default: null.
• config — object; optional.
• actions — array of AppAction; optional. Default: [].
• secrets — array of Secret; optional. Default: [].
• credentials — array of Secret; optional. Default: [].
• state_schema — object; optional.
• has_reset — boolean; optional. Default: false.
• has_snapshot — boolean; optional. Default: false.
• control — "owned" | "sandboxed" | "live" | "replayed"; optional. Default: "owned".

BenchmarkCategory

A named group of Tasks, scored together in results.

• id — string; required.
• name — string; required. Constraints: {"minLength": 1}.
• description — string; optional. Default: "".
• tasks — array of string; required. Constraints: {"minItems": 1}.
• taxonomy — string | null; optional. Default: null.

BenchmarkScoring

How a release turns attempts into one score per configuration.

• metric — string; optional. Default: "Score".
• description — string; optional. Default: "".
• aggregation — "weighted_mean"; optional. Default: "weighted_mean".
• score_range — array of JSON value; optional. Default: [0.0, 1.0]. Constraints: {"maxItems": 2, "minItems": 2}.
• success_threshold — number | null; optional. Default: 1.0.
• task_weights — object; optional.
• coverage_required — number; optional. Default: 1.0. Constraints: {"exclusiveMinimum": 0, "maximum": 1}.
• agent_failure — "zero" | "exclude"; optional. Default: "zero".
• infrastructure_error — "exclude"; optional. Default: "exclude".
• harness_timeout — "infrastructure_error" | "agent_failure"; optional. Default: "infrastructure_error".
• min_tasks_for_interval — integer; optional. Default: 5. Constraints: {"minimum": 2}.

DeclarativeImage

Portable subset of Daytona's declarative image builder.

• base — string; required. Constraints: {"minLength": 1}.
• pip_packages — array of string; optional. Default: [].
• environment — object; optional.
• workdir — string | null; optional. Default: null.

Resource

Data or application supplied by an Environment.

• kind — "data" | "application" | "file"; required.
• name — string; required. Constraints: {"minLength": 1}.
• path — string | null; optional. Default: null.
• uri — string | null; optional. Default: null.
• digest — string | null; optional. Default: null.
• content_type — string; optional. Default: "".
• config — object; optional.
• description — string; optional. Default: "".
• delivery — "descriptor" | "source" | "inline" | "resolver" | "blob"; optional. Default: "descriptor".
• content — string | null; optional. Default: null.
• resolver — string | null; optional. Default: null.

Runtime

Where an Environment runs: Docker, a trusted local process, or a remote sandbox.

• provider — string; optional. Default: "docker". Constraints: {"minLength": 1}.
• ref — string | null; optional. Default: null. Slug of the project Runtime this configuration was copied from. The machine fields hold that Runtime's settings as of the copy; credentials come from the project Runtime when a Job starts.
• placement — object; optional.
• variables — array of RuntimeVariable; optional. Default: [].
• image — string | null; optional. Default: null.
• snapshot — string | null; optional. Default: null.
• declarative_image — DeclarativeImage | null; optional. Default: null.
• build_context — string | null; optional. Default: null.
• dockerfile — string | null; optional. Default: null.
• network — NetworkMode; optional. Default: "public".
• allowed_hosts — array of string; optional. Default: []. Hosts the Runtime may reach when network is 'allowlist'.
• compute — ResourceRequirements; optional. CPU, memory, storage, and process limits for the Runtime.
• packages — RuntimePackages; optional. Packages installed in the Runtime before an episode. In environment.yaml, list Python ones under python_packages. On the local Runtime they must already be installed.
• read_only_root — boolean; optional. Default: false.
• targets — array of ExecutionTarget; optional. Constraints: {"uniqueItems": true}.
• persistent — boolean; optional. Default: false.
• compose — boolean; optional. Default: false.
• extra_capabilities — array of SandboxCapability; optional. Default: []. Constraints: {"uniqueItems": true}.
• timeout_seconds — number; optional. Default: 300. Constraints: {"exclusiveMinimum": 0}.
• build_timeout_seconds — number; optional. Default: 600. Constraints: {"exclusiveMinimum": 0}.
• allow_unsafe_local — boolean; optional. Default: false.

EvaluationTrack

Versioned rules that make results on one track comparable.

• id — string; required.
• version — string; optional. Default: "1.0.0".
• name — string; required. Constraints: {"minLength": 1}.
• kind — "models" | "agents"; required.
• description — string; optional. Default: "".
• attempts — integer; optional. Default: 1. Constraints: {"maximum": 32, "minimum": 1}.
• max_retries — integer; optional. Default: 0. Constraints: {"maximum": 10, "minimum": 0}.
• harnesses — array of string; optional. Default: [].
• instructions — "none" | "any"; optional. Default: "any".
• temperature — number | null; optional. Default: null.
• max_tokens — integer | null; optional. Default: null.
• tools — string; optional. Default: "Environment actions only.".
• max_turns — integer | null; optional. Default: null.
• max_seconds — number | null; optional. Default: null.
• max_cost_usd — number | null; optional. Default: null.

ExecutionLimits

Environment execution limits.

• max_turns — integer; optional. Default: 8. Constraints: {"maximum": 1000, "minimum": 1}.
• max_seconds — number; optional. Default: 120. Constraints: {"exclusiveMinimum": 0}.
• max_cost_usd — number | null; optional. Default: null.

ExecutionTarget

Execution isolation class.

Allowed values: "local" | "docker" | "remote".

ExternalAgent

An agent that already runs as a service of its own.

• url — string; required.
• signing_key — string; required.
• version — string; optional. Default: "".
• ca_bundle — string | null; optional. Default: null.

FileDeclaration

A file emitted or consumed by a package.

• path — string; required. Constraints: {"minLength": 1}.
• required — boolean; optional. Default: true.
• media_type — string; optional. Default: "application/octet-stream".

Guardrail

A plain-language Environment rule.

• name — string; optional. Default: "".
• rule — string; required. Constraints: {"minLength": 1}.

HarnessCapability

A capability requested by an agent harness.

Allowed values: "shell" | "file_read" | "file_edit" | "code_execution" | "web_search" | "browser" | "network_fetch" | "mcp" | "subagents" | "persistence".

HarnessDefinition

A declared executable Agent interaction strategy.

• name — string; required. Constraints: {"minLength": 1}.
• version — string; optional. Default: "0.1.0".
• description — string; optional. Default: "".
• command — array of string; required. Constraints: {"minItems": 1}.
• source — string; optional. Default: ".".
• digest — string | null; optional. Default: null.
• requirements — array of string; optional. Default: [].
• capabilities — array of HarnessCapability; optional. Default: []. Constraints: {"uniqueItems": true}.
• models — array of string; optional. Default: ["*"].
• auth — array of "environment" | "api_key" | "oauth" | "none"; optional. Default: ["environment"].
• secrets — array of string; optional. Default: [].
• env_vars — array of string; optional. Default: []. Names of environment variables passed through to the Harness process.
• healthcheck — array of string | null; optional. Default: null.
• outputs — array of FileDeclaration; optional. Default: [].
• artifacts — array of FileDeclaration; optional. Default: [].
• trajectory — string | null; optional. Default: null.
• tito — string | null; optional. Default: null.

HarnessPolicy

Environment ceiling for a Trial's optional Agent harness.

• mode — "allow_all" | "allowlist"; optional. Default: "allow_all".
• allowed_harnesses — array of string; optional. Default: [].
• allowed_capabilities — array of HarnessCapability | null; optional. Default: null.
• denied_capabilities — array of HarnessCapability; optional. Default: []. Constraints: {"uniqueItems": true}.

NetworkMode

Requested sandbox network policy.

Allowed values: "public" | "no-network" | "allowlist" | "public" | "no-network" | "allowlist".

Party

How many Agents may share one episode of the Environment, such as 2 to 8 at poker.

• min — integer; optional. Default: 2. Constraints: {"maximum": 100, "minimum": 2}.
• max — integer; optional. Default: 2. Constraints: {"maximum": 100, "minimum": 2}.
• isolation — "isolated" | "shared"; optional. Default: "isolated". `isolated: each seat's Harness runs as its own OS user, and the Apps are the only thing the seats share; needs Docker or Daytona. shared`: every seat runs as one user and can read and write the others' files and the Environment's, for a team that works on the same files, or to try a party on the local Runtime. A shared seat could also act for another seat.

ResourceRequirements

Optional compute limits.

• cpu — number | null; optional. Default: null.
• memory_mb — integer | null; optional. Default: null.
• pids — integer | null; optional. Default: null.
• disk_mb — integer | null; optional. Default: null.
• storage_mb — integer | null; optional. Default: null.

RubricCriterion

One Agent or Human rubric criterion.

• name — string; required. Constraints: {"minLength": 1}.
• description — string; required. Constraints: {"minLength": 1}.
• weight — number; optional. Default: 1. Constraints: {"exclusiveMinimum": 0}.
• min_score — number; optional. Default: 0.
• max_score — number; optional. Default: 1.

RuntimePackages

Packages installed in the Runtime before an episode, by package manager.

• python — array of string; optional. Default: [].

RuntimeVariable

A launch-time variable declaration. Values are supplied to Job, never saved here.

• name — string; required. Constraints: {"pattern": "^[A-Z_][A-Z0-9_]*$"}.
• description — string; optional. Default: "".
• required — boolean; optional. Default: true.
• secret — boolean; optional. Default: false.
• format — "text" | "url" | "integer" | "json"; optional. Default: "text".

SandboxCapability

Individual controls a provider can enforce.

Allowed values: "image" | "build" | "resources" | "network_none" | "network_allowlist" | "upload" | "download" | "timeout" | "cancel" | "persistence" | "compose" | "read_only_root" | "working_directory" | "environment" | "log_capture".

Secret

A named value an App needs, which it never stores itself.

• name — string; required. Constraints: {"pattern": "^[A-Z_][A-Z0-9_]*$"}.
• required — boolean; optional. Default: true.
• description — string; optional. Default: "".
• scopes — array of string; optional. Default: [].

VerifierRuntime

Machine for a verifier that needs its own sandbox.

• provider — string; optional. Default: "docker". Constraints: {"minLength": 1}.
• image — string | null; optional. Default: null.
• network — NetworkMode; optional. Default: "public".
• allowed_hosts — array of string; optional. Default: [].
• compute — ResourceRequirements; optional.
• timeout_seconds — number; optional. Default: 60. Constraints: {"exclusiveMinimum": 0}.