Agen

Platform advisor roster entry: a model the session's primary thread may consult mid-turn.

A Managed Agents agent.

A resolved agent reference with a concrete version.

Configuration for a specific agent tool.

Configuration override for a specific tool within a toolset.

Resolved default configuration for agent tools.

Default configuration for all tools in a toolset.

Input payload for the bash tool of the agent_toolset_20260401 toolset. All fields are optional; a normal invocation supplies command, while restart=true (with no command) reboots the runner-side bash session.

Input payload for the edit tool. Performs a string replacement in the named file; by default old_string must occur exactly once.

Input payload for the glob tool. Returns paths matching a doublestar glob pattern, newest first.

Input payload for the grep tool. Searches file contents for a regular expression, returning matching lines.

Configuration for built-in agent tools. Use this to enable or disable groups of tools available to the agent.

Input payload for the read tool. Reads file contents relative to the runner's working directory (or absolute when the runner permits).

Input payload for the write tool. Writes (overwriting) the entire file contents.

Tool calls are automatically approved without user confirmation.

Tool calls require user confirmation before execution.

A resolved Anthropic-managed skill.

An Anthropic-managed skill.

The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.

Configuration for the bash tool.

Configuration override for the bash tool.

A resolved user-created custom skill.

A user-created custom skill.

A custom tool as returned in API responses.

JSON Schema for custom tool input parameters.

A custom tool that is executed by the API client rather than the agent. When the agent calls this tool, an agent.custom_tool_use event is emitted and the session goes idle, waiting for the client to provide the result via a user.custom_tool_result event.

Configuration for the edit tool.

Configuration override for the edit tool.

High effort. Favors reasoning depth.

Low effort. Favors latency over reasoning depth.

Maximum effort. Favors reasoning depth over latency.

Medium effort. Balances latency and reasoning depth.

Extra-high effort. Not all models accept this level.

Configuration for the glob tool.

Configuration override for the glob tool.

Configuration for the grep tool.

Configuration override for the grep tool.

URL-based MCP server connection as returned in API responses.

Resolved configuration for a specific MCP tool.

Configuration override for a specific MCP tool.

Resolved default configuration for all tools from an MCP server.

Default configuration for all tools from an MCP server.

Configuration for tools from an MCP server defined in mcp_servers.

The model that will power your agent.

Model identifier and configuration.

An object that defines additional configuration control over model use

Whether the session's primary thread can consult an advisor model.

The agent has no advisor.

The agent has no advisor.

The session's primary thread can consult model mid-turn.

The session's primary thread can consult model mid-turn.

Whether the session's primary thread can consult an advisor model.

Resolved coordinator topology with a concrete agent roster.

A coordinator topology: the session's primary thread orchestrates work by spawning session threads, each running an agent drawn from the agents roster.

Whether the agent can define inline agents. The agent defines an inline agent itself, in a workflow run's plan or when it spawns a session thread, and the inline agent is not saved.

The agent cannot define inline agents.

The agent cannot define inline agents.

The agent can define inline agents.

The agent can define inline agents.

Whether the agent can define inline agents. The agent defines an inline agent itself, in a workflow run's plan or when it spawns a session thread, and the inline agent is not saved.

One agent in a predefined_agents list. It is an agent ID string, an agent reference with an optional version, or self for the agent that owns this configuration.

Sentinel roster entry meaning "the agent that owns this configuration". Resolved server-side to a concrete agent reference.

Whether the agent can spawn session threads.

The agent cannot spawn session threads.

The agent cannot spawn session threads.

The agent can spawn session threads.

The agent can spawn session threads. Each thread runs a predefined agent, which is a saved agent in predefined_agents, or an inline agent, which the agent defines when it spawns the thread and which is not saved. If inline_agents is disabled, predefined_agents must name at least one agent.

Whether the agent can spawn session threads.

Whether the agent can start workflow runs.

The agent cannot start workflow runs.

The agent cannot start workflow runs.

The agent can start workflow runs.

The agent can start workflow runs. Each run follows a plan, a program that the agent writes. A plan can use predefined agents, which are the saved agents in predefined_agents, and inline agents, which it defines itself and which are not saved. If inline_agents is disabled, predefined_agents must name at least one agent.

Whether the agent can start workflow runs.

Resolved multiagent configuration with three members, each enabled or disabled on its own.

Multiagent configuration with three members, each enabled or disabled on its own. On an update, if the agent's stored multiagent also has type multiagent_20261001, this configuration is merged into the stored one, level by level, instead of replacing it. A key that the update omits keeps its stored value. A key sent as null takes its default, on create as well, so "workflows": null enables workflows. An object sent with a type other than the stored one replaces the stored object, and the keys that it omits take their defaults. A predefined_agents list that is sent replaces the stored list. Every object that is sent needs its type, and an enabled advisor needs its model. Other validation applies to the merged result.

Configuration for the read tool.

Configuration override for the read tool.

Resolved agent definition for a single session_thread. Snapshot of the agent at thread creation time. The multiagent roster is not repeated here; read it from Session.agent.

Skill to load in the session container.

URL-based MCP server connection.

Approximate user location for search result localization.

Configuration for the web_fetch tool.

Configuration override for the web_fetch tool.

Every URL from this source may be fetched. This is the default.

Every tool's results contribute URLs that may be fetched, except the named tools' results.

This source contributes no URLs that may be fetched.

Only the named tools' results contribute URLs that may be fetched.

String form of a url_sources value that has no field other than its type: "all" means {"type": "all"} and "none" means {"type": "none"}.

Which tools' results contribute URLs that may be fetched.

Which tools' results contribute URLs that may be fetched. Accepts the string "all" or "none", or an object whose type is "all", "none", "only" or "except". Responses use the object form.

Names one tool in an only or except list.

Whether URLs in the text of user messages may be fetched.

Whether URLs in the text of user messages may be fetched. Accepts the string "all" or "none", or the object {"type": "all"} or {"type": "none"}. Responses use the object form.