Sessions

A Claude model id. The model must be permitted as an advisor for this agent's model.

The id the buffered agent.message will carry if it is emitted. Matches the event_id on this preview's event_delta events.

The agent ID.

The specific agent version to use. Omit to use the latest version. Must be at least 1 if specified.

The id the buffered agent.thinking will carry if it is emitted. Start-only — no event_delta events follow.

The agent ID.

Replacement MCP server list. Full replacement: the provided array becomes the MCP servers. Send an empty array to clear; omit to preserve the agent's servers.

Replacement model. Accepts the model string, e.g. claude-opus-5-5, or a model_config object. Omit to use the agent's model.

Replacement skill list. Full replacement: the provided array becomes the skills. Send an empty array to clear; omit to preserve the agent's skills.

Replacement system prompt. Up to 100,000 characters. Set to null to clear the agent's system prompt; omit to preserve it.

Replacement tool list. Full replacement: the provided array becomes the tool configuration. Send an empty array to clear; omit to preserve the agent's tools.

The specific agent version to use. Omit to use the latest version.

Branch name to check out.

Maximum list cost the session may accrue. List price is used regardless of any negotiated discount, so the cap fires at or before the actual charge.

Tokens used to create 1-hour ephemeral cache entries.

Tokens used to create 5-minute ephemeral cache entries.

Full commit SHA to check out.

A partial element of the content array at index, typed like the element itself — the same shape the buffered agent.message carries in content.

Which entry in the previewed event's content array this fragment lands in. Insert content as that entry when the index is new; append to the existing entry otherwise.

One fragment of the previewed event. The delta type is named for the previewed event's field it streams into: agent.message events stream content_delta fragments, each a partial element of the content array.

The id of the event being previewed. Matches event.id on the corresponding event_start and the buffered event that reconciles the preview.

ID of a previously uploaded file.

Mount path in the container. Defaults to /mnt/session/uploads/<file_id>.

Github URL of the repository

GitHub authorization token used to clone the repository. Required for private repositories; optional for public ones.

Branch or commit to check out. Defaults to the repository's default branch.

Mount path in the container. Defaults to /workspace/<repo-name>.

The memory store ID (memstore_...). Must belong to the caller's organization and workspace.

Access mode for the mounted store. Defaults to read_write. read_only mounts the store as a read-only filesystem.

Per-attachment guidance for the agent on how to use this store. Rendered into the memory section of the system prompt. Max 4096 chars.

When the outcome reached a terminal result. Null while pending/running/evaluating.

What the agent should produce.

Grader's verdict text from the most recent evaluation. For satisfied, explains why criteria are met; for needs_revision (intermediate), what's missing; for failed, why unrecoverable.

0-indexed revision cycle the outcome is currently on.

Server-generated outc_ ID for this outcome.

Current evaluation state. pending before the agent begins work; running while producing or revising; evaluating while the grader scores; satisfied/max_iterations_reached/failed/interrupted are terminal.

Number of server-executed web fetch requests.

Number of server-executed web search requests.

Execution principal snapshotted from the agent at session creation. Always present; {"type": "service_account"} when runs carry no customer-named principal.

Model identifier and configuration.

Resolved multiagent orchestration configuration. Null when the agent is single-threaded.

Replacement MCP server list. Full replacement: the provided array becomes the new value. Send an empty array to clear; omit to preserve.

Replacement tool list. Full replacement: the provided array becomes the new value. Send an empty array to clear; omit to preserve.

Full agent definitions the coordinator may spawn as session threads.

Whether the agent can define inline agents, which are not saved, when it spawns session threads.

Full agent definitions of the predefined agents, which are saved agents that this agent can spawn as session threads.

Whether a run's plan can define inline agents, which are not saved.

Full agent definitions of the predefined agents, which are saved agents that a run's plan can use.

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

Whether the agent can spawn session threads.

Whether the agent can start workflow runs.

Cumulative time in seconds the session spent in running status. Excludes idle time.

Elapsed time since session creation in seconds. For terminated sessions, frozen at the final update.

Unique identifier for this event.

Timestamp when the update was applied.

The session's effective agent configuration after the update. Present only when the update changed agent (tools or mcp_servers); when present it is the full materialised snapshot, not a diff.

The session's budget after the update: the new budget when set or replaced, or null when the update removed it. Present only when the update changed the budget.

The session's full metadata bag after the update. Present when the update set non-empty metadata; absent when metadata was unchanged or cleared to empty.

The session's new title. Present only when the update changed it.

Cumulative time in seconds during which the session had at least one thread in running status. Overlapping activity from concurrent threads is counted once, unlike stats.active_seconds, which sums each thread's own active time. This is the duration the session's runtime cost is priced on.

Tokens used to create prompt cache entries, broken down by cache TTL.

Total tokens read from prompt cache.

Total input tokens consumed across all turns.

Cumulative list cost of the session across all turns, priced at public list rates. Absent until cost tracking is available for the session.

Total output tokens generated across all turns.

Cumulative server-executed tool usage across all turns. Absent until server-tool tracking is available for the session.

Unique identifier for this event.

Timestamp when the snapshot was taken.

The session's cumulative usage at the snapshot time.

The session's configured budget at the snapshot time, or null when the session has no budget.

The previewed event's type and id. The event type determines which delta types the preview's event_delta events carry: agent.message events stream content_delta fragments; agent.thinking previews are start-only — no deltas follow, and the buffered agent.thinking with the same id concludes them.

The text content.

Unique identifier for this event.

System content blocks. Text-only.

Timestamp when this system message was processed.

Unique identifier for this event.

The id of the agent.tool_use event this result corresponds to, which can be found in the last session.status_idle event's stop_reason.event_ids field.

The result content returned by the tool.

Whether the tool execution resulted in an error.

Timestamp when this result was processed.

Set by the server to the subagent thread this result was routed to. Omitted when it was routed to the primary thread.