AgentEventDetails

Parameters and properties

The fields below cover all variants; the signature specifies their allowed combinations.

• kindRequired
  "model-route" | "subagent" | "message-usage" | "stderr" | "stopped" | "steer" | "reasoning" | "file-change" | "model-request" | "model-response" | "model-retry" | "model-error" | "hook" | "instructions-loaded" | "skills-loaded" | "tool-output" | "phase" | "summary" | "warning" | "text" | "text-delta" | "result" | "prompt" | "tool" | "tool-result" | "step" | "tool-denied" | "stop-prevented" | "compaction" | "conversation" | "usage" | "failure" | "quota" | "fallback" | "finished" | "raw"
  Event type selecting the payload: phase, summary, warning, text, text-delta, result, prompt, tool, tool-result, tool-output, tool-denied, file-change, reasoning, step, hook, instructions-loaded, skills-loaded, model-route, model-request, model-response, model-retry, model-error, stop-prevented, steer, compaction, subagent, conversation, usage, message-usage, quota, fallback, stderr, stopped, failure, finished or raw.
• stepVariant-dependent
  number
  Harness step for which this effective model was selected.
• choiceVariant-dependent
  string
  Effective candidate key after confidence and availability fallback rules.
• modelVariant-dependent
  AgentModel
  Selected normalized model with its own reasoning and output settings.
• reasonVariant-dependent
  "unavailable" | "selected" | "confidence" | "completion" | "idle-timeout" | "deadline" | "aborted" | "oversized-event" | "steered" | string
  On tool-denied, why the call was refused. On stopped, why Outpost ended the process: completion, idle-timeout, deadline, aborted, oversized-event or steered. On model-route, selected, confidence or unavailable explains the effective selection.
• confidenceVariant-dependent
  number | undefined
  Native decision confidence, present only when a valid choice response was received.
• idVariant-dependent
  string
  Conversation identifier for conversation events, or unique child execution identifier for subagent lifecycle events.
• callIdVariant-dependent
  string | string | undefined | string | string | undefined | string | string
  Tool call identifier linking the tool, tool-result, tool-output, tool-denied, file-change and subagent events of one call. CLI adapters copy the native identifier when the protocol has one.
• nameVariant-dependent
  string
  Phase name on phase events; tool name on tool, tool-result and tool-denied events; delegating tool name on subagent events.
• statusVariant-dependent
  "started" | "finished" | "failed" | number
  On summary, the exit status of the pass’s last turn, always 0 because a nonzero exit rejects the dispatch. On subagent, started, finished or failed.
• conversationVariant-dependent
  string | undefined
  Persisted child conversation identifier when the child enables transcript storage.
• tokensVariant-dependent
  Usage
  Token counts: an increment on usage, the pass total on summary, one message’s usage on message-usage.
• messageIdVariant-dependent
  string | undefined
  Claude Code identifier of the message whose usage a message-usage event reports.
• parentCallIdVariant-dependent
  string | undefined
  Parent tool call supplied by a CLI for a nested agent message or tool event.
• textVariant-dependent
  string
  Text of the event according to kind: agent text, streamed fragment, final answer, rendered prompt, steering instruction, reasoning, stderr line or tool output chunk.
• truncatedVariant-dependent
  boolean | undefined
  Whether the stderr fragment or oversized raw preview was bounded before delivery.
• modeVariant-dependent
  SteeringMode
  How a steer event reached the agent: injected into the running turn, or resumed in a new turn of the same conversation.
• changesVariant-dependent
  unknown
  File changes as Codex reports them, passed through unmodified.
• requestVariant-dependent
  unknown
  Model request observed only when the hub explicitly enables verbose payloads; may contain private instructions and messages.
• responseVariant-dependent
  unknown
  Model response observed only when verbose payloads are enabled, including tool calls and replayable content.
• attemptVariant-dependent
  number
  Retry attempt explicitly reported by a model provider; Outpost does not infer hidden retries.
• messageVariant-dependent
  string | undefined | string | string | string | string | string | string
  Message of a warning, failure, quota, fallback, model-retry or model-error event, or the message a stop hook sent back to the model on stop-prevented.
• phaseVariant-dependent
  string
  Harness hook phase that completed.
• changedVariant-dependent
  boolean
  Whether a hook returned a decision or changed its observable input.
• countVariant-dependent
  number
  Number of nonempty instruction sources resolved for this harness turn.
• namesVariant-dependent
  readonly string[]
  Skills newly loaded into the conversation and available to subsequent model steps.
• channelVariant-dependent
  "stdout" | "stderr"
  Command output stream for the tool-output event: stdout or stderr.
• agentVariant-dependent
  string | undefined
  Name of the agent running the pass, on phase events: claude, codex, antigravity, copilot or kimi, custom for a built-in harness agent, replay for a replay agent.
• branchVariant-dependent
  string | undefined
  Name of the work branch used or observed during execution.
• directoryVariant-dependent
  string | undefined
  Host workspace directory used for this execution.
• durationMsVariant-dependent
  number
  Duration of the pass in milliseconds, summed over its turns, on summary events.
• inputVariant-dependent
  unknown
  Raw arguments supplied to the tool named by this tool event.
• isErrorVariant-dependent
  boolean
  Whether the tool call failed or reported an error.
• previewVariant-dependent
  string
  First 2000 characters of the tool result.
• charactersVariant-dependent
  number
  Full length of the tool result before any truncation.
• indexVariant-dependent
  number
  One-based step number within a built-in harness turn.
• strategyVariant-dependent
  string
  Name of the context strategy that rewrote the history.
• messagesVariant-dependent
  number
  Number of messages in the history after compaction.
• cumulativeVariant-dependent
  boolean | undefined
  True when an adapter reports a running total for the native session instead of an increment. Outpost subtracts the session baseline and the usage already counted, so observers always receive usage events with increments and no cumulative flag.
• resetAtVariant-dependent
  string | undefined
  ISO timestamp at which a quota or fallback event says the usage or rate limit resets; present only when reported in structured form.
• fromVariant-dependent
  FallbackCandidate
  Candidate that stopped in a fallback event: its position, adapter name and model name when one was selected.
• toVariant-dependent
  FallbackCandidate
  Candidate that takes over in a fallback event: its position, adapter name and model name when one was selected.
• failureVariant-dependent
  FallbackTrigger
  Category that ended the previous candidate in a fallback event: quota or unavailable.
• valueVariant-dependent
  unknown
  Raw protocol line as the agent printed it; for an oversized line, only its first 2000 characters.
• bytesVariant-dependent
  number | undefined
  Observed UTF-8 byte size of an oversized protocol line or buffered prefix before termination.

Signature

export type AgentEventDetails =
  | ModelRouteEvent
  | {
      readonly kind: "subagent";
      readonly id: string;
      readonly callId: string;
      readonly name: string;
      readonly status: "started" | "finished" | "failed";
      readonly conversation?: string;
    }
  | {
      readonly kind: "message-usage";
      readonly tokens: Usage;
      readonly messageId?: string;
      readonly parentCallId?: string;
    }
  | {
      readonly kind: "stderr";
      readonly text: string;
      readonly truncated?: boolean;
    }
  | {
      readonly kind: "stopped";
      readonly reason:
        | "completion"
        | "idle-timeout"
        | "deadline"
        | "aborted"
        | "oversized-event"
        | "steered";
    }
  | {
      readonly kind: "steer";
      readonly text: string;
      readonly mode: SteeringMode;
    }
  | {
      readonly kind: "reasoning";
      readonly text: string;
      readonly parentCallId?: string;
    }
  | {
      readonly kind: "file-change";
      readonly changes: unknown;
      readonly callId?: string;
    }
  | {
      readonly kind: "model-request";
      readonly request: unknown;
    }
  | {
      readonly kind: "model-response";
      readonly response: unknown;
    }
  | {
      readonly kind: "model-retry";
      readonly attempt: number;
      readonly message?: string;
    }
  | {
      readonly kind: "model-error";
      readonly message: string;
    }
  | {
      readonly kind: "hook";
      readonly phase: string;
      readonly changed: boolean;
    }
  | {
      readonly kind: "instructions-loaded";
      readonly count: number;
    }
  | {
      readonly kind: "skills-loaded";
      readonly names: readonly string[];
    }
  | {
      readonly kind: "tool-output";
      readonly callId: string;
      readonly channel: "stdout" | "stderr";
      readonly text: string;
    }
  | {
      readonly kind: "phase";
      readonly name: string;
      readonly agent?: string;
      readonly branch?: string;
      readonly directory?: string;
    }
  | {
      readonly kind: "summary";
      readonly durationMs: number;
      readonly status: number;
      readonly tokens: Usage;
    }
  | {
      readonly kind: "warning";
      readonly message: string;
    }
  | {
      readonly kind: "text";
      readonly text: string;
    }
  | {
      readonly kind: "text-delta";
      readonly text: string;
    }
  | {
      readonly kind: "result";
      readonly text: string;
    }
  | {
      readonly kind: "prompt";
      readonly text: string;
    }
  | {
      readonly kind: "tool";
      readonly name: string;
      readonly input: unknown;
      readonly callId?: string;
      readonly parentCallId?: string;
    }
  | {
      readonly kind: "tool-result";
      readonly callId: string;
      readonly name: string;
      readonly isError: boolean;
      readonly preview: string;
      readonly characters: number;
      readonly parentCallId?: string;
    }
  | {
      readonly kind: "step";
      readonly index: number;
    }
  | {
      readonly kind: "tool-denied";
      readonly callId: string;
      readonly name: string;
      readonly reason: string;
    }
  | {
      readonly kind: "stop-prevented";
      readonly message: string;
    }
  | {
      readonly kind: "compaction";
      readonly strategy: string;
      readonly messages: number;
    }
  | {
      readonly kind: "conversation";
      readonly id: string;
    }
  | {
      readonly kind: "usage";
      readonly tokens: Usage;
      readonly cumulative?: boolean;
    }
  | {
      readonly kind: "failure";
      readonly message: string;
    }
  | {
      readonly kind: "quota";
      readonly message: string;
      readonly resetAt?: string;
    }
  | {
      readonly kind: "fallback";
      readonly from: FallbackCandidate;
      readonly to: FallbackCandidate;
      readonly failure: FallbackTrigger;
      readonly message: string;
      readonly resetAt?: string;
    }
  | {
      readonly kind: "finished";
    }
  | {
      readonly kind: "raw";
      readonly value: unknown;
      readonly bytes?: number;
      readonly truncated?: boolean;
    };

Related contracts

• FallbackCandidate
• FallbackTrigger
• ModelRouteEvent
• SteeringMode
• Usage