Struct ToolExecutionOutcome 

pub struct ToolExecutionOutcome {
    pub output: String,
    pub output_data: Option<Value>,
    pub success: bool,
    pub error_reason: Option<String>,
    pub duration: Duration,
    pub receipt: Option<String>,
}

Fields

output: String

Text handed to the model and persisted to provider history. The success path carries raw bytes; the failure paths of execute_one_tool fold a tool’s detailed error body (which can reflect a token or signed URL) into this text and credential-scrub it before storing it here.

output_data: Option<Value>

Structured output when the tool declared one (ToolOutput::data). Feeds SOP step capture and data-flow surfaces; the LLM sees only output. Stored raw — consumers scrub at their own rendering boundary.

success: bool

error_reason: Option<String>

Raw, unscrubbed failure text for trusted in-process consumers (SOP step capture, data-flow surfaces). Credential scrubbing is a rendering concern applied at each human-facing surface (observer events, post-execution log line, CLI progress) and, unlike this field, on the model-visible output.

duration: Duration

receipt: Option<String>

Cryptographic HMAC receipt proving this tool actually executed. Present only when tool receipts are enabled in config.