Struct ToolAuditEntry 

pub struct ToolAuditEntry {

    pub timestamp_unix_ms: u64,
    pub session_id: String,
    pub turn_id: String,
    pub tool_call_id: String,
    pub tool_name: String,
    pub arguments_hash: String,
    pub arguments_redacted: Option<Value>,
    pub result_hash: String,
    pub result_summary: Option<String>,
    pub duration_ms: u64,
    pub status: ToolAuditStatus,
    pub sandbox_policy: Option<String>,
    pub transport: Option<String>,
    pub server_address: Option<String>,
    pub server_port: Option<u16>,
    pub model_id: Option<String>,
    pub prompt_injection_flagged: bool,
    pub reason: Option<String>,
}

One audit row written per tool invocation.

The shape mirrors what tracing already records for MCP tool calls (mcp.tools.call span with provider / transport / server metadata) but adds enough context for offline forensics.

Fields

timestamp_unix_ms: u64

Unix epoch milliseconds when the entry was created.

session_id: String

Stable identifier for the session (e.g. UUID).

turn_id: String

Stable identifier for the turn within the session.

tool_call_id: String

Provider-side identifier for the tool call (matches tool_call_id).

tool_name: String

Canonical tool name (e.g. mcp::fetch::fetch, read_file).

arguments_hash: String

SHA-256 of the original (unredacted) tool arguments, hex-encoded.

arguments_redacted: Option<Value>

Optional redacted snapshot of the arguments (secrets removed).

result_hash: String

SHA-256 of the tool result, hex-encoded.

result_summary: Option<String>

Optional first N characters of the result for offline triage.

duration_ms: u64

Wall-clock duration of the tool call in milliseconds.

status: ToolAuditStatus

Final status (success / failure / timeout / cancelled / blocked).

sandbox_policy: Option<String>

Optional sandbox policy applied to this tool call (path to a JSON snapshot, or a short identifier).

transport: Option<String>

MCP transport when applicable (stdio, streamable_http, …).

server_address: Option<String>

Remote server address when applicable.

server_port: Option<u16>

Remote server port when applicable.

model_id: Option<String>

Identifier of the model that requested the call.

prompt_injection_flagged: bool

True when the static is_suspicious_instruction probe flagged the result content.

reason: Option<String>

Optional free-form reason (e.g. loop_detector, circuit_breaker_open).

Trait Implementations

impl Clone for ToolAuditEntry

fn clone(&self) -> ToolAuditEntry

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ToolAuditEntry

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ToolAuditEntry

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ToolAuditEntry

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.