Module audit_log 

Persistent audit logging for tool invocations.

§18.4.4 of The Hitchhiker’s Guide to Agentic AI calls for an audit log of every tool call (arguments, outputs, timestamp). The existing crate::command_safety::audit::SafetyAuditLogger only records command-safety decisions in memory; this module adds a complementary, opt-in sink that records every MCP and built-in tool invocation in a durable, append-only JSONL stream.

Sinks

• JsonlFileSink — appends entries to a JSONL file with size-based rotation.
• InMemorySink — keeps entries in memory for tests and short-lived tooling.
• NullSink — discards everything; zero-cost when audit is disabled.
• MultiSink — fan-out wrapper used by ToolAuditLogger.

Threading

Sinks are Send + Sync. Writes are non-blocking from the caller’s perspective: each sink uses an internal lock or background task to serialize I/O.

See also crates/codegen/vtcode-core/src/tools/untrusted_data.rs for the prompt-injection defense that pairs with this log.

Structs

InMemorySink

In-memory sink — useful for tests and ephemeral tooling.

JsonlFileSink

Append-only JSONL file sink with size-based rotation.

MultiSink

Fan-out sink — forwards every entry to a list of inner sinks.

NullSink

Null sink — accepts every entry but discards them.

ToolAuditEntry

One audit row written per tool invocation.

ToolAuditLogger

Top-level audit logger — currently a thin wrapper that owns a single sink.

Enums

ToolAuditStatus

Status of a single tool invocation for audit purposes.

Traits

ToolAuditSink

Trait every audit sink implements.