Intermediate · Safety

Agent Audit Trail

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. The chronological, tamper-resistant record of an agent's decisions, tool calls, and context sources.

Technical Definition

The chronological, tamper-resistant record of an agent's decisions, tool calls, and context sources.

How it works

An agent audit trail records the full decision path: the goal, each planning step, every tool invocation with arguments and results, which context segments were injected and from where, and any human approvals. Unlike application logs it is retained for compliance periods, protected from modification, and structured for replay so an investigator can reconstruct why the agent acted. Provenance of context is the part teams most often omit and most often need after an incident.

Related Concepts

• Agent Identity — The verifiable answer to 'which agent is this?' — a stable, attestable identifier distinct from the human or service behind it.
• Context Lifecycle — The stages a piece of context passes through: acquisition, validation, injection, use, compaction, persistence, and expiry.
• Agent Action Evidence — Verifiable records proving what an agent did, under whose authority, with what inputs and result.
• Agent Transparency — Making an agent's identity, authority, reasoning, and actions legible to the humans and systems it interacts with.