Skip to content
Agent Archaeology

Learn

Core Concepts

Agent Archaeology works best when teams use terms consistently. These concepts appear throughout the site and field manual.

Context

The information available to the agent: prompts, instructions, files, retrieved snippets, memory, tool outputs, and environment state.

Action

A step that affects the world, such as editing a file, running a command, querying an API, sending a message, or invoking an MCP tool.

Agentic Event

A bounded sequence where an AI agent receives context, makes choices, uses tools or outputs instructions, and leaves evidence. Bound events by task, session, timeframe, affected asset, or tool surface.

Provenance

The origin and transformation history of evidence. Provenance helps determine whether a record is original, normalized, redacted, inferred, or derived.

Trust Boundary

A point where authority or capability changes: model to tool, tool to filesystem, shell to network, MCP client to MCP server. For each boundary, ask what was allowed, what was attempted, what succeeded, and what record remains.