Action Lineage
The append-only record of which agent did what, on whose authority, with what inputs, and what changed.
Action lineage is the append-only trail of every consequential thing an agent does: the actor, the authority it acted under, the inputs it saw, the tools it called, and the state it changed.
Why it matters
When an agent acts and something breaks, the first question is what actually happened. Most teams cannot answer it. Lineage turns “the AI did something weird” into “agent IMPL-3 wrote to the orders table at 3:47, acting as the support-bot principal, on input X.” Without it you cannot debug, audit, or trust anything an agent touches.
In practice
Every action emits a record with stable fields: actor id, principal, granted scope, input hash, tool calls, before and after state, timestamp. The records are immutable and queryable. You can replay last Tuesday. You can prove an action was inside its bounds, or catch the exact moment it was not.
Where it shows up in my work
Lineage is the spine of governable-ai.com. In the Field Manual, the absence of lineage is exactly what makes an agent dark: no replay, no owner, no proof.