Inspect Agent Run With Prefactor CLI

Perform root-cause analysis on a specific agent run using Prefactor span data.

When to use

The user has already decided something in a run is wrong or worth investigating. They paste identifiers and a symptom; your job is to analyze the spans and deliver a findings report — what happened, what caused it, and where in the codebase it connects.

Do not prescribe fixes or recommended next actions. The human team owns remediation; your output equips their post-mortem with evidence and source locations.

Typical handoff:

• Agent ID (required)
• Agent Instance ID (when listing specific instances or resolving context)
• What went wrong, what surprised them, or what they want explained — may be ambiguous.

Run in the agent's codebase

Run this skill from the repository that deployed the agent under investigation — not from the Prefactor SDK repo or an unrelated project. Assume you are running in the correct repo and thus have access to the full codebase of the traces.