Agent Identities

Agent Identities lets you classify specific GitHub logins as AI coding agents (Claude Code running in CI, GitHub Copilot's coding agent, Devin, or a similar tool) — separately from routine automation like Dependabot or Renovate. Only logins classified AI Agent here count toward the Agent Activity measures in Query Builder (PRs opened, PRs merged, merge rate, review coverage).

Why this needs classifying by hand

GitHub can tell Vertly that an author is a bot or app account, but not what kind of bot it is — Dependabot, Renovate, and an AI coding agent all look identical from that signal alone. And some AI-agent setups don't use a GitHub App at all: a team running Claude Code from CI with a personal access token shows up as a completely ordinary user account, with no bot signal whatsoever. So Vertly surfaces what it can detect and asks you to confirm which of those are actually agents, rather than guessing.

Classifying a detected login

Every bot- or app-typed account that's opened a PR in a connected repo shows up under Settings → Agent Identities → Detected, needs classification. Mark each one AI Agent or Automation — Automation is recorded explicitly (not just left blank) so you can tell "reviewed and confirmed not an agent" apart from "haven't looked at this yet."

Adding a login manually

A personal-access-token-based agent (a plain user account, not a bot/app) never shows up in the detected list — add it directly under Add manually with its GitHub login and a classification.

Until you classify something

Agent Activity measures show zero for a newly connected org until at least one login is classified AI Agent — an unclassified bot/app account is excluded, the same way every other pull-request measure already excludes bot-authored PRs by default (see Query Builder).