feat(api): detect coding agents in User-Agent header

Jean-Gabriel Doyonrequested to merge
jgdoyon/detect-coding-agent-user-agent into main

Description

Appends a Coding-Agent/{name} token to glab's User-Agent header when an AI coding agent is detected, so GitLab can distinguish AI-agent traffic from human CLI usage.

  • Detects Claude Code, Codex CLI, Cursor, and OpenCode via their documented sentinel env vars
  • AI_AGENT env var takes priority and lets any agent self-identify without a glab release
  • AI_AGENT values are validated against [A-Za-z0-9._-]+ to keep the header well-formed
  • User-Agent becomes e.g. glab/v1.97.0 (darwin, arm64) Coding-Agent/claude-code; unchanged when no agent is detected
  • Existing ^glab/v server-side regex is unaffected (anchored to start)

Why env vars

Coding agents already set sentinel env vars in child processes. Reading those is zero-config and invisible to users. The AI_AGENT umbrella covers agents we haven't hardcoded yet.

Related Issues

Relates to gitlab-org/orbit/knowledge-graph#620 (closed)

How has this been tested?

Claude Code

image.png

image.png

Codex

image.png

OpenCode

Screenshot 2026-05-14 at 9.59.19 AM.png

Cursor

image.png

Edited by Jean-Gabriel Doyon

Merge request reports