AI Agent Deployments in Practice

How are Codex, Claude Code, OpenClaw and other agents deployed in practice?

Share anonymous configuration details and compare deployments worldwide.

Participants
Agent submissions
Manual script
Manual questionnaire

Participation Options

Two simple options:

Run via Agent Skill (Recommended) - less than 2 minutes

Install the skill from internetwarte/agent-setup-survey.

See installation instructions
Skill folder: ~/.codex/skills/
git clone https://github.com/internetwarte/agent-setup-survey
cp -r agent-setup-survey ~/.codex/skills/

Execute: Claude /agent-setup-survey or Codex $agent-setup-survey. The agent asks for consent and submits anonymized results automatically.

No network access? Submit manual script output instead

Manual questionnaire (less than 30 seconds).

If skills > 0, choose categories

Sandboxed vs Bare Metal

Agent (Auto + Manual Script)
Manual Questionnaire
Sandboxed Bare Metal

Environment Summary (Agent Only: Auto + Manual Script)

VM Technologies
Container Technologies
Bare-Metal OS

Why Participate?

Help build an open dataset on how local and user-controlled AI agents are deployed (local machines, self-hosted servers, VPS).

This data contributes to an ongoing research study on real-world AI agent execution environments.

More details in the footer.

What Are Agent Skills?

Agent skills are small extensions that teach your AI agent a new task.

A skill usually contains:

  • A SKILL.md file with instructions the agent can follow
  • Optional helper code (e.g., a small Python script)

In this case, the skill teaches your agent how to collect a few high-level configuration details and submit them anonymously.

You install the skill in your agent’s skill folder (e.g., ~/.codex/skills or ~/.claude/skills) and then execute it like any other command.

For the full technical specification, see: agentskills.io/specification

Other agents: skill folder paths
Agent Global Skill Folder
Amp / Kimi Code CLI / Replit / Universal~/.config/agents/skills/
Antigravity~/.gemini/antigravity/skills/
Augment~/.augment/skills/
Claude Code~/.claude/skills/
OpenClaw~/.openclaw/skills/
Cline~/.cline/skills/
CodeBuddy~/.codebuddy/skills/
Codex~/.codex/skills/
Command Code~/.commandcode/skills/
Continue~/.continue/skills/
Cortex Code~/.snowflake/cortex/skills/
Crush~/.config/crush/skills/
Cursor~/.cursor/skills/
Droid~/.factory/skills/
Gemini CLI~/.gemini/skills/
GitHub Copilot~/.copilot/skills/
Goose~/.config/goose/skills/
Junie~/.junie/skills/
iFlow CLI~/.iflow/skills/
Kilo Code~/.kilocode/skills/
Kiro CLI~/.kiro/skills/
Kode~/.kode/skills/
MCPJam~/.mcpjam/skills/
Mistral Vibe~/.vibe/skills/
Mux~/.mux/skills/
OpenCode~/.config/opencode/skills/
OpenHands~/.openhands/skills/
Pi~/.pi/agent/skills/
Qoder~/.qoder/skills/
Qwen Code~/.qwen/skills/
Roo Code~/.roo/skills/
Trae~/.trae/skills/
Trae CN~/.trae-cn/skills/
Windsurf~/.codeium/windsurf/skills/
Zencoder~/.zencoder/skills/
Neovate~/.neovate/skills/
Pochi~/.pochi/skills/
AdaL~/.adal/skills/
How do you interact with your agent?
Web UI
CLI
Messenger
Other

Participant Countries (0)

Top 10 ASNs (Agent Submissions)

Top 10 Agents (Agent Submissions)

Manual Questionnaire: Environment

Manual Questionnaire: Access

Manual Questionnaire: Internet

Manual Questionnaire: Skills Added

Research imprint
This study is conducted by researchers at IT:U Austria (Interdisciplinary Transformation University Austria) in collaboration with the University of Vienna.

The goal is to understand how AI coding agents are deployed across the broader developer community. Only user-agent and coarse network prefixes (/24 for IPv4, /64 for IPv6) are stored for rate-limiting and aggregation, alongside submitted survey answers. Aggregated results are shared publicly.

The endpoint is hosted by Hetzner in Germany. If you have questions about the study, please open an issue on GitHub.

Technologies

Area Technology
Agent skills agentskills.io
Visualizations Apache ECharts
Backend/API FastAPI + Nginx
Datasets IPinfo Lite, Pyasn, RouteViews, CAIDA AS-Org