Extremely high token burn: 13.6K tokens wasted on a single "Hi"

What version of the Codex App are you using (From “About Codex” dialog)?

0.116.0

What subscription do you have?

Team

What platform is your computer?

Darwin 25.3.0 arm64 arm

What issue are you seeing?

A completely fresh Codex CLI install (no AGENTS.md, no custom config, no MCP servers) consumes ~13.6K tokens out of 258K context window just to process a single "Hi" message and respond with one sentence.
That's 5% of the context gone before any real work begin

What steps can reproduce the bug?

Steps to reproduce:

1. Fully uninstalled codex from both brew and npm (brew uninstall codex + npm uninstall -g @openai/codex), removed ~/.codex config directory, then did a clean reinstall via brew
2. Clean install codex v0.116.0 via brew
3. Run codex with default settings (gpt-5.2, reasoning medium)
4. Type "Hi", press Enter
5. Run /status

Result: 13.6K used / 258K context window
Expected: <1K for a trivial greeting

What is the expected behavior?

Baseline token usage for a simple message should be minimal. System prompt overhead should be much smaller or clearly documented as a fixed per-session cost.

Additional information

• No AGENTS.md in working directory
• No custom MCP servers
• Default config, no customization
• Account: Team plan
• Session ID: 019d1b53-1408-7841-aa46-48e1d9d2fd25

<img width="1204" height="662" alt="Image" src="https://github.com/user-attachments/assets/f3478f70-1b65-47d2-af9d-6fcf00423662" />