📖 Glossary
Agentic loop
Codex’s core cycle: gather context (read/grep) → act (edit/run in a sandbox) → verify (tests) → repeat until done.
TUI
Text user interface — the full-screen interactive app you get after typing codex in your terminal.
AGENTS.md
Markdown project-memory file Codex reads automatically each session: commands, conventions, guardrails.
Sandbox
The guardrail around Codex’s actions: where it may write and whether it can reach the network (sandbox_mode).
Approval policy
When Codex pauses to ask before acting: untrusted, on-request, or never.
Reasoning effort
How long the model deliberates before acting: low, medium, or high.
MCP
Model Context Protocol — an open standard connecting Codex to external tools and data (GitHub, docs, databases).
Cloud task
Work delegated to Codex cloud: runs isolated on OpenAI’s infrastructure, in the background, and opens a pull request.
Headless
Running Codex non-interactively (e.g. codex exec) for scripts and CI — no session prompts.
UCC-1
A financing statement filed to perfect a lender’s security interest in a debtor’s collateral.
§9-503
The UCC rule requiring a registered organization’s exact legal name on a filing; a wrong name can be seriously misleading.