A coding agent for repository inspection, scoped changes, command execution, and reviewable development work.
Coding & development · Documented · Documentation checked 2026-10-04
Codex CLI, IDE extension, and desktop coding work; cloud environments need separate setup.
ChatGPT entitlement or supported API authentication, host capabilities, and workspace policy affect access.
What the platform is for
Codex helps turn a reproducible development problem into a change you can inspect. Its important components are the repository, instructions, model, tools, permissions, environment, and checks. A beginner should use a sample repo and learn to read a diff before giving an agent a broad implementation task.
Know the product boundary
CLI, IDE, desktop, and cloud development have different context and capability. ChatGPT projects do not mount a local repo automatically. Cloud environments require preparation and source access. Local memory differs from ChatGPT memory, and consumer entitlement does not establish the same API billing.
How the components connect
Repository instructions describe how to work. Skills supply repeatable procedures. Shell and MCP tools perform operations under permissions. Cloud setup makes dependencies and checks reproducible. Reviews and noninteractive runs create evidence for a human-controlled handoff. A successful agent turn does not itself authorize merge or deployment.
A beginner starting point
Begin by asking for a code explanation, then a scoped fix with a known reproduction. Inspect the change and actual test outcomes. Add deliberate instructions before extensions. Learn permission boundaries and cloud setup before recurring or noninteractive work.
Limits and data choices
Missing services, unread files, and unrun checks should remain explicit blockers. Prompt instructions are not access enforcement. Beta permission profiles and surface-specific plugin support require current documentation. This edition teaches the core coding loop rather than every configuration option, subagent API, or production deployment pattern.
The technical path
Continue from the official Codex / OpenAI Learn docs into configuration, MCP, AGENTS.md, evaluation, and developer APIs. Deployment depends on the repository’s own host and secrets. Use least-privilege test environments and a reviewable branch / diff as the handoff.
Feature map
Repository tasks and the coding loop
Inspect a repository, make a scoped change, run appropriate checks, and review the difference.
AGENTS.md, skills, and memory
Separate checked-in project instructions from reusable procedures and locally recalled context.
Shell tools, MCP, and permissions
Distinguish a requested command from the filesystem, network, and external access actually allowed.
Cloud environments and repository access
Prepare a repeatable remote environment with dependencies, scoped secrets, and a known test command.
Reviews and repeatable noninteractive runs
Use a code review or noninteractive task as evidence for a human-controlled change.
Complete workflows
A scoped repository bug fix
Reproduce a sample defect and hand off a checked diff.
A prepared cloud task and review handoff
Run a bounded repository task in a reproducible cloud environment.
Coverage and availability
Models, modes & conversation
Covered: Client model choices and task loop; current menus vary.
Search, research & sources
Covered: Search is a distinct tool; CLI / IDE do not inherit built-in Browser.
Files, analysis & code
Covered: Repository files, code edits, commands, and checks.
Images, video, audio & voice
Covered: Multimodal developer capability varies; not a consumer media studio guide.
Projects, instructions & memory
Covered: AGENTS.md, skills, local memory, and workspace scope.
Editing surfaces & deliverables
Covered: Code / generated files and reviewable diffs; visual preview is surface-specific.
Agents, skills, plugins & MCP
Covered: Shell, MCP, plugins, and permissions; all extension APIs are not covered.
Orchestration & scheduling
Covered: Noninteractive runs and scheduling boundary; host controls needed.
APIs, SDKs, deployment & evaluation
Separate product: Cloud environments and verification; production APIs / deployment require separate engineering.
Sharing, privacy & governance
Covered: Workspace and connected-system controls; source and output handling reviewed.
Plans, devices & accessibility
Not established: Entitlement and surface limits included; full quota and accessibility matrix not established.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Codex CLI · OpenAI
Using ChatGPT: Chat and Work · OpenAI
Projects and chats · OpenAI
Build skills · OpenAI
Memories across product surfaces · OpenAI
Permissions and approval policies · OpenAI
Plugins and supported surfaces · OpenAI
Cloud environments · OpenAI
Get started with Work · OpenAI
Scheduled tasks · OpenAI