Repository tasks and the coding loop

Home Tools Platforms Codex Repository tasks and the coding loop

Inspect a repository, make a scoped change, run appropriate checks, and review the difference.

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.

The worked example is illustrative and has not been tested in a signed-in product account.

What it does and how it connects

Inspect a repository, make a scoped change, run appropriate checks, and review the difference.

Codex can inspect files, edit code, and run commands in the environment it is allowed to use. The useful unit is a coding loop: understand the problem, change a small scope, check behavior, and review the difference. A CLI task, an IDE workspace, and a cloud task can see different files.

Walkthrough

Use the product named in the access panel. The steps below combine documented controls with the editorial example.

  1. Install / authenticate through the official CLI or supported client instructions and open an authorized sample repository.
  2. Record the branch and working-tree state; preserve existing changes.
  3. Ask Codex to explain the relevant behavior before requesting a narrowly scoped edit.
  4. Inspect changed files, command output, and the final diff before committing.

Worked example

Use a sample repository containing a broken search filter. Supply a reproducible input and expected result. A useful handoff includes the fix and a meaningful check of the reported failure, without unrelated formatting changes.

In this sample repository, reproduce the search-filter issue: [input] should return [expected result]. Inspect the relevant code first. Make the smallest justified change, preserve unrelated edits, and run an appropriate check. Report files changed, actual commands and outcomes, and anything unverified. Do not push or deploy.

Inspect and verify the result

Read the diff and reproduce the original case. Confirm tests actually ran and inspect failures. A completion message is not evidence of a passed check; request exact command outcomes and remaining uncertainty.

Limits and troubleshooting

Missing dependencies, unavailable services, or a different working directory can prevent verification. Ask for the blocked dependency rather than accepting invented results. Git checkpoints protect deliberate human changes beyond any chat undo feature.

Data, permissions, and actions

Repository files and command results become task context. Tools execute with the runtime’s effective permissions. Use a sample or approved repo, scope network access, and keep secrets and production credentials out of the learning exercise.

AGENTS.md, skills, and memory

Separate checked-in project instructions from reusable procedures and locally recalled context.

Explore AGENTS.md, skills, and memory

A scoped repository bug fix

Reproduce a sample defect and hand off a checked diff.

Explore A scoped repository bug fix

A prepared cloud task and review handoff

Run a bounded repository task in a reproducible cloud environment.

Explore A prepared cloud task and review handoff

Evaluations: does the system do the job?

Create representative tests for quality, tool behavior, refusal, escalation, and the effects of a change.

Explore Evaluations: does the system do the job?

Sources and verification

Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.

Codex CLI · OpenAI

Using ChatGPT: Chat and Work · OpenAI

Share this page