Reproduce a sample defect and hand off a checked diff.
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.
An illustrative recipe. This website does not run the automation.
The goal and the right kind of system
Reproduce a sample defect and hand off a checked diff.
An interactive coding agent fits investigation and iteration. A person selects scope and reviews the final change.
Inputs and connected services
Inputs: sample repo, clean checkpoint, reproduction, expected behavior, and actual test command.
Services: local Codex and repository tools; external services are excluded from the pilot.
The steps
An illustrative sequence to adapt and test before using real data or enabling ongoing work.
- Read AGENTS.md and inspect the relevant files.
- Reproduce or document the blocked reproduction.
- Make a scoped change and run meaningful checks.
- Review the diff and exact command outcomes.
Outputs and human checkpoints
Outputs: changed files, test evidence, and unverified concerns.
Human checkpoints: approve scope and inspect the diff before commit.
Exceptions, permissions, and stopping conditions
Permission boundary: sample workspace only, no push / merge / deployment. Missing dependencies remain blockers.
Stop on failing checks, uncertain reproduction, unexpected file changes, or a request for production access.
Adaptable prompt
Fix [reproducible sample defect] in this repo. Follow project guidance, preserve unrelated edits, run actual checks, and report outcomes. Stop on missing required access. Do not push, merge, or deploy.
Related reading
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.
Designing an AI workflow
Make a useful task repeatable with clear inputs, steps, outputs, exceptions, and human checkpoints.
Permissions and meaningful approvals
Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.
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