Instructions

To bring a long task in Codex to a result, break it into stages with verifiable ends. After each stage, save what is done, check results, and next action. Then continuation relies on project files.

Sequence from selection to local prototype:<a href=

Review meaningful changes separately while preserving sources and previous versions.

To test automation rerun, send one training event multiple times, including after an error. The system must distinguish new events from already processed ones. Result: a verification log showing which actions were performed and whether duplicates appeared.

To check AI agent access, first describe the allowed action, then try to exceed its limits in a training environment. If the agent must read notes, the check should show it does not change work records or send other files outside.

Start with source footage and a short plan, then check a sample with audio.

Agree on framing, action and important details first. A short sample helps test the chosen generation workflow.

From one verified material, you can prepare an article, a carousel, and a video script. First, highlight facts and limitations, then create each format from this common base. Result: consistent drafts referencing a single source version.

Connect MCP to Codex via configuration of a supported client. For CLI, below is an example server from official OpenAI documentation. After adding, check the actual tool call: the connection list shows the setup but does not prove the server responds.

To continue work in Codex in a new chat, save a short project record and ask the agent to verify it against current files. This avoids retelling the entire conversation and recalling which tasks are done.

To reduce Codex context, pass only files and information needed for the specific task. Verify that reduction does not worsen results. Do not promise a percentage saving without comparable runs.

To install a skill in Codex, select the exact source, check dependencies, and add one skill for a small task. After installation, perform a test request. Presence of file`SKILL.md` does not yet prove its scripts work in your environment.

To restore a project, check code, data, and uploaded files separately. Reverting the page to a previous version can still lose requests or images. Check result: restored training copy and report on what was recovered.

Prepare site for test deployment and domain connection. What settings, accesses, and checks are needed before publishing.

Check rights of two users: view, modify, and access others' data. What to log before app publication.

Assemble a training application form with Codex. Field validation, server response, text saving on error, and local example boundaries.

Choose a specific screen issue and test it in a working copy.

Choose app format by user action and needed device access. For reading knowledge bases, a website may suffice. Working with local files, phone, or another page's content requires separate capability checks.

Define form events and test them on a sample request. How to distinguish clicks, successful submission, and real result.

Choose a reader's question, write a clear answer, and check headings, links, and page accessibility. No promises of search rankings.

Check access and prepare test publication via API. How to separate draft, approval, and real submission.

Analyze posts by speech, frames, and structure. Create your content plan without copying others' text or fabricated statistics.