Practical Prompt Guide for ChatGPT, Work, and Codex
This guide explains how to craft effective prompts for ChatGPT, ChatGPT Work, and Codex by defining clear goals, providing background, specifying output formats, setting boundary rules, leveraging linked data sources and plugins, personalizing settings, and iteratively refining results with concrete examples for daily conversation, office tasks, and code scenarios.
Prompt Structure
Define the task by specifying Goal , Background , Output , and Boundary . Include only the parts that are needed for the specific request.
Result‑first description
State the desired final output and any audience or format constraints.
Turn meeting minutes into a brief project‑team progress update, placing decisions and next steps at the top.
Background information
Attach only essential references (documents, spreadsheets, PDFs) and indicate what should be extracted. For up‑to‑date answers, allow web search and require source citations.
Provide files when a summary, comparison, or transformation is needed.
When visual data is required, attach screenshots or charts and highlight the relevant region.
Linked data sources
If ChatGPT can access external sources, specify the query scope and goal without enumerating each search step.
Use the latest project plan in Google Drive and relevant decisions from the Slack channel to write a status report.
Plugins
Plugins expose reusable commands for Google Drive, Gmail, Slack, GitHub, and other tools. Invoke a plugin by typing @ in the input box.
Risk‑avoidance boundaries
Define constraints that prevent unwanted modifications.
Do not change confirmed dates or budget data.
Use only the provided information; flag missing data instead of guessing.
All suggestions must stay within the specified budget.
Draft messages must not be sent automatically.
Iterative refinement
Review the first response and request specific changes.
Make the opening more concise, keep the arguments, and move the recommendations after the background.
Real‑time guidance and queue commands
During a running Codex task you can:
Guide : add a message to adjust direction or supply missing information.
Queue : save a message for the next round when the current work must finish first.
Daily‑conversation prompt examples
Explain compound interest to a beginner using a concrete case and defining all financial terms.
Office‑scenario prompts
Prepare a one‑page project status report for Monday’s management meeting using the latest plan in Google Drive and decisions from the Slack channel. Highlight decisions and next steps at the top, keep confirmed dates and budget unchanged, and flag any conflicts.
Codex code‑scenario prompts
Typical workflow:
Open the relevant file(s) in the IDE.
Select the code snippet and add it to a Codex thread.
Provide a concise prompt describing the required function, constraints, and verification method.
Bug‑fix example
Bug: Clicking “Save” sometimes shows “Saved” but changes are not applied. Steps: 1. npm run dev → 2. Open /settings → 3. Toggle “Enable reminders” → 4. Click Save → 5. Refresh – the toggle reverts. Constraint: Do not modify the API; the fix should be minimal and include regression tests.
UI‑prototype example
# Constraints
- Use React, Vite, Tailwind, TypeScript
- Match spacing, layout, and typography as closely as possibleGenerate a new route, required components, and a README with run instructions based on the supplied design image.
Cloud‑hosted refactoring
$plan We need to refactor the authentication subsystem: - Split responsibilities (token parsing, session loading, permission management) - Reduce circular dependencies - Improve testability Constraints: No user‑visible feature changes, keep public APIs stable, provide a step‑by‑step migration plan.
Local code review and GitHub integration
/review 重点审查边界情况和安全问题After fixing issues, rerun the review command to confirm resolution.
Documentation update example
Update the “Advanced Features” documentation with authentication troubleshooting steps and verify that all links are valid.
Command‑line Codex workflow (file context)
Read @foo.ts and @schema.ts , explain the data structures and request/response flow, highlighting required and optional fields and compatibility rules.
Testing example
Add unit tests for the invert_list function in @transform.ts , covering normal cases and edge cases.
Voice input
In the desktop client, hold Ctrl+M to start speech‑to‑text transcription; edit the transcribed text before sending.
Key commands
/plan– let Codex research and propose a solution before editing. /goal – set a persistent objective for the session. $plan – invoke a structured planning request. /review – run a code review in the current workspace. @ – reference a file path or trigger a plugin. /mention – auto‑complete a file path in a command.
Workflow summary
1. Define the goal and constraints.
2. Provide minimal, relevant background data.
3. Choose plugins or linked data sources as needed.
4. Issue the prompt and receive a draft.
5. Iterate with guide or queue messages until the output meets the required format, length, and verification criteria.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
AI Architecture Hub
Focused on sharing high-quality AI content and practical implementation, helping people learn with fewer missteps and become stronger through AI.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
