OpenAI + Claude + Grok field guide

AI Commands Dictionary

Use this as a practical customer dictionary for OpenAI ChatGPT and Codex, Anthropic Claude and Claude Code, and xAI Grok. It separates official slash commands, @ mentions, coding-agent modes, model choices, and safe prompt patterns.

Browse Commands See @ and Mentions Official Sources
How commands work

Know what is official, then write the outcome.

Some commands are official product controls. Others are prompt conventions we recommend so customers can get better work from OpenAI, Claude, and Grok. This page labels the difference.

Command formula
platformOpenAI ChatGPT or Codex, Anthropic Claude or Claude Code, or xAI Grok.
/commandThe official slash command or recommended command-style prompt.
modelReasoning, fast, coding, multimodal, voice, local, or cost-sensitive lane.
toolsFiles, browser, connectors, GitHub, MCP, code execution, search, or apps.
boundaryPrivacy, citations, approval, budget, output format, and no-go areas.
OpenAI is strongest for broad app workflows.ChatGPT, Codex, Canvas, apps, connectors, and coding-agent reviews cover many business surfaces.
Claude is strongest for careful long-form and code workflows.Claude Code slash commands, skills, memory, model switching, and security review are useful for teams.
Grok is strongest when you want agent extensions and X-aware context.Grok Build exposes slash-command access to plugins, hooks, skills, MCP servers, and headless agent modes.
Visual platform guide

See the command surface before you copy it.

Each platform has a different mental model. Customers should know whether they are opening a workspace, calling a coding agent, routing context, or approving a tool before they run the command.

OpenAI-style workflow with a writing workspace, app routing, review checklist, and approval node.
OpenAI + Codex

Workspace first, then review and ship.

Use ChatGPT for drafts, Canvas-style iteration, app routing, and customer explanation. Use Codex when files, pull requests, tests, screenshots, and deployment proof matter.

/canvas/review@codex@ app
Claude-style workflow with long context, code review, security review, and model controls.
Claude + Claude Code

Long context, careful review, clean handoff.

Use Claude for deep reading, tone, documents, and careful explanation. Use Claude Code for compacting context, choosing model lanes, inspecting config, and security review.

/compact/model/config/security-review
Grok Build-style workflow with plugins, hooks, skills, MCP connections, and a controlled agent launch.
Grok + Grok Build

Extensions, agent modes, and fast experiments.

Use Grok Build when the job needs plugins, hooks, skills, MCP servers, headless prompts, or a controlled agent workflow that can be repeated.

/plugins/hooks/skills/mcps
Slash command dictionary

97 commands, routes, and safe patterns in one place.

Filter by platform or search by outcome. Each card says whether the entry is an official product control or a recommended prompt convention.

/canvasOpenAI official

Open a structured ChatGPT workspace.

Use Canvas when the task needs editable writing, code, or a longer draft you want to improve section by section.

ChatGPT /canvas Create a first draft of a service page for an Edmonton AI automation offer.
ChatGPT / shortcutsOpenAI official

Use ChatGPT app shortcuts for common actions.

In supported ChatGPT apps, slash-style shortcuts can speed up actions such as searching, reasoning, and image work.

ChatGPT: use the app shortcut menu, then ask for search-backed notes before writing the page.
/statusCodex official

Check the coding agent state.

Use status-style checks in Codex when you need to know what changed, what is running, and what still needs proof.

Codex /status Summarize changed files, tests run, deploy state, and remaining risks.
/reviewCodex official

Ask Codex to review code or a change.

Use review mode before deploying, merging, or publishing customer-facing work.

Codex /review Focus on customer-visible regressions, mobile overflow, broken links, and missing tests.
/modelCodex official

Switch or inspect the model lane.

Use when the work should move between fast editing, deeper reasoning, or coding-agent behavior.

Codex /model Use the strongest reasoning lane for the final architecture and deploy checks.
/fork + /sideCodex official

Explore alternatives without losing the main path.

Useful when you want one branch to stay stable while another explores a design or implementation option.

Codex /fork Try a second homepage menu structure, then compare against the current live navigation.
/prompts:nameCodex official

Reuse a saved prompt pattern.

Codex supports reusable prompt-style commands so repeat work can be invoked consistently.

Codex /prompts:mobile-qa Check the page like a phone customer and report blockers first.
/permissionsCodex official

Control what Codex can do without asking.

Use when a task needs to move between read-only, auto, and tighter approval rules.

Codex /permissions Tighten approvals before touching checkout, DNS, customer data, or deploy settings.
/ideCodex official

Pull editor context into the prompt.

Use when open files, current selection, or IDE state matter to the next request.

Codex /ide Use the current open file and selection to explain what this UI component does.
/appsCodex official

Browse app connectors.

Use when a customer job should pull in a connected app, service, or approved data surface.

Codex /apps Attach the right connector before asking for a calendar, email, or CRM summary.
/pluginsCodex official

Browse installed and discoverable plugins.

Use when the work needs document, browser, spreadsheet, creative, data, GitHub, or other plugin capability.

Codex /plugins Find the plugin that can inspect this file type before trying to solve it manually.
/hooksCodex official

Inspect lifecycle hooks.

Use when automation runs before or after tools, edits, builds, or verification steps.

Codex /hooks Review trusted hooks before letting the agent run a deploy or write files.
/compactCodex official

Compress a long Codex session.

Use after long implementation runs so the next turn keeps decisions, blockers, and proof without carrying every detail.

Codex /compact Preserve the goal, changed files, tests, deploy target, and unresolved risks.
/diffCodex official

Review the working diff.

Use before tests, commits, deploys, or handoff so the user sees exactly what changed.

Codex /diff Show customer-facing files first, then summarize risky changes.
/approveCodex official

Approve one denied retry.

Use when the auto reviewer blocked an action and the human wants to permit one retry.

Codex /approve Retry only the denied verification command, not the whole deploy.
/memoriesCodex official

Configure memory behavior.

Use when a session needs to inspect, enable, disable, or tune persistent memory behavior.

Codex /memories Check what durable project instructions may affect this answer.
/skillsCodex official

Browse task-specific skills.

Use when the job has a specialized workflow such as browser control, docs, data, design, GitHub, or PDF work.

Codex /skills Pick the correct skill before editing a spreadsheet, PDF, or product page.
/initCodex official

Create persistent project instructions.

Use to scaffold an AGENTS.md guide so future AI work understands the repo rules.

Codex /init Create instructions for build, test, deploy, mobile QA, and protected files.
/mcpCodex official

List connected MCP tools.

Use when a workflow depends on external tools, records, APIs, or authorized systems.

Codex /mcp Show available servers before asking the agent to inspect live data.
/mentionCodex official

Attach a file or folder.

Use when the next prompt needs a specific local file instead of vague project context.

Codex /mention Attach the screenshot and the page file before asking for a mobile fix.
/fastCodex official

Toggle a faster service lane.

Use when the model catalog exposes a fast tier and the task is routine enough for speed.

Codex /fast Use speed for simple copy edits, not for final security or payment checks.
/planCodex official

Ask for a plan before edits.

Use when the next change is risky, broad, or needs approval before implementation.

Codex /plan Redesign the mobile menu, but wait for approval before changing files.
/goalCodex official

Give Codex a tracked finish line.

Use for multi-step work where the agent should keep driving toward a clear objective across turns.

Codex /goal Add the command dictionary page, verify mobile, deploy, and report proof.
/ps + /stopCodex official

Monitor or stop background terminals.

Use when a dev server, build, test, or deploy is running in the background.

Codex /ps Check the dev server output, then /stop if the run is stale.
/resume + /newCodex official

Continue or reset a session.

Use resume for prior work and new when the customer wants a fresh context inside the same tool.

Codex /resume Continue the last shop QA session, or /new before a separate client task.
/usageCodex official

Check token and account usage.

Use when a task is large, long-running, or near rate-limit pressure.

Codex /usage Check limits before starting a large multi-page rewrite.
/theme + /keymapCodex official

Customize the terminal interface.

Use for syntax theme, key bindings, Vim mode, status line, and terminal title preferences.

Codex /theme Pick a readable terminal theme, then /keymap if the shortcuts slow you down.
/agentCodex official

Switch the active agent thread.

Use when a spawned subagent is working and the user needs to inspect or continue that branch.

Codex /agent Inspect the mobile QA subagent before merging its findings into the main plan.
@ appChatGPT official

Call a connected ChatGPT app.

Where supported, @ mentions let a user choose a connected app or assistant inside ChatGPT.

@YourConnectedApp Summarize this customer request and propose the safest next action.
@codex reviewOpenAI official

Request a GitHub pull request review.

On supported repos, @codex can be mentioned on a pull request to request a code review.

@codex review Please check the mobile menu, sitemap change, and redirect rules before merge.
@DocumentsCodex plugin route

Create or edit document artifacts.

Use when the customer needs Word-style proposals, handoffs, reports, redlines, or polished written deliverables.

@Documents Create a one-page proposal from these notes and render it for visual QA.
@PDFCodex plugin route

Read, create, and verify PDFs.

Use for extracting evidence, checking page layout, generating PDFs, or confirming a rendered document is presentable.

@PDF Inspect this proposal PDF and flag layout, crop, or missing-page issues.
@SpreadsheetsCodex plugin route

Work with sheets, CSVs, formulas, and charts.

Use when a customer gives you numbers, pricing, leads, revenue, products, or operational data.

@Spreadsheets Clean this CSV, add formulas, and summarize the top conversion opportunities.
@PresentationsCodex plugin route

Create and edit slide decks.

Use for pitch decks, client recaps, training decks, sales explainers, and board-ready summaries.

@Presentations Turn this AI readiness audit into a 10-slide customer deck.
@Template CreatorCodex plugin route

Turn a good artifact into a reusable template.

Use when one customer deliverable should become a repeatable format for future work.

@Template Creator Convert this proposal format into a reusable Codex template.
@BrowserCodex plugin route

Inspect and test a web page in the in-app browser.

Use for clicking, screenshots, layout checks, route checks, and customer-eye review of public pages.

@Browser Open the live shop page on mobile and check menu, CTAs, and overflow.
@ComputerCodex plugin route

Operate local Mac apps when the UI matters.

Use when the job cannot be solved by files or browser alone and needs real app clicks or visual inspection.

@Computer Use the visible app state to complete this admin step carefully.
@Creative ProductionCodex plugin route

Explore marketing visuals and campaign assets.

Use for ad directions, image prompts, offer boards, product scenes, and polished launch visuals.

@Creative Production Generate three click-worthy visual directions for this AI service offer.
@Data AnalyticsCodex plugin route

Turn data into reports, charts, and dashboards.

Use when the answer should be backed by structured rows, KPIs, metrics, or source-backed analysis.

@Data Analytics Build a small KPI dashboard from this revenue export and explain the trend.
@GitHubCodex plugin route

Work with repositories, issues, PRs, and checks.

Use when a customer asks about pull requests, CI failures, GitHub reviews, or repo history.

@GitHub Summarize the failing checks on this PR and list the safest fixes.
@GmailCodex plugin route

Search and summarize mail safely.

Use for inbox triage, thread summaries, draft preparation, and follow-up extraction before sending anything.

@Gmail Find the latest customer thread and draft a reply, but do not send it.
@Google CalendarCodex plugin route

Check calendars, events, and availability.

Use when a customer needs scheduling help, meeting prep, free windows, or event updates.

@Google Calendar Find two open windows for a strategy call next week.
/helpClaude official

List available Claude Code commands.

Use this first when you are inside Claude Code and need the active command surface for that environment.

Claude Code /help Show available slash commands, model controls, and project options.
/compactClaude official

Compress context during long work.

Useful when a coding or analysis session is getting long but needs to preserve the important state.

Claude Code /compact Keep the deploy target, changed files, test results, and unresolved issues.
/modelClaude official

Pick the Claude model for the job.

Use model controls when moving between speed, cost, long-context writing, or stronger coding/reasoning.

Claude Code /model Use the strongest suitable model for refactoring and final QA.
/configClaude official

Open Claude Code configuration.

Use configuration controls for model behavior, permissions, tools, and project-level setup.

Claude Code /config Review permissions and tool access before connecting private project data.
/security-reviewClaude official

Review for security risk.

Use before publishing code, connecting credentials, handling customer data, or changing authentication.

Claude Code /security-review Check auth, secrets, payment links, redirects, and customer data exposure.
/debug + /code-reviewClaude official

Invoke bundled or skill-backed coding help.

Claude Code skills can expose reusable slash commands for debugging, reviewing, and repeated team workflows.

Claude Code /code-review Focus on bugs, regressions, and missing verification before release.
/<skill-name>Claude official

Run a reusable Claude skill.

Teams can define project or personal skills that appear as slash commands, such as deploy checklists or code review routines.

Claude Code /mobile-audit Check the page for phone contrast, spacing, tap targets, and overflow.
/style-nameClaude app official

Use a Claude style shortcut.

Claude app styles can create slash-command shortcuts for tone and writing behavior.

Claude /customer-ready Rewrite this explanation in a practical buyer-facing style.
/add-dirClaude official

Add another working directory.

Use when Claude Code needs file access outside the current project root.

Claude Code /add-dir ../shared-assets Give the session read access to the shared asset folder.
/agentsClaude official

Manage agent configurations.

Use when a team wants specialized Claude agents for review, implementation, research, or QA roles.

Claude Code /agents Inspect available agents before assigning a background review task.
/background + /bgClaude official

Detach work into a background agent.

Use when a long task should keep running while the terminal is freed for other work.

Claude Code /background Continue the audit and report only blockers when finished.
/batchClaude official

Split large codebase changes into parallel units.

Use for migrations or broad edits that need planning, worktrees, tests, and multiple PRs.

Claude Code /batch migrate these service pages to the new navigation pattern.
/branch + /btwClaude official

Explore or ask without disrupting the main thread.

Use branch for alternate directions and btw for quick side questions that should not take over the session.

Claude Code /btw Is this claim source-backed, or should we soften it?
/cdClaude official

Move the session to another directory.

Use when the active project changes but the conversation should keep its useful context.

Claude Code /cd ../client-site Move this session into the client repo before editing files.
/clear + /newClaude official

Start fresh while keeping prior sessions available.

Use when a task is done and the next customer job should not inherit old context.

Claude Code /clear shop-fix-complete Start a clean conversation for the next page.
/contextClaude official

Visualize context usage.

Use when a long session may be bloated, stale, or close to the context limit.

Claude Code /context all Check what is taking space before compacting this session.
/diffClaude official

Open an interactive diff viewer.

Use to inspect current changes and per-turn diffs before merging or deploying.

Claude Code /diff Review the mobile menu and CSS changes before release.
/doctorClaude official

Diagnose installation and settings.

Use when Claude Code behavior feels broken, misconfigured, or missing expected capabilities.

Claude Code /doctor Verify the local install before blaming the project.
/effortClaude official

Set reasoning effort for the current model.

Use when a task should move between fast routine work and deeper architecture or review.

Claude Code /effort high Use deeper reasoning for the final checkout-risk review.
/forkClaude official

Spawn a background subagent.

Use when a side task can proceed independently and report back later.

Claude Code /fork Audit the public links while I continue the page copy.
/goalClaude official

Set an outcome Claude keeps working toward.

Use when the work needs persistence across turns until a named condition is satisfied.

Claude Code /goal Finish the page, test mobile, and deploy only after verification passes.
/mcpClaude official

Manage MCP server connections.

Use to reconnect, enable, disable, or inspect external tool servers.

Claude Code /mcp reconnect analytics Reconnect the approved data server before analysis.
/memoryClaude official

Edit project memory and auto-memory settings.

Use when persistent Claude instructions need to be inspected, cleaned, or updated.

Claude Code /memory Check whether old project instructions are steering this answer.
/permissionsClaude official

Manage allow, ask, and deny rules.

Use before a workflow touches tools, files, sensitive data, or irreversible operations.

Claude Code /permissions Require ask before network, send, delete, or payment actions.
/plugin + /reload-pluginsClaude official

Manage Claude Code plugins.

Use to list, install, enable, disable, or reload plugins without restarting the session.

Claude Code /plugin list Confirm the plugin exists before asking it to run.
/resumeClaude official

Resume a prior conversation.

Use when customer work spans days and the previous state should be recovered instead of rebuilt.

Claude Code /resume Open the last AGI Times session and continue from its live-check state.
/rewindClaude official

Return to an earlier point.

Use when the work went in the wrong direction and needs a checkpoint-style correction.

Claude Code /rewind Go back before the theme change and keep the content edits.
/run + /verifyClaude official

Launch and observe the app.

Use when tests are not enough and the agent must actually view the running product.

Claude Code /verify Launch the page, capture mobile, and confirm the button works.
/scheduleClaude official

Create or run cloud routines.

Use for recurring checks, maintenance routines, or scheduled follow-ups that are safe to automate.

Claude Code /schedule Check this public page every Monday and report broken links.
/simplifyClaude official

Clean up changed code.

Use after a working change exists and the question becomes whether it can be simpler or better placed.

Claude Code /simplify Review this CSS patch for unnecessary duplication.
/status + /tasksClaude official

Inspect session health and background work.

Use status for model/account/connectivity and tasks for work running in the background.

Claude Code /tasks Show what is still running before I ask for another deploy.
[Image #N]Claude official

Reference uploaded images precisely.

Claude supports image references in chat context; this is not an @ mention, but it acts like a target pointer.

Claude: Use [Image #1] to audit the phone menu and explain what should change.
/pluginsGrok official

Open Grok Build plugin controls.

Use when you need to install, inspect, or manage extensions for agent workflows.

Grok Build /plugins Open available plugins before building the customer support workflow.
/hooksGrok official

Manage lifecycle rules.

Hooks are useful when you need checks around tool use, edits, or workflow steps.

Grok Build /hooks Add a verification hook before any public deploy action.
/skillsGrok official

Open skill management.

Skills package reusable instructions and workflows so a team can repeat high-quality work.

Grok Build /skills Create or choose a skill for local SEO page audits.
/mcpsGrok official

Manage MCP connections.

Use when a workflow needs external tools, data sources, or authorized systems exposed through MCP.

Grok Build /mcps Connect the approved source before asking the agent to inspect records.
/<skill-name>Grok official

Invoke a user-created skill.

Grok skills can be called directly as slash commands for repeatable research, coding, or operating procedures.

Grok Build /shop-audit Review product cards, checkout friction, trust proof, and mobile speed.
Shift+Tab modesGrok official

Cycle agent modes.

Not a slash command, but important: Grok Build uses modes to control how much the agent acts versus asks.

Grok Build: use the appropriate mode before allowing a long-running multi-step task.
grok -pGrok official

Run a headless prompt.

Use command-line prompting when a workflow needs automation, scripting, or non-interactive output.

grok -p "Summarize this log and return the three highest-risk deployment issues."
--resume / --continueGrok official

Continue a previous session.

Session flags preserve context so long work can continue without losing state.

grok --continue -p "Finish the QA summary and include remaining blockers only."
--output-formatGrok official

Control output shape.

Useful for scripts that need JSON, streaming JSON, or predictable output rather than a normal chat answer.

grok -p "Return product QA findings" --output-format json
grok agent stdioGrok official

Run Grok as an agent server.

Use for advanced setups where another client or tool needs to talk to Grok through an agent protocol.

grok agent stdio --name opcelerate-review-agent
/home + /newGrok official

Return home or start fresh.

Use when a Grok Build session should reset or move back to the welcome surface.

Grok Build /new Start a clean session for the next customer workflow.
/sessions + /shareGrok official

Find, rename, and share sessions.

Use when a team needs to retrieve prior work or send a session link for review.

Grok Build /sessions Find yesterday's agent setup, then /share if it is safe to send.
/session-info + /contextGrok official

Inspect session state and context.

Use before continuing a long task or switching models so you know what the session already carries.

Grok Build /context Check context pressure before asking for a long implementation pass.
/modelGrok official

Switch the active Grok model.

Use when a task needs a different model lane for speed, reasoning, or multimodal work.

Grok Build /model Choose the model lane before starting a long tool workflow.
/always-approveGrok official

Toggle always-approve mode carefully.

Use only for low-risk environments, never for money, customer messages, secrets, or destructive changes.

Grok Build /always-approve Keep off when the workflow touches customer data.
/multiline + /compactGrok official

Control prompt input and session density.

Use multiline for bigger prompts and compact controls for managing conversation history or UI density.

Grok Build /multiline Paste the full acceptance checklist before starting the task.
/theme + /usageGrok official

Manage interface, feedback, and usage.

Use theme for readability, feedback for product notes, usage for credits, and logout on shared machines.

Grok Build /usage Check limits before asking for a long media-generation workflow.
/flush + /memory + /dreamGrok official

Manage memory and consolidation.

Use when a session needs persistent memory written, searched, edited, or consolidated offline.

Grok Build /memory Search for prior project rules before editing the site again.
/imagineGrok official

Generate an image from text.

Use for concept images, social cards, product visuals, and fast creative exploration.

Grok Build /imagine Create a bright hero image for an AI training page.
/imagine-videoGrok official

Generate video from text.

Use for motion concepts, ads, explainers, and social tests when video generation is available.

Grok Build /imagine-video Make a short product explainer concept from this script.
/quit + /rewindGrok official

Exit or return to an earlier point.

Use quit to leave the app and rewind when a session should back up to a safer state.

Grok Build /rewind Go back before the plugin install and keep the research notes.
/citeRecommended pattern

Force source-backed work.

Use for news, model claims, pricing, legal-adjacent topics, medical-adjacent topics, and customer education.

/cite Use official OpenAI, Anthropic, and xAI sources. Separate facts from recommendations.
/protectRecommended pattern

Add approval and privacy gates.

Use before an AI system touches customer data, sends messages, modifies checkout, publishes, or runs tools.

/protect Identify private data, approval gates, rollback steps, logging needs, and anything the AI must not do alone.
Model lanes

Pick the platform by job, not hype.

The right model choice starts with the work surface: customer drafting, coding agent, long-context review, search-aware research, or controlled automation.

OpenAI ChatGPT

Customer-facing drafting and app workflows.

Use for Canvas writing, app-connected tasks, broad ideation, customer education, and turning rough notes into useful page copy.

/canvas@ app/cite
OpenAI Codex

Repo work, code review, and deployment proof.

Use when the work involves files, tests, GitHub review, local verification, mobile QA, redirects, builds, and public release checks.

/review/model@codex
Anthropic Claude

Careful writing, long context, and review.

Use for long documents, reasoning-heavy edits, precise tone control, screenshots, and customer-friendly explanations that need nuance.

/style[Image #N]/goal
Claude Code

Structured coding sessions with safety controls.

Use for CLI-based code work, model/config controls, skills, debugging, code review, security review, and context compaction.

/compact/config/code-review
xAI Grok

Research, X-aware context, and Grok Build extensions.

Use for Grok Build sessions, plugin/skill/MCP extension workflows, headless prompts, and API-backed Grok model experiments.

/plugins/skillsgrok -p
Human approval

Money, security, legal, and promises.

Use a human checkpoint before checkout changes, customer messages, legal or medical claims, credentials, deletion, DNS, or irreversible deploys.

/protect/citeverify
@ and context routing

@ means route context, but each platform is different.

Do not assume every AI tool treats @ the same way. Use @ only when the product supports it, and use files, screenshots, skills, apps, or MCP when that is the real routing method.

OAI
OpenAI @ appsChatGPT context

Use when ChatGPT exposes a connected app, GPT, or tool as a selectable target inside the conversation.

CODEX
@codex reviewGitHub review

Use on supported pull requests to request a Codex review of changed code before merge.

DOC
DocumentsArtifact route

Use when the work is a proposal, guide, redline, report, Word-style artifact, or polished written handoff.

PDF
PDFRender route

Use for PDF extraction, creation, page-image QA, layout proof, and final printable files.

XLS
SpreadsheetsData file route

Use for CSVs, Excel files, formulas, tables, lead lists, pricing sheets, and chart-ready data.

PPT
PresentationsDeck route

Use when the customer needs a sales deck, strategy deck, training slides, or board-ready summary.

TPL
Template CreatorReuse route

Use when a good deliverable should become a reusable template for future customer work.

WEB
BrowserLive page route

Use for in-app browser checks, screenshots, clicks, route tests, and customer-eye page review.

MAC
ComputerApp UI route

Use when a native Mac app or admin screen must be operated visually and carefully.

CP
Creative ProductionVisual route

Use for ad concepts, image prompts, offer boards, product scenes, and creative exploration.

DATA
Data AnalyticsDecision route

Use when rows, KPIs, dashboards, reports, charts, or metric-backed recommendations matter.

CL
Claude filesContext pointer

Use uploaded images, file paths, project instructions, and named context instead of inventing an @ mention that Claude does not support.

SKILL
Claude skillsSlash route

Use /<skill-name> when the reusable workflow lives as a Claude Code skill.

GX
Grok extensionsBuild route

Use /plugins, /hooks, /skills, and /mcps to route work through Grok Build extensions.

MCP
MCP serversTool context

Use MCP when the AI needs authorized tools, docs, records, or systems instead of pasted context.

GIT
GitHubRepo route

Use for repositories, pull requests, issues, code review, CI checks, and deploy-related proof.

MAIL
GmailInbox route

Use for thread summaries, inbox triage, reply drafts, follow-ups, and safe no-send reviews.

CAL
CalendarSchedule route

Use for availability, meeting prep, calendar conflicts, free windows, and scheduling suggestions.

URL
Links and sourcesEvidence route

Paste official URLs for model claims, pricing, legal-adjacent guidance, medical topics, and news.

HITL
Human checkpointApproval route

Name the human owner when the next step affects money, customer trust, security, policy, or publishing.

Customer workflow

One strong request beats ten vague prompts.

Use this sequence when you are choosing between ChatGPT, Codex, Claude, Claude Code, Grok, or Grok Build for a real customer job.

1. Choose platformChatGPT for broad work, Codex for repos, Claude for careful context, Grok for Build/X-aware workflows.
2. Pick commandUse an official slash command when available; otherwise use a clearly labeled prompt pattern.
3. Attach contextAdd files, screenshots, URLs, docs, source links, repo state, or the exact customer note.
4. Set boundaryName privacy rules, approval gates, model lane, output format, budget, and forbidden actions.
5. Require proofAsk for citations, tests, screenshots, live URL checks, limitations, and the next safe move.
Source desk

Official references behind the dictionary.

These links are the starting point for checking whether a command is current. Product surfaces change, so high-stakes work should re-check the official page before rollout.

Approval rules

Commands speed up work. They do not remove judgment.

The safest AI systems are explicit about what can move automatically and what needs a human checkpoint.

Always ask before these actions.

  • Charging money, changing checkout, or modifying payment settings.
  • Sending emails, SMS, DMs, outreach, or customer-facing promises.
  • Deleting data, overwriting active work, changing DNS, or touching credentials.
  • Publishing legal, medical, financial, security, or high-stakes claims.

Ask for this proof when it matters.

  • Live URL and HTTP status after deployment.
  • Mobile and desktop screenshots for important pages.
  • Source links for news, technical claims, and citations.
  • Known remaining risks, test gaps, and next recommended move.
Ready starter

Copy this when you do not know where to begin.

Use this as a safe first prompt in ChatGPT, Claude, Grok, or with a coding agent. It gives structure without forcing you to become a prompt engineer.

Get Free AI Scan Open Service Shop Book Strategy Call
/goal Help me choose the right AI tool for this job. Compare ChatGPT/Codex, Claude/Claude Code, and Grok/Grok Build. Ask for missing context, cite official sources for product claims, identify privacy and approval risks, then give the first practical step.