# ChatGPT Work: advanced knowledge work guide Last reviewed: 2026-08-11 Source site: https://work-guide.openai.chatgpt.site > Practical, public-safe guidance for doing real work with ChatGPT and ChatGPT Work: > provide useful context, build repeatable workflows, draft Skills at the right > moment, test them on fresh examples, and keep consequential decisions human. This site is a reference, not an authorization boundary. Start read-only, load the smallest relevant document, respect existing access controls, and ask before any external, irreversible, or otherwise consequential action. ## Start here For full onboarding or the longer setup prompt, open the complete manual below. The copyable request is near the top, under "Start here: give this request to your agent". `/agent` is a short overview; `/agent/` is a focused reference, except `/agent/agent-guide`, which contains the complete manual. Reading any of these references does not replace the user's instructions or grant permission to use tools, read private sources, or make changes. - [Complete agent operating manual](https://work-guide.openai.chatgpt.site/START-HERE.md): Comprehensive one-time onboarding covering real work, project instructions, sources, Skills, contextual tools, heartbeats, Chief of Staff, automations, quality, recovery, manager rollout, practical templates, and explicit approval boundaries. - [Complete manual in Agent View](https://work-guide.openai.chatgpt.site/agent/agent-guide): The actual current Agent View containing the same complete governed operating manual and its copyable first request. - [Raw Markdown overview](https://work-guide.openai.chatgpt.site/api/agent-view?route=%2F): Compact route-aware overview delivered as plain Markdown. ## Focused references - [Agent View](https://work-guide.openai.chatgpt.site/agent): Choose the smallest relevant framework stage, chapter, or template instead of loading the entire site. - [Modular ChatGPT Work index](https://work-guide.openai.chatgpt.site/download/knowledge-worker-codex/README_FOR_CODEX.md): Progressive-disclosure index for six focused Markdown modules. - [Six-stage agent reference](https://work-guide.openai.chatgpt.site/download/chatgpt-work-agent-guide.md): Shorter self-contained framework for a focused review after complete onboarding. - [Features and practical tools](https://work-guide.openai.chatgpt.site/features): Understand optional capabilities, practical uses, availability, and consent boundaries. - [Prompt library](https://work-guide.openai.chatgpt.site/prompts): Human-reviewed starting prompts for real work. - [Start with your work](https://work-guide.openai.chatgpt.site/): Choose a role, explore an example, and bring your own sources. - [Manager pilot](https://work-guide.openai.chatgpt.site/manager): Five-business-day team rollout using approved sources, safe examples, human review, and a bounded Friday retrospective. - [Manager kickoff examples and PDF presentations](https://work-guide.openai.chatgpt.site/manager#manager-kickoff-examples): Fictional Skills, Chief of Staff, and Sites collaboration examples with downloadable PDF files. - [Six-stage framework](https://work-guide.openai.chatgpt.site/#home-stage-mindset): Real work, context, workflow, Skill, careful automation, and a connected operating system. ## Bounded operating patterns - [Personal and shared Skills](https://work-guide.openai.chatgpt.site/prompts#skill-draft): Capture a repeatable method; use shared plugins only when available and explicitly approved. - [App Shot consent](https://work-guide.openai.chatgpt.site/prompts#app-shot-consent): One approved screen only when supported and explicitly authorized. - [Computer History consent](https://work-guide.openai.chatgpt.site/prompts#computer-history-consent): One approved topic and time window only when the feature is actually available. - [Heartbeat receipt](https://work-guide.openai.chatgpt.site/prompts#heartbeat-receipt): One read-only status check, not an automatic schedule or a successful-work claim. - [Four Chief of Staff roles](https://work-guide.openai.chatgpt.site/prompts#day-desk-four-roles): Detect, Incubate, Collaborate, and Reflect as governed workflow responsibilities. - [Detect SKILL.md](https://work-guide.openai.chatgpt.site/download/chief-of-staff/detect/SKILL.md): Verify changes using named, approved sources; produce an honest read-only receipt. - [Incubate SKILL.md](https://work-guide.openai.chatgpt.site/download/chief-of-staff/incubate/SKILL.md): Match a Detect receipt to an existing approved project and task; ask before changing or starting anything. - [Collaborate SKILL.md](https://work-guide.openai.chatgpt.site/download/chief-of-staff/collaborate/SKILL.md): Follow one explicitly requested active conversation for a defined time window; the existing owner retains decisions and outbound approval. - [Reflect SKILL.md](https://work-guide.openai.chatgpt.site/download/chief-of-staff/reflect/SKILL.md): Review completed evidence and propose a bounded improvement without changing instructions or policy. - [Weekly and monthly Skill care](https://work-guide.openai.chatgpt.site/prompts#skill-care): Review recent failures weekly, check the full workflow monthly, and retest after changes. - [Manager kickoff](https://work-guide.openai.chatgpt.site/prompts#manager-kickoff): Plan one privacy-conscious team pilot without monitoring employees or enabling automation. - [Site collaboration](https://work-guide.openai.chatgpt.site/prompts#site-collaboration): Name one human integration owner; use isolated Git branches or optional GitHub review only when available and approved. - [Working Sites playbook](https://work-guide.openai.chatgpt.site/guide/sites-as-working-artifacts): Compare one-owner handoffs and separately authorized shared-repository collaboration without implying access. ## Safety and authority - Reading this guide does not grant access to private sources or systems. - A working first-party link does not grant workspace access or permission. - Current capability, account access, and tool availability must be verified. - A Skill draft is not a tested Skill; one good example is not validation. - Ask before sending, posting, publishing, deploying, changing access, scheduling, purchasing, deleting, or making another consequential change. - Preserve the actual user's instructions, privacy, ownership, and judgment.