# Knowledge Worker companion for ChatGPT Work

This companion is generated from the same canonical framework as the human
Site. Use progressive disclosure: load only the chapter and template needed for
the current job.

## Read these first

- [Complete onboarding manual](https://work-guide.openai.chatgpt.site/START-HERE.md) teaches the full public-safe operating system during initial onboarding.
- [Complete Agent View](https://work-guide.openai.chatgpt.site/agent/agent-guide) shows the same manual in its actual current agent-readable surface.
- [Machine-readable index](https://work-guide.openai.chatgpt.site/llms.txt) lists the approved discovery and reference entry points.
- [Optional features and practical tools](https://work-guide.openai.chatgpt.site/features) explain useful capabilities, availability, and consent boundaries.
- [Remote and Voice walkthrough](https://work-guide.openai.chatgpt.site/remote) explains desktop pairing, the mobile Remote screen, and a new Voice conversation.
- [Focused prompt library](https://work-guide.openai.chatgpt.site/prompts) provides copyable starting requests for one specific job.
- [Sites and build prompts](https://work-guide.openai.chatgpt.site/sites) provides practical layouts and five portable prompts for your own sources.
- [Manager kickoff examples](https://work-guide.openai.chatgpt.site/manager#manager-kickoff-examples) provide a practical team path and fictional presentation examples.
- [Safe Site collaboration](https://work-guide.openai.chatgpt.site/prompts#site-collaboration) distinguishes a single integration owner from separately approved Git branches or optional GitHub review.

Opening any link remains subject to the actual account, existing workspace
access, available capability, and applicable human approval.

Available tools,
connectors, screenshots, computer history, Skills, and automations depend on
the user's account, workspace settings, permissions, and explicit consent.

## Default behavior

1. Start read-only.
2. Identify the decision, owner, sources, constraints, and approval boundary.
3. Load the smallest relevant stage module.
4. Use a template only when the stage calls for one.
5. Stop before sending, publishing, deploying, deleting, changing access, or
   taking another consequential action without explicit authorization.

## What good operation looks like

- Keep one project per real workstream and retain its existing owner.
- Separate global preferences from project-level AGENTS.md instructions.
- Inspect app shots or computer history only within an explicitly approved scope.
- Build personal or approved shared Skills from corrected real work; check them weekly and retest monthly.
- Use shared plugins only when available, explicitly approved, and appropriate for the existing owner.
- Record what a read-only heartbeat actually observed or advanced.
- Keep automations bounded, reviewable, deduplicated, and separately authorized.
- Use the Chief of Staff roles Detect, Incubate, Collaborate, and Reflect only when useful.
- Keep source quality, privacy, approval boundaries, and final decisions visible.

## Starting request

Read https://work-guide.openai.chatgpt.site/START-HERE.md and only the module needed for my
current job. Ask for the workstream, human owner, trusted sources, expected
result, checks, protected files, and required approvals. Help with one real or
clearly synthetic example, explain what you verified, and stop before any
external action or unapproved change. Suggest a reusable Skill only after the
method proves useful.

If the guide or module cannot be read here, ask once for that file and continue
with any fully specified approved draft. Label it a partial application of the
method; do not claim the guide was reviewed or bypass access.

Use the complete bundle only for a one-time, read-only assessment of a working
system. Do not load the entire guide into routine tasks.

## Routing

- **Mindset:** `chapters/01-mindset.md`
- **Context:** `chapters/02-context.md`
- **Workflow:** `chapters/03-workflow.md`
- **Skill:** `chapters/04-skill.md`
- **Automation:** `chapters/05-automation.md`
- **Operating system:** `chapters/06-operating-system.md`

See `manifest.yaml` for version, currentness, and module metadata.