Use a project for a coherent area of work
The project documentation distinguishes ChatGPT projects with uploaded or connected sources from local projects with folder access. Naming a ChatGPT project after a folder does not grant access to that folder. Choose the project type according to the actual sources and intended work location.
Example: a reliable proposal baseline
Our proposal for a service business: include the approved service description, current proposal structure and a few suitable writing examples. Add client-specific information for each assignment. Do not add private notes, old prices or third-party contract drafts to the shared context without review.
Require the draft to flag unsupported service commitments as open questions. The domain review can then focus on real decisions: what is included, what is excluded and which assumption needs to be clarified with the client?
Separate workflow guidance from source material
Project instructions describe the shared working agreement. A skill can package a reusable workflow with instructions, resources and optional scripts. For our proposal process, that workflow could identify gaps in the brief, create sections and list commitments for human approval.
- Sources: what is factually authoritative, and which version establishes it?
- Instructions: how should conflicts, tone and missing information be handled?
- Output: which file or decision is actually needed?
Keeping knowledge current is a team responsibility
Assign each central source an owner and a review trigger, such as a new service or revised proposal terms. Replace superseded versions consistently. A small source register with status and validity makes conflicts easier to spot than adding more prompt rules.
After a material change, test a familiar request. If the new draft still uses an old service or wording, look for remaining sources and context copies. Add further material only when it improves the actual assignment.
Organize access and handover
Extensions involve several control layers: plugin availability, the underlying app, allowed actions and source-system permissions. Test with a typical team role. A successful administrator sign-in does not prove that the intended employees can use the same information appropriately.
A Starter Setup is useful when knowledge exists but briefs and outputs vary widely. In the AI System Check, describe the recurring assignment and the sources people currently need to gather for it.
Keep it verifiable
Primary sources
- Projects and chatsSource checked:
- Build skillsSource checked:
- Plugin controlsSource checked:
