cinder sun
I would like clearer integration between ChatGPT Projects, Saved Prompts, and Workspace Agents.
Saved Prompts can include capabilities/tools, and Workspace Agents can appear as selectable capabilities. That is useful, but it creates a key product question:
If I run a Saved Prompt inside a Project and it invokes a Workspace Agent, what context and runtime does it actually use?
Projects have their own context layer: Project instructions, files, chats, memory/chat history where available, and the current ChatGPT model/reasoning mode.
Workspace Agents also appear to have their own execution layer: agent instructions, files, skills, memory, connected apps/tools, approval rules, and optional extra reasoning effort.
So it is unclear whether a Saved Prompt inside a Project behaves like:
- manually pasting that same prompt into the Project chat;
- invoking the selected Workspace Agent with its own files, skills, and memory;
- a hybrid of Project context plus Agent context;
- or only a lightweight UI shortcut.
This matters for reliability. A user may expect Project files and instructions to apply, while the Agent may rely on its own files, skills, memory, and tools. It is also unclear whether the current ChatGPT model/reasoning setting applies, or whether the Agent uses a hidden runtime.
Suggested improvements:
- Show what runtime is used.
- Show whether the current Project model/reasoning setting applies.
- Clarify whether the selected Agent uses its own runtime.
- Show which context sources were available and used.
- Clarify whether Project files and Agent files are merged, isolated, or selectively included.
- Preserve approval gates for write actions.
- Let users bind Saved Prompts to Projects as visible workflow buttons.
- Let users choose Project context, Agent context, or both.
Core need: transparency. Saved Prompts could become workflow launchers inside Projects, but users need to know what context, tools, files, memories, and runtime they are using.