Starting to use OpenSpec on a huge project

[colinmollenhour]

Does anyone have any experience or guidance or prompts and such to help with starting fresh with OpenSpec on a huge existing project (mainly PHP)? I imagine this needs to be done in several "layers" of discovery and what I don't want to end up with is thousands of highly detailed specs that are infeasible for any one person to ever even read or a huge burden to maintain. I have resources within the code repo but also wikis and docs pages I could import temporarily and MCPs I could connect to (ClickUp, GitLab), but there is just so much to start with.

Also, what models are best for handling large specs? I know some models have larger context windows than others, but also they decay at different rates. Lastly, what agents work best in your experience? I've seen that some seem to be way faster and more efficient at reading files. I've not done a formal test but for example opencode seems to work better than claude at grep/read if I remember correctly.

Thanks in advance!

[clay-good]

You don't need the layers of discovery — and the outcome you're trying to avoid is the one OpenSpec is designed to avoid too.

From Using OpenSpec in an Existing Project: "You do not document your whole codebase to start. You write specs only for what you're about to change."

So on a large PHP codebase: don't import the wikis, don't sweep the repo. Take one real change, run /opsx:explore to let the agent read just that area, then /opsx:propose → /opsx:apply → /opsx:archive. You end up with a spec for that slice only. Repeat, and specs/ grows around the work you actually do — never thousands of specs nobody reads.

On models: nothing OpenSpec-specific. Because specs stay change-sized rather than codebase-sized, context length stops being the binding constraint.

Closing as answered — reopen or comment if your case is different.