Load a handover
Pick a project up cold in a new thread, model, or provider · without drift · using soil list and soil load.
You've saved a handover. Now start fresh · a new thread, a different model, even a different provider · and continue without re-explaining everything.
Open a fresh conversation
Start a new thread, or switch models/providers entirely. As long as Soil Cloud is connected (or you use the copy-paste backstop), the new model can reach your handovers.
Find the handover
Ask: "soil list" (or "list my Soil handovers"). The model shows your saved
handovers newest-first, each with an exact load code (#001, #002, …) and its
readiness band. For a shared team project, soil list @project lists the
project's threads: code, title, contributor, readiness. (You can also browse
everything in the console dashboard.)
Load it
Load by exact code · soil load #001 · or by name · soil load #my-thread, if you named the save (soil save #my-thread). The model pulls in
the restore prompt: a comprehensive, paste-ready brief assembled from all
the saved sections · project identity, current task, locked decisions and their
reasons, next steps, and the rest · that boots the model back into the work.
A numeric #NNN is an exact code: if it's not found, Soil says so rather than
loading a different handover. It never silently substitutes another one.
Working in a project? Reach it with the @project
scope · @ says where, # says which:
soil load @acme-app #002orsoil load @acme-app #auth-flow· one thread inside the project, by its code or by its name · the everyday load.soil load @acme-app· the project in one restore: the project knowledge in full, a table of contents of every thread attributed per contributor, and the newest threads restored inline. Any older thread is listed rather than included · read one whole withsoil load @acme-app #NNN. The accumulated state, without the whole archive landing in your context.
A bare soil load #002 is always personal · reaching a project thread needs
the @project scope.
Continue · and trust the labels
The model now has the executive summary, workflow rules, architecture, implementation state, next steps, and risks. The provenance labels (Stated / Inferred / Unknown) tell it how much to lean on each claim · Stated is solid, Inferred is worth a check, Unknown is a known gap to resolve.
The restore also carries the save's recorded working-style observations · the answers the saving model gave to a few situational questions, in a clearly labeled block. They are raw observations for the new model to match its behavior against, not a scored profile.
One more trick worth knowing: if the session starts to drift later · the tone slips, or a question you already settled quietly reopens · re-paste the restore prompt. The same document that boots a fresh session also re-anchors a running one, pulling the model back to the locked decisions and the project's own terms.
No client support? No problem
If your tool can't connect over MCP, open the console
dashboard, open the handover, and copy its restore
prompt · the same assembled brief soil load produces. Paste it into the
conversation. No connector needed; the copy-paste backstop works anywhere.
Save a handover
Capture a live conversation's project state into a checked, graded handover with one command · soil save.
Rescue a dead ChatGPT thread
Recover a stuck ChatGPT chat where Soil's tools went missing · run the rescue prompt, paste the result into Import, and get a real handover with its own load code.