Native Soil
Soil Cloud (Remote MCP)

The soil commands

The commands you use once Soil Cloud is connected · save, list, load, and info.

Once Soil Cloud is connected, you use four soil commands. You invoke them by asking the model in plain language ("save a Soil handover for this project"); the model calls the matching tool.

The commands

soil save

Captures the current conversation as a checked, graded handover. One command does the whole job: behind it, the model extracts the project's state against Soil's recipe and Soil runs the verification pipeline · extract → verify → drift check · before anything is stored.

  • A complete, high-confidence extraction passes the checks and is saved in one call, with a load code and a readiness band.
  • An incomplete one comes back "Needs work · missing: …" with the exact gaps to fill. Incomplete input is never saved.
  • A borderline one comes back asking for a drift-check pass: the model re-reads the handover cold, fixes what a fresh reader would have to guess, and saves again.

A bare soil save makes a personal, standalone snapshot, auto-named from the work. To name it yourself, write soil save #my-thread; every named save is kept, so saving the same name again preserves both as distinct handovers (#my-thread-001, -002, …) · it never overwrites. To save into a shared project you belong to · where your work accumulates over time · write soil save @project (the leading @ marks a project), or soil save #auth-flow @project to name the thread in the same breath. @ says where, # says which, in either order. Without @, the save is a personal snapshot. Everything stored is a safe projection.

As part of every save, the model also answers a few situational working-style questions. The answers are stored raw with the handover · never scored, never blocking the save · and render as a clearly labeled block when the handover is loaded.

soil list

Lists everything you can load: your saved handovers newest-first (title, project, saved-at, an exact load code, and the readiness band) as safe, code-first projections, plus the shared projects you belong to, so you can discover what @project addresses exist. soil list @project lists one project's threads (code, title, contributor, readiness) so you know what to load.

soil load

Pulls a saved handover into the current conversation as a paste-ready restore brief so a fresh thread or different model can continue:

You typeWhat loads
soil load #001One of your own handovers, by exact code.
soil load #my-threadOne of your own handovers, by name.
soil load @project #001One thread inside a project, by its code.
soil load @project #nameOne thread inside a project, by its name (e.g. soil load @project #auth-flow).
soil load @projectThe project: the project knowledge, a table of contents of every thread attributed, and the newest inline.

An unknown numeric code is reported as not-found · Soil never substitutes a different handover. A bare soil load #001 is always personal: a project thread needs the @project scope. The restore also carries the save's recorded working-style observations, clearly labeled · personal loads included.

soil info

Reports connection and account status: a masked account label (never your raw email), your project and handover counts, a few most recent handovers, and the shared team projects you belong to. The friendly first call after connecting · confirms your client reached Soil and is authenticated.

Under the hood

Tool names follow soil_* (e.g. soil_save, soil_list, soil_load, soil_info). There is also a helper tool, soil_prepare: it hands the model the extraction recipe at the start of a save · the sections to fill, the working-style questions for this save, and (for an @project save) the project-knowledge digest. The full recipe arrives as text in every client; clients with structured-tool-result support also get it as a machine-readable payload. The model calls it automatically as part of soil save · it is never something you type. In conversation you can just describe what you want · "save a Soil handover", "load my last Soil handover" · and the model does the rest.

Typical flow

  1. soil info · confirm you're connected.
  2. soil save · the model extracts, Soil verifies and stores (or tells you exactly what's missing).
  3. Later, in a new thread or model: soil listsoil load → continue.

See save a handover and load a handover for worked walkthroughs.

On this page