Three modes of one tool, not three commands. They answer "what do I have pending?", "what is blocked and why?" and "how is the roadmap going?" reading frontmatter only, across every organization at once — and the root's own work along with them, grouped separately.
How it is run
.os/core/lib/sweep.sh --brain . --mode pending
.os/core/lib/sweep.sh --brain . --mode blocked
.os/core/lib/sweep.sh --brain . --mode roadmap
The environment table —the one mapping a mounted node's remote to its local path on this machine—
is read on its own from mounts.md, at the root of the brain: --mounts <path> is an override and
does not need to be passed. With no table the scan covers the brain only and says so; with a mount
declared and not cloned, it reports it as unreachable and continues. mount-repo writes it.
What to do with the output
The output is the state. It is shown exactly as it came out and it is not recomputed, nor completed with inferences, nor summarized. Above it goes a one-line reading and a suggested next action.
Whatever shows up under "Sin clasificar" is never guessed. A head without status or without
horizon has no state the scan can deduce: the operator is offered the chance to write it, with the
path.
The body of an initiative is loaded only once the operator picks a focus. The scan reads frontmatter: enough to choose, not enough to work.