The setup for Senso's Shared Context path — getting a first document into the organization's knowledge base so the team's agents have something to retrieve, and starting the owner-and-status convention with it. Use when senso-shared-context hands here because the knowledge base is empty, or when a user says "set up Shared Context". Not for the Verification Loop's setup (senso-verification-loop-setup), and not for a user who named material to save (senso-context-layer).
npx @senso-ai/shipables install senso-ai/senso-shared-context-setupShared Context needs one thing before its loop can run: a document in the knowledge base. That
is the whole of setup. This skill owns that chore, tags the document the way the loop expects, and
hands to senso-shared-context, whose opener carries the setup close. No industry, no brand name,
no content type, no brand kit — those are the Verification Loop's and this path does not need them.
senso whoami first, once. If the router already ran it, reuse the organization's name. Say the
name in a sentence; never the record.
Never hijack a stated request. "Save revenue.pdf to Senso" is a save — senso-context-layer's,
even on an empty organization. An empty context layer is not permission to run setup.
Lead. Do not ask permission to continue. The one question that stays a question is which door their first document comes through. "Shall I continue?" is never one.
Never narrate the machinery — and the line during a wait is in their terms. "Saving that and checking any agent on your team can find it…" — never "polling the ingestion status".
A choice is offered as options, not as a sentence — the recommended option first, and in the message body too. Free text is never boxed.
Nothing is saved that the user has not read. The written-up note is shown, corrected, then saved.
Say what the context layer is, once, where the term lands — the shared knowledge base every agent on your team reads from — and what it buys them: what one person saves is what the next person's agent starts from.
senso kb stats --output json --quiet # total_files — anything in there yet?
Something already in (total_files > 0): nothing to do. Load senso-shared-context and go.
Never redo work they have done, and never congratulate someone on it.
Nothing in: the one chore, below.
Your context layer is empty — that's the shared knowledge base every agent on your team reads from, so what one person saves is what the next person's agent starts from. The first step is one document in it. Tell me in a few lines what your team does and what your agents should know, and I'll write it up — or point me at a file you have.
Options: I'll write it up (Recommended) · Upload a file.
Then senso-context-layer → Add something → Which door — write it up, or upload it — the
module owns both doors and every command:
create-raw. Markdown goes in
through create-raw, never kb upload — the upload endpoint rejects .md by filename.Name the folder yourself and say so — there is nothing to choose between yet. shared-context
is the name: the default home for what the team settles through this loop, and theirs to
reorganize — subfolders, renames, a scheme of their own — whenever they like. The module's rule
that a folder is never invented silently has this one exception, and the line "I've put it in a
folder called shared-context — the default home for what your team settles; reorganize it however
you like" is what makes it not silent. An organization that already has material and its own
folders is not a fresh one — this chore only runs on an empty knowledge base, so it never
imposes the folder on a team that has its own way.
Then tag it — senso-context-layer → Organize, kb tags set with status:approved and
owner:<the user's name, as they give it>. Ask the name in the same breath as the yes — "and
I'll mark it approved, owner you — what name should that be?". This is where the loop's convention
starts: an approved note with an owner is what agents may treat as settled; the loop's skill owns
the convention, this chore only applies it to the first document.
Then senso-context-layer → After any write — poll until it is searchable and confirm with
one search that cites it. Nothing is said about that here; it is the Landed fact the setup close
carries.
Load senso-shared-context and say nothing here. The message that closes setup — "That's setup
done — and you won't do it again", the document's title, and that any agent on the team can
retrieve it — is the first block of the loop's opener: senso-shared-context → When setup ran
this session, the opener's first block is the setup close. A close sent from here, before the
loop's reads, is not produced on every model; the opener is the message that ends the turn, so it
carries the close. Keep in hand what that block needs: the document's title, the folder, which
door it came through, and the owner's name.
No Verification Loop chores — no industry, no brand name, no content type, no brand kit. Those
are senso-verification-loop-setup's and Shared Context does not need them.
No knowledge base commands of its own — senso-context-layer owns every one, including the
tags.
No second document. One is setup; the rest is the loop, and the user saying what to save.
| They want | Skill |
|---|---|
| Save, find or organize named material | senso-context-layer |
| The Verification Loop, or its setup | senso-verification-loop |
| To get started, path not yet chosen | senso-quickstart |