dip carries a set of skills — instructions written for a coding agent, not for you — that teach
Claude Code, Codex or Cursor how to work with your tenant: how to connect a data source, how
to use the MCP server well, and how to author a data app or governed agent. They ship inside the dip binary itself,
so they describe your installed version of the platform and need no network access to read.
The installed skill names keep their altimate- prefix. They describe Dippr workspaces; use the
names listed below when selecting or installing a skill.
If dip update --check shows your machine pinned to an older beta while your workspace uses stable,
run dip update to clear the pin and install the latest stable. A beta newer than stable is not
behind it; keep an intentional beta pin unless you want to return to stable. When the workspace
itself pins a beta, use the reported tag with dip update --version <tag>.
What each skill teaches
An agent picks these up on its own once they’re installed — you don’t invoke them by name, and
their trigger phrases fire whenever a conversation is shaped like the thing they teach (connecting a
source, using the MCP tools, authoring an app manifest, and so on).
Sharing data with your team
When you ask your agent for something other people will open, such as “a dashboard of my team’s open deals by region that I can share with my team”, the skills and the MCP server tell it to build a data app rather than a page of copied rows. A data app queries your tenant as the person viewing it, so each colleague sees only the rows their own access allows. A chat artifact, an HTML page, a file, a screenshot or a spreadsheet made from rows your agent read would show everyone the rows you can see, so the skills tell your agent not to use one for data from your tenant. Members publish and install their own static apps by default.dip app push needs Docker with buildx running on your
computer; when it is not, and the Dippr MCP tools are connected, your agent publishes a read-only page
with the MCP app_publish tool instead, so you get a link without installing anything. Docker comes up
later, only for apps that need it (Python, Streamlit, certified metrics).
Your agent asks you to confirm the install once, then gives you a link for your team. If your role
has own apps turned off, the skills tell your agent to say who can turn them back on. See
Build and share your first app.
Installing them
dip login installs (or refreshes) every skill as one of its last steps, right before
registering your MCP servers — most people never run
anything on this page directly. Pass --no-skills to dip login to skip it.
~/.claude/skills for Claude Code,
~/.agents/skills for Codex and Cursor) by default — they’re the same platform knowledge in every
project you open, so their scope matches your sign-in, not any one checkout. Pass --dir DIR to
install into a project instead (the opt-in for CI, or a repo that wants a pinned copy); that path
never overwrites a file it didn’t create without --force, and even --force refuses a file you
wrote yourself. --agent claude|codex|cursor|all narrows which tool’s discovery path gets written.
Staying current
Every filedip installs starts with a marker naming the version that wrote it. A newer dip —
including one that just replaced itself through its own automatic
update — rewrites only the files carrying that
marker to match, on its next command, and leaves anything you wrote yourself alone. Run dip update --check to see whether your installed skills are behind without changing anything.
The write path: what an agent can actually configure
Beyond reading,dip gives your agent commands that change your tenant directly — the same
onboarding steps a person would otherwise click through in the console:
dip login (they’re not reachable with an alt_ API key from the
MCP server itself) and, beyond the login scopes, your own role has to permit the action — the same
rule that governs the console:
See Roles and permissions matrix for what each built-in role includes.
Nothing here writes a secret to the command line — a source’s credential is always read from
standard input, never a flag, so it never lands in your shell history.
This write path configures your tenant’s connections and infrastructure. It’s separate from
the MCP server’s own tools, which only ever read your data and certified metrics — see MCP
tools. An agent can draft metrics and business concepts:
dip semantic create/import/update and dip ontology write drafts and propose submits them for review,
and over MCP suggest_metric files a suggestion. Certifying one stays a console-only, human
decision.Routine investigations
Theroutines skill turns a recurring question into an instruction with verified sources,
thresholds, branch rules, stop rules and stable finding keys. It tests once with
run_routine_now, reads get_routine_run, and shows you the result and coverage before proposing
save_routine. Saving without explicit start stays pending until you confirm in a fresh console session. An explicit personal start begins without console confirmation.
It uses the platform runner rather than asking you to build an agent image.
See Routine tools for access and availability.
Related
Install dip and sign in
Where skill installation fits into signing in.
Connect an AI agent
The MCP connection these skills teach your agent to use.