rosetta-d360-builder¶
Six step-by-step builder walkthroughs for Salesforce Data 360. Each skill
walks a build in order on the Salesforce Data 360 MCP tools (d360_*).
Every create or run step is followed by a read-back that checks it
landed. No skill ever calls a delete tool.
Every skill is DRAFT and UNVERIFIED. Each one opens with a banner that says so. Run them against a sandbox org only. A banner lifts only when a sandbox smoke passes and the skill is re-rendered. Until then the tool bindings behind the walks are drafted from a prose reference of the Data 360 tools, not from their real schemas.
| Skill | What it builds |
|---|---|
customer-360-foundation |
Ingest a CRM profile source, map DLO fields to DMOs, run identity resolution, verify unified profiles |
segment-to-activation |
Segment unified profiles, stand up a channel target, activate with a governed refresh |
zero-copy-federation |
Federate a Snowflake warehouse without ingestion |
unstructured-to-ai-grounding |
Index unstructured content and ground AI on it |
insight-and-semantic-layer |
Build a Calculated Insight and surface it through a semantic data model |
data-graph-real-time-action |
Act on customer data in real time through a data graph and a data action |
Each skill starts with preflight gates. These are decisions to settle before any tool call: org topology, the source ruling, the governance envelope and the economics.
What it needs¶
- A Claude Code session connected to a Salesforce Data 360 MCP server
that serves the
d360_*tools, pointed at a sandbox org. - Nothing from Rosetta. The skills are self-contained and call no Rosetta tool.
How the skills were made¶
The skills are compiled, not written by hand. Rosetta's d360-distiller
0.2.0 read this library's CP/, ADR/, EF/ and BI/ files at commit
60ce178 (snapshot content-2026-10-16-v1.2.0), in run
run-20261016-01. It compiled each walk onto tool bindings v0.1.2, and
each skill's colophon records its inputs: pattern uids, ADR ids, the
snapshot and the bindings version.
customer-360-foundationis the ratified reference render, filled from its fixture.- The other five are compiled renders. Their text is unchanged since 0.1.0, because no input they are compiled from has changed (see CHANGES).
Never edit a skill file here. A wrong step is fixed in the bindings or in
the graph, and the skill is re-rendered. To report one, propose a
correction as contributions/README.md describes.
This page as Markdown — for pasting into Claude or another assistant · llms.txt