Salesforce Data 360, made agent-ready

The Data 360 kit for teams that want answers with proof.

Data360 Beast packages the preflight contract, phase proof matrix, limits-source precedence, model-gallery lessons, RAG retrieval playbook, proof ledger evidence, and validation loop an AI agent needs to work with Salesforce Data 360 without guessing.

Read llms.txt Preflight View skill pack Limits rule RAG playbook MCP installs
17 agent skills 12 proof routes 14 model diagrams 45 RAG pages 194 API operations 813 schemas 100 docs indexed 
npx skills add architect-bertie/data360beast

Use Data360 Beast when:
- running target-org/data-space preflight
- designing a Data 360 architecture
- choosing the required proof path
- finding the right Connect API endpoint
- designing RAG, search indexes, and retrievers
- creating query, segment, activation, or data action payloads
- validating with org metadata, status, counts, and readback

Agent-ready by design

Readable by people. Ingestible through one project-owned agent entry point.

Inspired by the Agentforce Vibes Library pattern: simple repository shape, explicit skill folder, and compact machine-readable entry points.

01

Skill pack

skills/ ships the Beast router plus 16 specialist Data 360 skills.

02

llms.txt

/llms.txt points models to the right files before they crawl the repo randomly.

03

Manifest

/agent-manifest.json exposes topics, scores, entry points, and validation status.

04

Proof matrix

phase-proof-matrix.json maps phases to sources, tools, proofs, and assumptions to avoid.

Full specialist layer

Install the repo. Get the Data 360 specialists.

The pack now includes the orchestration skill, Connect API skill, and specialist phase skills for the full Data 360 delivery lifecycle.

data360beast sf-datacloud sf-datacloud-connectapi sf-datacloud-connect sf-datacloud-prepare sf-datacloud-harmonize sf-datacloud-governance sf-datacloud-retrieve sf-datacloud-calculated-insights sf-datacloud-segment sf-datacloud-act sf-datacloud-automation sf-datacloud-semantic-layer sf-datacloud-ai-models sf-datacloud-unstructured-retrieval sf-datacloud-analytics sf-datacloud-metadata-agentic

The system

One loop. Five truth sources. No folklore.

Data 360 work fails when teams mix stale docs, guessed payloads, and unverified org state. Beast mode makes the source order explicit.

  1. Run preflight. Capture target org, API version, data space, persona, lifecycle, authorization boundary, tools, and proof target.
  2. Classify the phase. Use the proof matrix to route connect, prepare, harmonize, govern, retrieve, insight, semantic, AI/search, segment, act, automate, or package work.
  3. Choose the model path. Use the public model-gallery map for anchor DMO, grain, and relationship design.
  4. Design retrieval deliberately. For RAG, choose ADL or manual setup, field roles, chunking, search type, retriever filters, and prompt scope.
  5. Apply limits precedence. Use current Data 360 limits first; use legacy CDP limits only for explicit legacy scope or comparison.
  6. Fetch official docs. Use Salesforce docs for setup, permissions, behavior, and exact limit readback.
  7. Search OpenAPI. Use the catalog for method, path, params, schemas, and version gates.
  8. Apply proof ledger evidence. Start from tested caveats and known validation paths.
  9. Validate in org. Prove with status, counts, metadata, data space, and readback.
Data360 Beast operating map showing docs, OpenAPI, proof ledger, skills, and live org proof.

Proof ledger

The schema told us what was possible. The proof ledger records what actually worked.

We tested core Data 360 Connect API surfaces in a live Data 360 lab org and documented the evidence, caveats, and promotion status that docs and OpenAPI alone do not reveal.

Live-tested

  • Data spaces and metadata
  • Query SQL and Profile API
  • DBT segment create/read/count
  • Data action target and action create
  • Activation target create/read
  • Search index and calculated insight list

Captured gotchas

  • includeDbt.models.models[] is the working DBT segment create shape.
  • Readback normalizes DBT models to includeDbt.models[].
  • Approximate segment count can fail when the feature is disabled.
  • Segment members use a delta window; status/count readback is primary proof.
  • Activation target readback was reliable by returned ID.

Scorecard

The API skill moved from useful to field-ready.

The site is public; the raw Salesforce Help cache is not. That keeps the repo clean, legal, and useful for teammates, customers, and agents.

Connect API readiness 5.5 9.7
Core Data 360 architecture 8.0 9.5
Model and DMO design 6.5 9.4
Unstructured retrieval and RAG 6.8 9.5
Governance and data spaces 7.5 9.0
Segment, act, automation 7.5 9.2
Deterministic proof routing 7.0 9.6
Overall Beast mode 7.3 9.8

Repository map

Small enough to scan. Structured enough to ingest.

data360beast/
|-- skills/data360beast/SKILL.md
|-- skills/sf-datacloud-connectapi/SKILL.md
|-- skills/sf-datacloud-*/SKILL.md
|-- AGENTS.md
|-- manifest.json
|-- llms.txt
`-- docs/
    |-- index.html
    |-- llms.txt
    |-- llms-full.txt
    |-- agent-manifest.json
    |-- agent-quickstart.md
    |-- beast-preflight.md
    |-- beast-evals.md
    |-- phase-proof-matrix.json
    |-- mcp-dependencies.md
    |-- skills.md
    |-- operating-model.md
    |-- proof-ledger.md
    |-- scorecard.md
    `-- data360/
        |-- architecture-engine-map.md
        |-- interoperability-decision-map.md
        |-- limits-source-precedence.md
        |-- model-gallery-implementation-map.md
        |-- rag-search-index-retriever-playbook.md
        |-- help/index.md
        `-- developer/index.md

Next frontier

The remaining work is targeted, not vague.

External activation destinations and tested search-index creation evidence need specific enabled assets. When those exist, the same loop applies: create a controlled lab case, test it live, capture the evidence, and promote it into the skill.

Open the repo