Cisco Data Fabric Setup

Cisco Data Fabric Setup

Prerequisites

Workflow Overview

┌───────────┐   ┌───────────────┐   ┌───────────────┐   ┌─────────────────┐
│ Preflight │ → │ Render/review │ → │ Apply/handoff │ → │ Validate evidence │
└───────────┘   └───────────────┘   └───────────────┘   └─────────────────┘

When to Activate

• Users need Cisco Data Fabric architecture, feature or product coverage, readiness assessments, gap analysis, machine-data activation, federation targets, storage tiering, AI-ready data, or AgenticOps data foundation requests. Distinguish.
• Preview and review the cisco data fabric setup workflow before any live apply phase.
• Diagnose failed prerequisites, generated assets, configuration, or validation evidence.

Scope

Follow the documented read-only or render-first path whenever it is available. This skill does not imply permission to mutate live systems. Require explicit apply flags, protected credentials, and operator review for state changes.

Examples

Inspect the supported setup modes before selecting one:

bash skills/cisco-data-fabric-setup/scripts/setup.sh --help

Expected output: usage, supported modes, and required arguments are displayed without changing the target environment.

Inspect validation modes before running completion checks:

bash skills/cisco-data-fabric-setup/scripts/validate.sh --help

Expected output: offline, live, and completion options are displayed when the skill supports them; help exits without mutation.

Troubleshooting

Treat Cisco Data Fabric as the architecture powered by the Splunk Platform, not as a single installer, SKU, API, or storage product. Render a complete coverage packet first, then execute only reviewed child-skill plans whose public contracts and required non-secret inputs are available.

Decision Workflow

1. Read references/feature-matrix.md for the architecture, feature stages, product owners, and boundaries.
2. Read references/research-ledger.md before changing availability claims, federation targets, model status, or product naming.
3. Collect a non-secret intake from template.example.
4. Render and validate the packet.
5. Review gap-register.md, doctor-report.md, and the product and federation matrices before executing delegated renders.
6. Apply or validate state through the owning child skill, never through a fabricated Cisco Data Fabric API.

Architecture Lanes

1. Data access and management: collect, inspect, filter, shape, redact, route, tier, and monitor machine data with supported Splunk ingestion, Edge Processor, Ingest Processor, SPL2, and Ingest Monitoring workflows.
2. Federation: route Splunk-to-Splunk and current Splunk Cloud Data Management connection/dataset work to splunk-federated-search-setup. Track Amazon S3, Microsoft Azure, Azure Databricks, Snowflake, DDSS, and Amazon Security Lake separately; do not infer equal stage or entitlement. Product lifecycle and tenant access are separate fields: for example, Amazon S3 federation and Federated Analytics can be GA while still requiring sales activation, scan entitlement, a premium add-on, or topology gates.
3. Storage and catalog: distinguish the real-time Splunk index, external stores, DDSS/DDAA/SmartStore adjacencies, and the alpha Machine Data Lake. Built-in Data Catalog and Machine Data Lake remain readiness handoffs until stable public administration contracts exist.
4. Context and governance: cover schema and catalog metadata, knowledge objects, CIM/OCSF, ITSI/business context, RBAC, lineage, audit, human approval, and downstream data-readiness evidence.
5. AI activation and action: delegate AI Toolkit/PSC/DSDL, hosted-model readiness, and MCP Server. Distinguish the available open Cisco Time Series Model 1.0 from the GA hosted Cisco Deep Time Series Model integration, keep Agent Launchpad at its documented GA boundary with its region, egress, connection, and enabled-agent gates, and keep AI Canvas at its CA boundary.
6. Cross-domain experience: represent SecOps, ITOps, Engineering/DevOps, NetOps, Splunk Enterprise Security, ITSI, Observability Cloud, Cisco Cloud Control, AI Canvas, and Cisco product telemetry as consumers or handoffs, not as interchangeable Data Fabric components.

Safe First Command

bash skills/cisco-data-fabric-setup/scripts/setup.sh --help

Primary Workflow

Render and validate the complete evidence-backed packet:

bash skills/cisco-data-fabric-setup/scripts/setup.sh \
  --render --validate \
  --spec skills/cisco-data-fabric-setup/template.example \
  --output-dir cisco-data-fabric-rendered

Run the gap/readiness doctor:

bash skills/cisco-data-fabric-setup/scripts/setup.sh \
  --doctor \
  --spec skills/cisco-data-fabric-setup/template.example \
  --output-dir cisco-data-fabric-rendered

Preview delegated commands without writing or executing:

bash skills/cisco-data-fabric-setup/scripts/setup.sh \
  --execute data-management,federation,ai-activation,context-governance \
  --dry-run --json \
  --spec skills/cisco-data-fabric-setup/template.example

Execute only reviewed delegated render/doctor commands:

bash skills/cisco-data-fabric-setup/scripts/setup.sh \
  --execute data-management,ai-activation \
  --accept-execute \
  --spec skills/cisco-data-fabric-setup/template.example \
  --output-dir cisco-data-fabric-rendered

CLI Contract

setup.sh supports --render, --validate, --doctor, --execute SECTION[,SECTION], --accept-execute, --dry-run, --json, --spec PATH, and --output-dir DIR.

Delegated sections:

• data-management
• federation
• ai-activation
• context-governance

Handoff-only sections, which refuse explicit execution:

• storage-catalog
• experience

Child commands are render, doctor, or plan operations. Applying their output requires the child skill's own explicit approval gates. A missing child spec, tenant URL, MCP URL, entitlement, or public API is reported as a gap rather than silently converted into a successful apply.

Non-dry-run delegation fails before any child command when gap-register.json contains an error, a selected section is handoff-only, or any selected executable section has no reviewed command. This prevents partial execution when a later lane is missing required intake.

Non-Negotiable Boundaries

• Do not claim a direct Cisco Data Fabric management API.
• Do not call Machine Data Lake or its built-in Data Catalog GA; current public material identifies Machine Data Lake as alpha.
• Do not collapse store-specific federation stage, region, role, catalog, and entitlement requirements into a generic "Federated Search is GA" claim.
• Do not use activation_required as a product lifecycle. Record lifecycle in product_stage and tenant/commercial gates in access_requirement.
• Do not create new legacy Amazon S3 federated providers on Splunk Cloud 10.5; the old provider/index path is deprecated and migrated to the Data Management connection/dataset model.
• Do not promote announcement dates to current availability. Cisco Time Series Model 1.0 is published as an open Apache-2.0 model and the hosted Cisco Deep Time Series Model integration is generally available since AI Toolkit 6.0.0, but they remain separately governed layers and neither one settles the other.
• Do not pair AI Toolkit 6.0.2 with a PSC release below 4.3.4. The audited baseline is 6.0.2 with PSC 4.3.4 on Python 3.13; PSC 4.3.2 only applies back at AI Toolkit 5.7.4.
• Do not describe Splunk AI Toolkit Agent Launchpad as alpha, private preview, or a Fall 2026 GA target. Current AI Toolkit documentation makes it generally available since 6.0.0. Report it as unreachable only when a readiness gate actually fails: an unsupported AWS region or a missing region egress IP in the stack apiAllowlistIP, no supported LLM connection, or no enabled agent. Splunk Enterprise reaches it through the Splunk Cloud Connect app rather than being unsupported.
• Do not treat Cisco Security Analytics and Logging (SAL) as Splunk Machine Data Lake or as an automatically configured federation target.
• For AI Canvas with Splunk, require Cloud Control enablement, Splunk Cloud 10.5.2605.3, current AI Assistant and MCP Server, and mcp_tool_execute; retain the 100-row-per-card and forbidden-SPL-command limitations in the production handoff.
• Never accept raw tokens, passwords, API keys, client secrets, or private keys in chat, argv, specs, or rendered artifacts.

Validation

bash skills/cisco-data-fabric-setup/scripts/validate.sh \
  --output-dir cisco-data-fabric-rendered

python3 -m py_compile \
  skills/cisco-data-fabric-setup/scripts/render_assets.py

Read reference.md for the rendered artifact contract and delegated owner map.