01 Get started
Choose a harness
Use Claude Code, Codex or another declared adapter without changing the proof standard.
Choose the working environment, keep the proof standard
A harness is the application that gives an assistant its tools, instructions and skill discovery. Analytical Spine targets those harness capabilities. Changing the model or chat window does not by itself establish that the required filesystem, shell, permission or execution facilities are present.
Choose a route your team can inspect and operate. Claude Code has a documented plugin route. Codex and other declared adapters use the deterministic library installer. Both routes still require approved distribution access, and neither supplies source permissions or business approval.
| Route | What to inspect |
|---|---|
| Claude Code plugin | Approved distribution, plugin discovery and explicit skill invocation. |
| Codex adapter | Target plan, parameters, installed discovery and readiness. |
| Other declared adapters | Required capabilities, stated degradation and the adapter selected by the plan. |
Claude Code: use the approved plugin source
The documented plugin is named analytics-spine. Once its distribution source has been made available through an approved access route, the installation command is:
/plugin install analytics-spine@analytical-solutionsThis is not evidence of a public marketplace listing. Public publication and clean-install verification remain release steps. Do not use an internal repository address as a public download link.
Invoke the front door explicitly. If another plugin owns the short name, use its namespaced form:
/analytics-spine/analytics-spine:analytics-spineThe front door guides the project workflow. A request to initialise a new project must not overwrite an existing proof. Review the evidence and current state before resuming work, rather than starting a new history to escape a failed gate.
Codex and other harnesses: inspect a deterministic plan
From the approved library checkout, select the target and adapter explicitly:
node tools/agent-skills.mjs plan suites/analytics-spine.json --target <target-repository> --profile full --agents openai-codexFor another harness, use the adapter declared by its plan. The suite documentation describes generic discovery for an Agent Skills harness. Do not assume every adapter has identical hooks, sandbox controls or invocation behavior. A missing required capability should remain visible instead of being silently replaced with an assistant's promise.
Follow the install guide for the matching apply, verification and readiness steps. Preserve the generated manifest and discovery ownership; manual copying makes later upgrades harder to verify.
Keep repository scope and user scope distinct
The Analytical Spine suite examples install into a repository. The library also supports user installation for individual packages that explicitly opt in. That facility does not make an entire suite a user-scope operation. Check the package contract and documented supported route before using it.
Repository installation gives the project a reviewable tool set and deployment state. It does not grant an assistant unrestricted source reads, network actions or target overwrites. Keep outward-action consent tied to the actual operation and exact plan.
Treat the assistant provider as a separate boundary
Information pasted into an external assistant follows that provider's controls. The Hosted ledger's value-free upload contract does not cover everything you might put into a prompt. Use synthetic metadata for learning and review what a tool or assistant receives before involving client evidence.
Read Claude Code and MCP for the read-only preview boundary and useful status questions.