Skip to content

Connect your first system

Treat each connection as a governed delivery slice: adopt or open one public issue with scope, non-scope, dependencies, measurable acceptance criteria, negative probes, evidence, recovery and non-claims. Link implementation pull requests and evidence to that issue; keep unpatched security-sensitive details in the private path defined by SECURITY.md. Locally validated describes evidence maturity, while released describes published bytes. Neither label proves live-system or production fitness.

This guide distinguishes the runnable current public release from released local-synthetic contracts, locally validated work and product plans. The historical v0.1.0 tag is the initial public baseline, not the current release or its direct predecessor. A roadmap or issue is not runtime evidence.

What you can run in the public snapshot

The released connection path is the bundled, synthetic EspoCRM-to-Dolibarr demo. It does not connect an existing company system.

  1. Read the Quickstart, Architecture and Known Limitations.

  2. On a supported disposable or development host, run from the release root:

    sh
    ./demo/install.sh
  3. Confirm that the installer reports READY_VERIFIED. The installer selects the safe local profile, creates file-backed random demo secrets, loads only fictional fixtures, performs the governed synthetic flow and verifies the provider result through readback.

  4. Inspect the shipped connection inputs rather than replacing them with real credentials or data:

    • demo/manifests/catalog/crm-erp-playable-v1.json — bundled capability and template catalog;
    • demo/manifests/identity/panskys-zoo-v1.json — fictional identity and provider-role mapping;
    • demo/manifests/authority/SAFE_GUIDED-v1.json — safe authority default;
    • demo/manifests/network/local-egress-policy-v1.json — local network boundary;
    • demo/readback.sh — semantic provider-readback implementation.
  5. Reset installer-owned state with:

    sh
    ./demo/uninstall.sh --purge

This proves only the pinned local demo path. It does not prove production identity, arbitrary adapters, tenant isolation, provider rollback or safe use of real data.

Governed onboarding blueprint

Use this sequence to design a future connection. Steps marked PLANNED are requirements, not executable arbitrary-system instructions in the current release.

1. Choose the outcome and source

PLANNED: Name one bounded use case, its authoritative source system, the objects and fields required, freshness needs, expected evidence and an owner. Start with the smallest useful read-only slice. A direct database write is not a safe default.

2. Check profile, adapter and prerequisites

PLANNED: Select only a versioned, supported system profile or adapter and verify its API/export version, scopes, network route, rate limits, tenant model, reset behavior and readback support. If no compatible profile exists, stop at design; do not treat a roadmap issue or catalog name as implementation proof.

3. Formalize the knowledge contract

RELEASED LOCAL-SYNTHETIC AUTHORING/VALIDATION CONTRACT: Prepare a vendor-neutral System Advisor Guide, machine-readable Machine Manifest and explicit Capability Mapping. The published contracts and synthetic fixtures can author and validate typed objects, dependencies, cause/effect/context, supported operations, safe defaults, failure modes, setup/reset and required evidence without granting Authority. JSON, YAML and Markdown are the knowledge formats; MCP may be an access channel but is not the definition.

PLANNED LIVE REALIZATION: discovering those facts from an existing system, proving them against a tenant/provider, and activating the resulting binding are not implemented by this authoring surface.

4. Bind identity and boundaries

PLANNED: Assign a dedicated Workload Identity and tenant. Refer to credentials through a secret handle or file-backed secret; never store plaintext secrets in Guides, manifests, source control, logs or contribution bundles. Declare outbound network destinations and deny direct database access unless a separately reviewed design requires it.

The released local-synthetic Microsoft contract does not register an application, acquire or validate a live token, request consent, store a credential or activate a tenant. Its Entra profile and the five-operation Power Platform read connector bind exactly cm.discovery.read. cm.operator.read is reserved for a future separate administrative-read Profile and is invalid on the released connector.

5. Select the owner profile

  • SAFE_GUIDED is the required default for planning and initial use.
  • CUSTOM content remains untrusted until provenance, compatibility and owner review are complete; importing content never activates authority.
  • FULL_CONTROL_LAB is only for an isolated disposable lab with explicit owner risk acceptance. It may bypass PanSphaira action and Approval gates up to the host process's OS/host ceiling, so bypassed layers are outside SAFE_GUIDED/Canon security claims. It is not a production shortcut and must bind reset/rollback/recovery and reset to SAFE_GUIDED on restart, revoke or cleanup.

The current public release contains setup-planning and Builder contract code, tests and synthetic fixtures for these fail-closed rules. It does not ship a user-facing arbitrary-system builder or automatic adapter generator. The contract surface is RELEASED WITH LOCAL-SYNTHETIC EVIDENCE; live onboarding and production fitness remain unproven.

6. Preflight effective rights

PLANNED: Run a dry-run/preflight before connectivity. Produce a Permission X-ray of the effective identity, tenant, scopes, capabilities, network access, policy constraints and requested effects. Missing, broader-than-declared or cross-tenant rights must fail closed. The current release has no user-facing command for this arbitrary-system preflight.

The released setup-planning and Builder contracts can validate closed local-synthetic inputs and produce inactive plans. That is not credential preflight, live tenant discovery or provider activation.

7. Connect read-only first

PLANNED: Establish a typed, schema-checked, least-privilege read-only connection through a supported application API or export. Prove tenant and field boundaries, deterministic mapping, pagination, error handling and source readback before considering writes.

Only then may an owner separately approve a narrow, controlled and reversible write capability. Each write needs a typed request, policy/use-time check, readable diff or equivalent preview, explicit authority, idempotency strategy, provider readback and a tested reversal or recovery path.

8. Reconcile, receipt and recover

PLANNED: Compare the result with the authoritative source, reconcile canonical identifiers and record a digest-bound receipt with provenance, tenant, policy, input and readback references. Define Reset, provider Rollback, authority Revoke and installer Cleanup separately; they are not synonyms.

9. Add BI and share deliberately

PLANNED, OPTIONAL: Add a versioned BI Semantic Contract and Dashboard Pack only after the read-only mappings and lineage are verified. The released Builder contract can create a closed, synthetic, digest-only contribution bundle, but it remains opt-in and carries no publication authorization. The released HMI contribution preflight can prepare canonical, digest-bound local-synthetic contribution bytes. Neither surface submits an issue, publishes content, writes externally, uses credentials or carries publication authority. Publication still requires a separate authorized Owner action outside these contracts. Shared or untrusted knowledge never grants authority, and private company data is never published automatically.

Maturity summary

MaturityScope
Released local PoCPinned synthetic CRM/ERP installation, SAFE_GUIDED local default, generated file-backed demo secrets, governed synthetic effect, provider readback, receipt and ownership-scoped cleanup.
Released local-synthetic contractsSetup-planning, System Advisor/Machine Manifest authoring and validation, Builder planning, generic inactive scaffolds, synthetic evidence, two-system fixture reuse, HMI contribution preflight and sanitized opt-in packaging. The Entra/Power Platform contracts fix five reads to cm.discovery.read. There is no user-facing onboarding UI or CLI for external systems, and no submission/publication/external-write effect.
PlannedLive-system discovery and profiles, credentials, registration/consent, Workload Identity/tenant/provider activation, cm.operator.read administrative profile, Permission X-ray UI, live external connectors, production reversible-write onboarding and BI packs.

Keep planned work behind its own evidence gate. Promotion requires shipped artifacts, reproducible tests and current readback evidence—not an issue, design document or catalog entry alone.

Local synthetic proof of concept — not a production release or security certification.