Oci terraform authoring

OCI Terraform authoring

Create reviewable infrastructure artifacts without contacting OCI. Contact OCI only for discovery, planning, or an explicitly approved apply/destroy after a named-context preflight.

Workflow

1. Establish the owner. Default durable resources to terraform; never give the same resource to direct CLI and Terraform.
2. Scaffold with ../../scripts/oci_tf.sh scaffold <dir> --name <name> immediately; it is a safe offline action.
3. Resolve resource fields from terraform providers schema -json after terraform init, or from the current official OCI provider docs. Never invent a field.
4. Keep credentials out of HCL and variable files. Use provider config, environment variables, workload/resource principals, or Vault references.
5. Validate with oci_tf.sh validate <dir> and run .tftest.hcl tests with mocked providers where possible. If a local prerequisite is missing, deliver the scaffold and the exact next local command rather than stopping the workflow.
6. Preflight the named context (./scripts/oci_preflight.sh -c "$COMPARTMENT_OCID") only before creating a binary reviewed plan with oci_tf.sh plan; stop if the resolved tenancy/compartment does not match the intended target.
7. Inspect the metadata-only plan summary. Stop on unexpected replacement/deletion, public exposure, or secret-bearing resources.
8. Apply the unchanged plan with oci_tf.sh apply; use destroy only with a separately reviewed destroy plan and destructive approval.
9. Verify resource state and reconcile any prior CLI break-glass change.
10. For team state, prefer the native OCI Object Storage backend when the reviewed runtime supports it; treat the legacy S3-compatible backend as a deprecated fallback and never place credentials in backend HCL.

Read terraform-authoring.md for command contracts, provider-schema grounding, discovery, packaging, and state rules.

Common multi-step flows

Request	Sequence
Write Terraform	intake → scaffold → provider schema/docs → HCL + tests → fmt/init/validate
Discover existing resources	preflight → empty destination → discover → review generated HCL → remove secrets → choose import/ownership
Deploy locally	preflight → validate → plan → inspect summary → exact-plan apply → service verification
Deploy with Resource Manager	author here → package source + schema.yaml → hand off stack/job operations to oci-resource-manager
Adopt existing resources	prove inventory/ownership → author resource and declarative import blocks → reviewed plan/import → verify zero unintended change
Refactor addresses	retain moved blocks → plan for no replacement → review consumers → exact-plan apply → verify state addresses
Investigate drift	refresh-backed plan → classify change → update HCL or restore through owner → reviewed plan → verify convergence
Configure team state	prefer native OCI backend → verify auth/protection/concurrency → migrate once → test recovery without exposing state
Destroy	refreshed destroy plan → dependency review → exact destructive approval → apply reviewed plan → verify absence

Safety boundaries

• Missing context, preflight, live credentials, or approval never blocks offline authoring, tests, validation, review, or documentation; it blocks only the later live operation that requires it.
• Never print or commit state, plan binaries, .terraform/, wallets, private keys, or terraform.tfvars.
• Reject symlinked or non-empty discovery destinations.
• A plan review sidecar is context- and content-bound; changed bytes require a new review.
• Resource discovery is a starting point, not migration proof.
• State can contain secrets even when values are marked sensitive. Encrypt and access-control it, serialize writers, and never log or attach it.
• Match authentication to the execution context: named local profile or short-lived token, instance/resource principal, OKE workload identity, or Resource Manager. Never embed key material in provider/backend configuration.
• Pin Terraform, provider, and module constraints; commit reviewed lock metadata; review module provenance, checksums, licenses, nested dependencies, provisioners, public exposure, and upgrade/migration notes.
• Resolve realm/region endpoints from the installed provider and current official docs. Do not hardcode commercial-realm domains; use the applicable FIPS-compatible provider where Oracle requires it.
• CLI may inspect or recover. A CLI mutation against a Terraform-owned resource is break-glass and must be followed by terraform plan reconciliation.

Expected output

Report artifact paths, source/schema used, validation result, plan action counts, approval state, verification, and rollback. Never reproduce state or sensitive planned values.

Official documentation

OCI Terraform provider · Resource discovery · Terraform on OCI. Full list in the terraform-authoring reference.

Open Knowledge Format grounding - every doc link here is registered and liveness-checked in the oracle-docs.md index (the pack's single source of truth). When extending this skill, cite the most specific official page through that index; the non-official MCP gateway is never a source of truth.