Atmos Terraform

Atmos Terraform and OpenTofu Orchestration

Atmos wraps the Terraform or OpenTofu CLI to provide stack-aware orchestration of infrastructure operations. Instead of manually managing workspaces, backends, variable files, and authentication for each component, Atmos resolves the full configuration from stack manifests and handles all of these concerns automatically.

Everything in this skill applies identically to Terraform and OpenTofu. The atmos terraform command namespace is the same regardless of which binary is configured -- atmos terraform plan runs tofu plan when the binary is set to tofu. Use the user's terminology in responses (if they say "OpenTofu," say "OpenTofu").

Terraform or OpenTofu: Binary Selection

Atmos defaults to the terraform binary. Switch to OpenTofu by setting components.terraform.command: tofu in atmos.yaml. The setting cascades through CLI > env > config > defaults precedence and can be overridden at multiple levels for mixed setups.

Global (whole project on OpenTofu)

# atmos.yaml
components:
  terraform:
    command: tofu                # All atmos terraform commands invoke `tofu` instead of `terraform`
    base_path: components/terraform

Or via environment variable:

export ATMOS_COMPONENTS_TERRAFORM_COMMAND=tofu
atmos terraform plan vpc -s dev   # Runs: tofu plan

Per-stack override

Use terraform.overrides.command in a stack manifest to switch the binary for everything in that stack:

# stacks/orgs/acme/plat/prod/_defaults.yaml
terraform:
  overrides:
    command: tofu

Per-component override

Set command on an individual component to run a single component on a different binary than the rest of the stack (useful for legacy components that haven't been validated on OpenTofu, or new components testing OpenTofu-specific features):

components:
  terraform:
    legacy-vpc:
      command: terraform           # This component stays on Terraform
      vars: ...
    new-eks:
      command: tofu                # This component runs on OpenTofu
      vars: ...

Per-invocation override

Pass --terraform-command on the CLI to override for a single command:

atmos terraform plan vpc -s dev --terraform-command=tofu

Installing and Pinning the Binary via Toolchain

The Atmos toolchain installs and pins both Terraform and OpenTofu so the same binary version runs on every developer machine and in CI. Binary selection (command: terraform vs command: tofu) and binary version pinning (dependencies.tools.terraform vs dependencies.tools.opentofu) are independent settings -- command says which binary Atmos invokes, dependencies.tools.<name> says which version the toolchain installs. Always pin both together; setting command: tofu without a matching opentofu dependency leaves you at the mercy of whatever tofu is on PATH.

Pinning can be applied at four scopes, in increasing precedence:

1. Project-wide -- .tool-versions (asdf-compatible) at the repo root for defaults everyone shares.
2. Stack-wide -- dependencies.tools in a stack default to pin a whole stack scope.
3. Component-type -- terraform.dependencies.tools to default every Terraform/OpenTofu component.
4. Per-component -- dependencies.tools on an individual component for migrations or one-offs.

Atmos installs declared tools when running the component. Use atmos toolchain install only to pre-warm a cache, bootstrap a shell, or troubleshoot a specific binary. For full YAML examples, resolution order, ad-hoc installs, and the inspection workflow, see references/toolchain-pinning.md and the atmos-toolchain skill.

OpenTofu-specific considerations

• State encryption (OpenTofu 1.7+) -- a Terraform-incompatible feature; if enabled, state can no longer be read by terraform. Switching back is not zero-effort.
• removed blocks -- supported by both binaries in recent versions; no Atmos-side difference.
• Provider/module registry -- OpenTofu uses registry.opentofu.org by default; pinned modules sourced from registry.terraform.io still work in OpenTofu but consult the relevant registry availability when the user reports a missing module.
• terraform.required_version -- OpenTofu respects this constraint; pin appropriately.
• terraform { backend ... } block -- identical syntax in both binaries; no Atmos-side change needed.

For users running mixed Terraform/OpenTofu deployments, the per-component or per-stack override pattern is the right tool. Do not propose a project-wide switch unless the user has validated all components against OpenTofu.

How Atmos Orchestrates Terraform

When you run any atmos terraform command, Atmos performs the following sequence:

1. Resolves stack configuration -- Reads and deep-merges all stack manifests to produce the fully resolved configuration for the target component in the target stack.
2. Generates backend configuration -- Writes a backend.tf.json file in the component directory with the correct backend settings (S3 bucket, key, region, etc.) derived from the stack config.
3. Generates variable file -- Writes a terraform.tfvars.json file containing all vars defined for the component in the stack.
4. Provisions backend infrastructure -- If provision.backend.enabled: true, creates the backend storage (e.g., S3 bucket) before Terraform init.
5. Runs terraform init -- Initializes the working directory with the generated backend config, skipping it automatically when nothing relevant changed (init.mode), and adding -reconfigure/-upgrade only when required (init.reconfigure/init.upgrade). Cleans .terraform/environment first.
6. Selects or creates workspace -- Calculates the Terraform workspace name from context variables and selects it (or creates it if it does not exist).
7. Executes the requested command -- Runs terraform plan, apply, destroy, etc. with the generated varfile and any additional flags.

This means a single command like atmos terraform plan vpc -s plat-ue2-dev replaces what would normally require multiple manual steps: configuring the backend, writing tfvars, running init, selecting the workspace, and then running plan.

Core Commands

plan

Generates a Terraform execution plan showing what changes would be made.

atmos terraform plan <component> -s <stack>

By default, Atmos saves the plan to a file using the naming convention <context>-<component>.planfile. This planfile can later be used with --from-plan to apply the exact reviewed changes.

# Basic plan
atmos terraform plan vpc -s plat-ue2-dev

# Skip planfile generation (useful for Terraform Cloud)
atmos terraform plan vpc -s dev --skip-planfile

# Plan with custom output path
atmos terraform plan vpc -s dev -out=/tmp/my-plan.tfplan

# Plan only specific resources
atmos terraform plan vpc -s dev -target=aws_subnet.private

apply

Applies Terraform changes. Supports interactive approval, planfile-based apply, and auto-approve.

atmos terraform apply <component> -s <stack>

# Interactive apply (prompts for confirmation)
atmos terraform apply vpc -s plat-ue2-dev

# Apply from a previously generated plan
atmos terraform plan vpc -s dev
atmos terraform apply vpc -s dev --from-plan

# Apply a specific planfile
atmos terraform apply vpc -s dev --planfile /tmp/my-plan.tfplan

# Auto-approved apply (no confirmation prompt)
atmos terraform apply vpc -s dev -auto-approve

deploy

Combines plan and apply with automatic approval. This is the most common command for CI/CD pipelines.

atmos terraform deploy <component> -s <stack>

Key differences from apply:

• Automatically sets -auto-approve -- no interactive confirmation
• Supports --deploy-run-init to control whether init runs
• Designed for automated, non-interactive deployments

# Deploy a component
atmos terraform deploy vpc -s plat-ue2-dev

# Deploy from a previously generated plan
atmos terraform deploy vpc -s dev --from-plan

# Deploy a specific planfile
atmos terraform deploy vpc -s dev --planfile /tmp/vpc-plan.tfplan

destroy

Destroys all resources managed by a component in a stack. Accepts -auto-approve and -target=... just like upstream Terraform: atmos terraform destroy vpc -s dev.

init

Atmos runs terraform init automatically before plan, apply, and deploy, but only when needed (init.mode), adding -reconfigure/-upgrade only when required (init.reconfigure/init.upgrade); --skip-init disables auto-init for one invocation. Upstream init flags pass through when invoked directly (-reconfigure, -upgrade, -migrate-state): atmos terraform init vpc -s dev -reconfigure.

Streaming UI (--ui)

plan, apply, deploy, init, and destroy support a --ui flag that renders a live, Docker-build-style progress view (spinners, a dependency tree, per-resource status) instead of raw Terraform output:

atmos terraform apply vpc -s dev --ui

Enable it by default via components.terraform.ui.enabled: true in atmos.yaml or ATMOS_TERRAFORM_UI=true; pass --ui=false to override a config-enabled default for one run. The UI auto-disables when output is piped, in CI (CI=true), or on unsupported commands. refresh doesn't support it — Terraform's refresh doesn't emit the structured -json output the UI depends on; --ui refresh prints a warning and falls back to standard output.

Multi-Component Operations

Atmos supports executing Terraform commands across multiple components simultaneously using filter flags. These work with plan, apply, and deploy.

# All components in all stacks
atmos terraform plan --all

# All components in a specific stack
atmos terraform plan --stack prod

# Specific components across stacks
atmos terraform deploy --components vpc,eks

# Only components affected by git changes (in dependency order)
atmos terraform deploy --affected

# Affected with dependents included
atmos terraform deploy --affected --include-dependents

# Filter by YQ query expression
atmos terraform plan --query '.vars.tags.team == "eks"'

# Combine filters
atmos terraform plan --affected --stack prod

# Always preview first with --dry-run
atmos terraform deploy --all --dry-run

Workspace Management

Atmos computes the Terraform workspace name from stack context (namespace, tenant, environment, stage, component), runs terraform init -reconfigure (always, regardless of init.reconfigure), and selects (or creates) the workspace on every invocation. Explicit management is also available: atmos terraform workspace vpc -s plat-ue2-dev.

For stable workspace keys across component implementations, use metadata.name on an abstract base component (name: vpc, component: vpc/v2); for dynamic naming, workspace_key_prefix under backend: accepts Go templates. Toggle the runtime behavior via components.terraform.init.reconfigure (supersedes init_run_reconfigure) and workspaces_enabled in atmos.yaml. Full examples are in references/backend-configuration.md.

Backend Configuration and Auto-Generation

With components.terraform.auto_generate_backend_file: true in atmos.yaml, Atmos reads backend_type and backend from the resolved stack config, deep-merges through the stack hierarchy (org → tenant → environment → stage → component), and writes backend.tf.json into the component directory before terraform init. Add backend.tf.json to .gitignore.

Regenerate manually with atmos terraform generate backend vpc -s plat-ue2-dev. For full examples covering S3, GCS, Azure, and remote backends, plus workspace key prefix patterns, see references/backend-configuration.md.

Variable File Generation

Atmos generates terraform.tfvars.json from the vars section in the stack configuration. This happens automatically before plan/apply/deploy, but can also be invoked manually:

atmos terraform generate varfile vpc -s plat-ue2-dev

# Output to a custom file
atmos terraform generate varfile vpc -s plat-ue2-dev -f vars.json

Planfile Generation

Generate planfiles in JSON or YAML format for review or integration with tools like Checkov:

atmos terraform generate planfile vpc -s plat-ue2-dev
atmos terraform generate planfile vpc -s dev --format=json
atmos terraform generate planfile vpc -s dev --format=yaml --file=planfile.yaml

Authentication Configuration

Terraform commands can use Atmos auth identities from atmos.yaml or component-level auth.identity. Keep detailed provider/profile setup in atmos-auth; this skill should only recommend the Terraform runtime controls:

atmos terraform plan vpc -s prod --identity prod-admin
atmos terraform plan vpc -s dev --identity ""

Backend Provisioning

Backend configuration and backend provisioning are different:

• backend_type and backend tell Terraform where state lives and generate backend.tf.json.
• provision.backend.enabled: true tells Atmos to create the backend storage before first use.

Set provision.backend.enabled: true in stack config to auto-provision backend infrastructure, solving the Terraform bootstrap problem. Manual provisioning is available via atmos terraform backend create/list/describe/update/delete. Backend provisioning currently applies to Terraform components. See references/backend-configuration.md for details.

Source Provisioning and Workdirs

Use source for just-in-time component provisioning and pair it with provision.workdir for isolated per-instance execution. Workdirs prevent shared .terraform, lockfile, backend, and varfile collisions when the same component source is used by multiple stacks or runs.

components:
  terraform:
    vpc:
      source:
        uri: github.com/cloudposse-terraform-components/aws-vpc.git
        version: 1.450.0
      provision:
        workdir:
          enabled: true

With workdirs enabled, Atmos stages the provisioned source into the instance workdir and runs Terraform there. Without workdirs, source provisioning targets the component path.

Interactive Shell

The shell command drops you into a shell pre-configured with all the context for a component in a stack. Varfiles, backend config, and workspace are all set up so you can run native Terraform commands directly.

atmos terraform shell vpc -s plat-ue2-dev

Inside the shell:

• The working directory is the component's folder
• terraform.tfvars.json and backend.tf.json are generated
• The correct workspace is selected
• All required ENV vars are set
• ATMOS_SHLVL tracks shell nesting level

Customize the shell prompt in atmos.yaml:

components:
  terraform:
    shell:
      prompt: "atmos [{{.Stack}}] {{.Component}} $ "

Additional Commands

Atmos supports output, validate, state, clean, console, fmt, get, import, show, taint/untaint, force-unlock, refresh, graph, and providers -- all standard Terraform subcommands with the same atmos terraform <cmd> <component> -s <stack> syntax. For the complete reference, see references/commands-reference.md.

# Common examples
atmos terraform output vpc -s dev vpc_id
atmos terraform state list vpc -s dev
atmos terraform clean vpc -s dev

Common Flags

Use -- to pass flags directly to Terraform: atmos terraform plan vpc -s dev -- -refresh=false. For a default that should apply on every run instead of being retyped, declare it under components.terraform.flags (or a stack/component-level flags: block) — see Configuration in atmos.yaml below. For the complete flag reference, see references/commands-reference.md.

Path-Based Component Resolution

You can use filesystem paths instead of component names:

cd components/terraform/vpc
atmos terraform plan . -s dev
atmos terraform apply . -s dev

Supported path formats: ., ./component, ../sibling, /absolute/path. If a path matches multiple components, Atmos prompts for selection in interactive mode.

Debugging

• atmos describe component vpc -s plat-ue2-dev -- show the fully resolved configuration (merged vars, backend, workspace name, metadata, settings).
• atmos terraform plan vpc -s dev --dry-run -- preview what Atmos will do without executing.
• TF_LOG=DEBUG atmos terraform plan vpc -s dev -- enable upstream Terraform debug logging.

Configuration in atmos.yaml

Key settings under components.terraform include auto_generate_backend_file, init.mode, init.reconfigure, init.upgrade (see init above), workspaces_enabled, deploy_run_init, apply_auto_approve, and plan.skip_planfile. Each has a matching ATMOS_COMPONENTS_TERRAFORM_* environment variable override. See references/backend-configuration.md for complete configuration details.

components.terraform.flags sets fleet-wide defaults for terraform CLI execution flags (lock_timeout, lock, parallelism, refresh, compact_warnings) — e.g. lock_timeout: "5m" so concurrent runs retry a held state lock instead of failing on Terraform's 0s default. The same flags: block can be set at the stack level (root-level terraform: block) and per component, each overriding the layer below field-by-field. See Flags.

Best Practices

1. Use the two-stage plan/apply workflow for production. Run plan first, review the output, then apply --from-plan to ensure exactly the reviewed changes are applied.

2. Use deploy for automated pipelines. It combines plan and apply with auto-approve, ideal for CI/CD.

3. Always preview multi-component operations with --dry-run before executing --all or --affected.

4. Let Atmos manage backend configuration. Set auto_generate_backend_file: true and define backend settings in stack manifests rather than hardcoding in Terraform modules.

5. Use atmos describe component to debug configuration resolution issues. It shows the fully merged result of all stack manifest inheritance.

6. Add generated files to .gitignore. The backend.tf.json and terraform.tfvars.json files are generated at runtime and should not be committed.

7. Use atmos terraform shell for interactive debugging. It sets up the full context so you can run native terraform commands directly.

8. Enable backend provisioning (provision.backend.enabled: true) to solve the Terraform bootstrap problem and ensure backends exist before first use.

9. Use source provisioning with workdirs when components are pulled via source, especially in CI or any multi-stack workflow that can run concurrently.

Additional Resources

• For the complete list of all atmos terraform subcommands, see references/commands-reference.md
• For backend configuration patterns (S3, GCS, Azure, remote), see references/backend-configuration.md
• For toolchain-based Terraform/OpenTofu version pinning at each scope, see references/toolchain-pinning.md