---

name: pulumi description: Manages Pulumi stacks and deployments via native pulumi_* tools and optional Pulumi MCP. Use for stack inspect, preview, up, refresh, and destroy with confirmed org/project/stack targets.

Pulumi

Purpose

Operate Pulumi infrastructure: list and inspect stacks, preview changes, run up/refresh/destroy, and use MCP tools when routed—always confirming organization/project/stack and integration token before mutating state.

When to Use

• List stacks or read stack outputs, resources, history, deployments.
• Preview, update (up), refresh, or destroy a named stack.
• Pulumi MCP workflows when MCP is the selected route (list_tools).

When NOT to Use

• Authoring new IaC patterns from scratch → pulumi-best-practices, author-iac, or migration skills.
• Provider version bumps → pulumi-upgrade-provider.
• Unconfirmed destroy on production stacks.

Expected Outcome

• Target stack triple confirmed before preview/up/destroy.
• Preview summarized from tool output before apply when user asked to preview first.
• Destructive ops only after explicit user confirmation when appropriate.

Inputs to Gather

• organization / project / stack (or equivalent from tool schema).
• Whether route is native pulumi_* or Pulumi MCP (self-hosted default unless configured otherwise).
• Integration token / connection status before writes.

Workflow

1. Verify Pulumi integration and token when writes are needed.
2. Confirm organization/project/stack with the user for mutating commands.
3. Inspect: pulumi_list_stacks, pulumi_get_stack, outputs, resources, history as needed.
4. Plan: pulumi_preview before pulumi_up when the user wants to see changes.
5. Apply: pulumi_up or pulumi_refresh; destroy only with confirmation.
6. For MCP route: discover tools via list_tools; follow live schemas.

Domain rules

1. Native vs MCP: pulumi_* for curated control plane; MCP when platform routes to MCP.
2. Self-hosted MCP default unless hosted override is configured.
3. Token and target checks before preview, up, refresh, destroy.
4. Confirm destructive actions (especially destroy).

Main tools

• Native: pulumi_list_stacks, pulumi_get_stack, pulumi_get_stack_outputs, pulumi_get_stack_resources, pulumi_get_stack_history, pulumi_list_deployments, pulumi_get_deployment, pulumi_preview, pulumi_refresh, pulumi_up, pulumi_destroy
• MCP: tools from list_tools when MCP is selected

Examples

List stacks: pulumi_list_stacks (or MCP equivalent); return names and state.

Preview dev: Confirm target stack, run preview, summarize resource changes from response.

Tool Availability Rules

Access	Behavior
Full tool access	Inspect and mutate per user confirmation.
Read-only	List/get/preview only.
No token/integration	Do not run up/destroy; report blocker.

Related tool sets

• pulumi

Review / Decision / Execution Criteria

• Never destroy without clear user intent.
• Summarize preview diffs in plain language.

Output Format

1. Stack target and command.
2. Preview summary or apply result.
3. Errors.
4. Recommended follow-up (refresh, drift check).

Quality Bar

• No fabricated resource counts or URNs.
• Distinguish preview vs applied state.

Safety and Boundaries

• Confirm destroy and production up when impact is unclear.
• No secrets in stack config in chat.

Escalation / Dispatch Rules

• IaC authoring/migrations → related pulumi-* skills.
• MCP schema missing → stop and report.

References

• Legacy: skills/old_skills.json (pulumi).
• skills/skill.instruction.md, skills/meta.instructions.md