aibot88 Pulumi Specialist
Structured investigation for Pulumi infrastructure codebases. Five phases: gather context, diagnose, design, recommend, verify.
Arguments
$0— stack name, project path, or problem description. Required.
Phase 1: Context Gathering
- Identify the Pulumi project:
cat Pulumi.yaml pulumi stack ls - Inspect the active stack config:
pulumi config --show-secrets pulumi stack output - Glob for Pulumi source files:
find . -name "*.ts" -o -name "*.py" -o -name "*.go" | grep -v node_modules | sort - Check for Automation API usage:
grep -r "LocalWorkspace\|RemoteWorkspace\|createStack\|selectStack" . \ --include="*.ts" --include="*.py" --include="*.go" -l
Phase 2: Diagnosis
Stack health:
pulumi preview --diff
pulumi stack --show-ids
Resource graph:
pulumi stack graph --dependency-graph /tmp/graph.dot
dot -Tsvg /tmp/graph.dot -o /tmp/graph.svg # if graphviz available
Secret exposure check:
pulumi config --show-secrets | grep -i "key\|secret\|password\|token"
pulumi stack output --show-secrets
Test coverage:
find . -name "*.test.ts" -o -name "*_test.go" -o -name "test_*.py" | sort
Phase 3: Design / Root-Cause Analysis
Map symptoms to causes:
| Symptom | Common Causes | Check |
|---|---|---|
| Unexpected resource replace | Input property changed that triggers replacement | pulumi preview --diff — look for [replace] |
| Stack output leaks secret | Output not marked secret: true |
pulumi stack output --show-secrets |
| ComponentResource missing output | Output not registered in registerOutputs |
Check this.registerOutputs({...}) call |
| Automation API stack timeout | Long provisioning, no timeout configured | Add OnEvent handler + timeout options |
| ESC env not loaded | esc env open not called; ESC not linked to stack |
pulumi config env ls |
Cite file:line for every finding.
Phase 4: Recommendations
Output findings in priority order:
[CRITICAL] <title>
Resource: <logical name or file:line>
Issue: <one sentence>
Evidence: <preview output or code snippet>
Fix: <specific change, with code diff>
Trade-off: <alternative and its downside, if meaningful>
Order: CRITICAL → WARNING → INFO.
Phase 5: Verification
After fixes are applied:
pulumi preview— zero unexpected changes; no[replace]for unintended resources.pulumi stack output— no secrets in plaintext output.- Unit tests pass:
npm test/go test ./.../pytest. - For Automation API changes: run the program with
--previewmode and verify event callbacks fire. - For ESC changes:
esc env open <org>/<project>/<env>and confirm values resolve correctly.
Reference Docs
| File | When to use |
|---|---|
stack-design.md |
Stack topology, StackReference, multi-stack patterns |
component-resources.md |
ComponentResource design, inputs/outputs, lifecycle |
automation-api.md |
Embedding pulumi up/destroy in Node/Python/Go |
secrets.md |
Config secrets, ESC, output sensitivity, encryption |
testing.md |
Unit mocks, integration tests, Automation API test harness |