AWS Native Runtime Investigation

AWS Native Runtime Investigation

Use this skill to find the relevant AWS Native lifecycle boundary and keep the

change small. It is a routing aid, not a required planning framework or an

edge-case checklist.

Start With The Repository Guide

Read docs/provider-runtime.md, following

only the Pulumi engine and CloudControl references relevant to the reported

behavior. Read ARCHITECTURE.md for repository

boundaries and AGENTS.md for generated-file and command

rules.

Trace the motivating issue, repro, or failing test through the current code. Do

not ask the user to explain facts that can be read from the repository.

Keep The Change Narrow

• Identify the concrete resource and lifecycle transition that is wrong.
• Name the boundary that owns it: Pulumi engine behavior, AWS Native

translation/runtime behavior, CloudFormation schema metadata, or

CloudControl behavior.

• Propose the smallest end-to-end change that fixes the reported path.
• Treat adjacent metadata and lifecycle cases as context to disposition, not

requirements to implement.

• Preserve existing behavior when stronger semantics cannot be established

safely.

• Before adding machinery for an uncommon case, check whether it occurs in the

current schemas, tests, or reported user behavior. Report its incidence,

fallback, and follow-up option when support is costly.

• Broaden across Read, Diff, or patch generation only when those boundaries

must agree for the motivating behavior to be correct.

• Prefer a focused test that proves the user-visible regression. Add lower-level

tests when they provide distinct proof or useful failure localization rather

than duplicating the lifecycle test.

Do not create a design document merely because this skill was loaded. Use the

conversation for narrow changes. Create a durable plan only when the user asks

for one or when unresolved cross-boundary decisions need to survive a handoff.

Stop And Ask

Return to the maintainer with concrete options when:

• Pulumi, AWS Native, and CloudControl evidence point to different owners;
• two plausible semantics have different user-visible results;
• the safe fallback is unclear;
• the implementation is becoming substantially broader than the reported

problem; or

• a rare case requires disproportionate complexity and cannot be deferred.

Follow AGENTS.md for validation and generated artifacts. Do not hand-edit

generated SDKs, generated schema or metadata, CI workflows, toolchain files, or

upstream submodule content.