The meaning of various cdk assets and implications of deleting them

[ustulation]

I was wondering if someone could let me know of the implications of getting rid of various "types" of assets in cdk assets directory.

• For CodePipeline I end up with
  • cdk-asset dir cdk-hnb659fds-assets-<acc-no>-<region>: This mostly has json CFn template files for the pipeline stack itself. My pipeline stack doesn't have anything else like a lambda and so on. I suppose if it had say a Lambda which needed a source code zip then that zip would be here too.
  • Per pipeline pipelines-artifact bucket: Each of these belong to a pipeline and have 2 dirs inside them: one that seems to contain a zipped cdk.out produced by cdk synth each time it executes in the pipeline and another dir which seems to contain zipped result of a git clone of the source repo that the pipeline is listening to (via codestar connection to GitHub in my case) for source code changes.
• For various stages that the pipeline deploys to (different accounts in my case), there's again a cdk-asset bucket per stage. That bucket contains zip files which are source code for lambdas in that stage's stack(s). Similarly there is a cdk-ecr repo that contains images for ECS services.

1. Given all that is it safe to delete all the json templates from cdk-asset dir in the pipeline account above? CFn seems to keep its own copy of the template anyway (in some s3-external.amazonaws.com bucket which i can see from CFn console if I manually create a stack) - so I don't know when would these template jsons be ever needed - even during rollbacks.
2. Is it safe to just get rid of everything inside code-pipelines artifact bucket (which has a zipped cdk.out and a zipped source code from github, per deployment)? When are these needed and what's the drawback of say creating a lifecycle policy to just get rid of all objects > 1 day old in these buckets?
3. For other assets like the zipped source code for lambda and images in ECR, I suppose it's not safe to get rid of them as they are either currently in use or might be needed again during update-rollbacks by CFn. I'm planning to run some code that checks all templates in an account+region and gets rid of all the remaining zip assets and images which have no mention in the template provided there's no CFn stack in in-progress state (whether create-in-progress or roll-back-in-progress etc). If it's in progress then it's not safe to delete anything because I wouldn't know if the template i got by querying CFn was the new one which is in progress or the previous one before the progress.

(3) Above could be much simpler if cdk did a unique prefix (or bucket) per stack. Then I could just delete all the artifacts not referenced by a template, after it has successfully been deployed, by creating a post-deployment action in the pipeline. However since all other unrelated stacks share the same bucket+prefix this becomes impossible to do since some of them might be in some in-progress state or the other.

---

Q) However does (1) and (2) sound reasonable or what are the caveats?