Preview pending infrastructure changes in a specified Marin Pulumi stack without applying them. Trace each change to supporting code and merged pull requests when deployment history provides a reliable Git baseline.
Run a read-only preview for a specified Marin infra/pulumi stack and trace each pending resource change to merged pull requests since its latest successful update when that update records a clean Git checkout. Use when a stack has unapplied changes or drift and someone needs to know which PRs explain the preview.
Review the instructions the agent will receive before you install.
Work only with the marin-iac project in infra/pulumi. Require an explicit stack name.
Read infra/pulumi/README.md and satisfy its credentials and stack-specific prerequisites. Fetch origin/main and run from a clean checkout at that commit. Preserve existing work; use another worktree when necessary.
Prepare the environment from the repository root:
uv sync --package marin-iac --extra deploy --frozen
export PULUMI_PYTHON_CMD="$PWD/.venv/bin/python"
Run the preview without changing the selected stack:
pulumi -C infra/pulumi preview --stack <stack> --diff --color never
Keep the raw preview local. It may contain decrypted identifiers. If the preview is empty, report that the stack has no pending diff and stop.
Read recent history without secrets:
pulumi -C infra/pulumi stack history \
--stack <stack> --json --page-size 100 > /tmp/marin-pulumi-history-<stack>.json
jq -r '
map(select(.kind == "update" and .result == "succeeded"))
| first
| [.startTime, .environment["git.head"], .environment["git.dirty"]]
| @tsv
' /tmp/marin-pulumi-history-<stack>.json
Use git.head only when it is a commit on origin/main and git.dirty is false. If either condition fails, report that an exact Git baseline is unavailable and stop attribution.
List merged commits after the deployed commit:
git log --first-parent --format='%H%x09%s' <deployed-sha>..origin/main
Inspect each candidate with git show. Match concrete preview resource names and changed properties to the code or configuration diff. Do not attribute a change merely because its commit is in the range.
Map a matching commit to its pull request:
gh api repos/marin-community/marin/commits/<commit>/pulls \
--jq '.[] | [.number, .title, .html_url] | @tsv'
Report each preview change with its matching PR and the file or hunk that explains it. Report changes with no matching commit as live drift or unresolved attribution. A preview can contain changes from multiple PRs.
Never run pulumi up, pulumi refresh, pulumi destroy, an import, or a state mutation. Call out any NodePool replacement or deletion from the preview.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Compare Zephyr changes using the same prepared dataset and matching execution settings. Analyze Finelog metrics for each processing stage to distinguish CPU efficiency, elapsed time, and peak memory changes, then report improvements, regressions, or tradeoffs.
Register Hugging Face datasets for Marin by inspecting their structure without downloading the full dataset. Identify text fields, configurations, and data splits, then add a dataset module and validate the mapping with a sample.
Turn a specified GCP access request into a reviewable Pulumi change and pull request. Select the narrowest suitable role and encrypt personal email identities, covering resource permissions and Cloud Run viewer access through IAP, an access gateway.
Add or update Pallas/Mosaic kernels, specialized computation routines for JAX. The skill covers reference implementations, output and gradient checks, public wrappers, fallback behavior, performance benchmarks, and hardware-specific tuning when requested.
Preserve old experiment scripts with a dated Git tag, a named snapshot of the code, and add comments linking related GitHub issues to the archived files and cleanup pull request.
Launch and continuously monitor a specified Zephyr pipeline on Iris when asked. It checks coordinator and worker health, tracks stage progress, identifies stalled or repeated failures, and requires explicit approval before stopping or restarting jobs.