Review a Marin access-change pull request by decrypting user identities and explaining their roles and resources. After explicit human confirmation, merge the request and apply only the verified permission changes through Pulumi.
中文简介正在准备中。当前显示英语简介。
Review an explicitly identified marin-iac grant PR that edits IAM data or a deploy-target module, confirm its decrypted principals and roles, then apply only the confirmed grant.
安装前可以查看交给 AI 代理的指令内容。
A grant PR (usually from the add-grant skill) uses opaque human-NNN
references whose emails are KMS ciphertext in iam_data.yaml. This skill
reveals the real grant, gets an explicit human confirmation, then lands it and
applies it.
Read first:
infra/pulumi/README.md — the marin-iac stacks and the pulumi up
prerequisites (you need roles/cloudkms.cryptoKeyEncrypterDecrypter on the key
and permission to update the marin stack).Never approve or merge before the user confirms the decrypted grant. The whole point is that a second person sees the actual identity and access before it is applied.
gh pr view <n> --repo marin-community/marin --json title,body,headRefName,files,url
gh pr checkout <n> # pull the branch into the worktree
git fetch origin main
Confirm the diff only touches grant surfaces under infra/pulumi/src/iac/gcp/:
iam_data.yaml and/or a deploy-target IAM module. If it
changes anything else (code, other Pulumi resources), stop and review it as an
ordinary PR, not a grant.
Turn the changed opaque principal references into real emails:
git diff origin/main...HEAD -- infra/pulumi/src/iac/gcp \
| uv run --package marin-iac --extra deploy \
python infra/pulumi/iam_principal.py decrypt --diff
Each output line is + user:<email> (added) or - user:<email> (removed). Map
each back to the role and resource it sits under in the diff. The decryptor
shows the principal; read the surrounding role and container
(project_grants, a specific bucket/secret/repository/service account, or a
deploy-target module) from the diff hunk.
Print a plain-language summary, one line per grant, and ask the user to confirm. For example:
PR #1234 grants:
+ alice@openathena.ai → roles/storage.objectViewer on project hai-gcp-models
+ alice@openathena.ai → IAP viewer on evaldash.oa.dev
- bob@openathena.ai → roles/bigquery.dataViewer (revoked)
Apply this? (yes/no)
Call out anything that looks off: a broader role than the resource needs, a principal you do not recognize, a domain wildcard, or a revocation that might cut off active access. If the user does not clearly approve, stop and report back — do not merge.
After the user confirms:
gh pr review <n> --repo marin-community/marin --approve
gh pr merge <n> --repo marin-community/marin --squash
git checkout main && git pull origin main # land the merged change locally
Wait for the merge to land on main and pull it before applying, so pulumi up
runs against the committed state.
The change is not live until pulumi up runs — CI never applies. Identify the
affected stack(s) from the diff:
infra/pulumi/src/iac/gcp/ → the marin stack in
infra/pulumi.Prompt the user: they can run it themselves, or ask you to. If you run it, per stack:
cd infra/<dir>
pulumi stack select <stack>
pulumi preview # confirm ONLY the intended grant is added/removed, no other drift
pulumi up
Read the preview before applying. For the marin stack, expect only the
IAMMember create/delete for this grant — any NodePool or other unexpected
replace/delete means stop and reconcile, exactly as the pulumi README
warns. Once up is clean, tell the user the grant is live.
Comment on the merged PR that pulumi up ran and the grant is live, so the
requester and any watcher see the change reached production, not just main:
gh pr comment <n> --repo marin-community/marin \
--body "🤖 \`pulumi up\` on the \`<stack>\` stack succeeded — the grant is live."
An agent comment must begin with 🤖 (see AGENTS.md). If pulumi up did not run
(the user is applying it themselves), skip this and let them confirm instead.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
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.
中文简介正在准备中。当前显示英语简介。
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.
中文简介正在准备中。当前显示英语简介。
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.
中文简介正在准备中。当前显示英语简介。
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.
中文简介正在准备中。当前显示英语简介。
Try model or training changes in a copied Grug or Grugformer experiment, document the differences, and bring successful changes back into the base template while removing superseded experiment code and checking behavior.
中文简介正在准备中。当前显示英语简介。
Organize Marin experiment reports into a maintained index, or archive explicitly named older experiments using lasting tags and issue references. Existing conclusions and canonical links are preserved so experiment records remain traceable.
中文简介正在准备中。当前显示英语简介。