Marin is an open-source research program, software platform, and community for developing foundation models such as large language models. Researchers use it for data preparation, tokenization, pretraining, posttraining, evaluation, and related experiments, including work on audio-text, DNA, and protein models. The catalogue entries are add-ons that support workflows around Marin.
Borrowing it
Nothing to install: this file belongs to marin-community/marin. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/marin-community/marin/main/.agents/skills/review-grant/SKILL.mdgit clone --depth 1 https://github.com/marin-community/marinWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/marin-community/marin/review-grant)<a href="https://agentmods.dev/skills/marin-community/marin/review-grant"><img src="https://agentmods.dev/badge/skills/marin-community/marin/review-grant/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/marin-community/marin/review-grant"><img src="https://agentmods.dev/badge/skills/marin-community/marin/review-grant.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00037 | $0.01087 |
| Opus 5 | $0.00018 | $0.00544 |
| Sonnet 5 | $0.00007 | $0.00217 |
| Haiku 4.5 | $0.00004 | $0.00109 |
Grade A, and why
review-grant scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 118 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill: Review a user grant
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 thepulumi upprerequisites (you needroles/cloudkms.cryptoKeyEncrypterDecrypteron the key and permission to update themarinstack).
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.
1. Fetch the PR
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.
2. Decrypt the changed principals
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.
3. Present the grant and get confirmation
Print a plain-language summary, one line per grant, and ask the user to confirm. For example:
PR #1234 grants:
+ [email protected] → roles/storage.objectViewer on project hai-gcp-models
+ [email protected] → IAP viewer on evaldash.oa.dev
- [email protected] → roles/bigquery.dataViewer (revoked)
Apply this? (yes/no)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 118 lines · 37 tokens per session scan A 88a5bb7e5d1b
review-grant is a skill published in the GitHub repository marin-community/marin (3,607 stars, last pushed today), licensed Apache-2.0. It adds 37 tokens to every session and 1,087 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
gke-compute-classes
Configures, optimizes, and troubleshoots GKE ComputeClasses. Use when configuring Spot VMs with on-demand fallback, targeting specific accelerators (GPUs/TPUs) or machine families, restricting ComputeClass access, or debugging pending pods related to node pool auto-creation. Do not use for cluster-level Node Auto…
gke-reliability
Improves GKE workload reliability, using PDBs, health probes, and topology spread constraints. Use when configuring GKE workload reliability, setting up PDBs, or configuring GKE health probes (liveness, readiness, startup). Don't use for disaster recovery setup or full cluster backups (use gke-backup-dr instead).
gke-workload-security
Audits, configures, and hardens workload-level security controls for Google Kubernetes Engine (GKE) applications and namespaces. Covers running cluster security audits (auditcluster.sh), configuring Workload Identity Federation (impersonation, KSA/GSA binding, and pod setup), enforcing Network Policies (default-deny…
azure-mgmt-botservice-dotnet
Azure Resource Manager SDK for Bot Service in .NET. Management plane operations for creating and managing Azure Bot resources, channels (Teams, DirectLine, Slack), and connection settings. Triggers: "Bot Service", "BotResource", "Azure Bot", "DirectLine channel", "Teams channel", "bot management .NET", "create bot".
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.
cloud-architect
Designs cloud architectures, creates migration plans, generates cost optimization recommendations, and produces disaster recovery strategies across AWS, Azure, and GCP. Use when designing cloud architectures, planning migrations, or optimizing multi-cloud deployments. Invoke for Well-Architected Framework, cost…