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/babysit-zephyr/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/babysit-zephyr)<a href="https://agentmods.dev/skills/marin-community/marin/babysit-zephyr"><img src="https://agentmods.dev/badge/skills/marin-community/marin/babysit-zephyr/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/babysit-zephyr"><img src="https://agentmods.dev/badge/skills/marin-community/marin/babysit-zephyr.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Output Handling · line 64 Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.Fix: Set explicit limits on output length, generation count, and rate. Use max_tokens and truncation to prevent unbounded output.
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 | $0.00031 | $0.01044 |
| Opus 5 | $0.00015 | $0.00522 |
| Sonnet 5 | $0.00006 | $0.00209 |
| Haiku 4.5 | $0.00003 | $0.00104 |
Grade A, and why
babysit-zephyr 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 11d 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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Babysit a Zephyr job
Zephyr Job Structure
A job has a one-task *-coord coordinator and a *-workers pool. Sequential
pipelines produce different p<N> child names; retries produce different hashes
with the same p0. Child names follow
<hash>-p<pipeline>-a<attempt>-{coord,workers}. Old coordinators may linger.
Iris Config
Resolve the requested cluster to its file under lib/iris/config/; substitute
that file for <CONFIG> below. See lib/zephyr/OPS.md for the dashboard and
coordinator query reference.
Starting a Job
Get the run command from the user:
uv run iris --config <CONFIG> job run --region <REGION> --no-wait -- python <SCRIPT>
The entrypoint container defaults to 1GB memory. For long-running pipelines that accumulate state (GCS clients, logging), increase with --memory:
uv run iris --config <CONFIG> job run --region <REGION> --memory 5GB --no-wait -- python <SCRIPT>
The command prints a job ID on success. Note it for monitoring.
Stopping a Job
Always ask the user before stopping. Stopping kills all child jobs (coordinators, workers).
uv run iris --config <CONFIG> job cancel <JOB_ID>
Monitoring
Health Checks
Check child job states via the Iris CLI (returns per-task state and resourceUsage):
# diskMb is updated every ~60s. On K8s it is always 0 (workdir lives inside the pod).
uv run iris --config <CONFIG> rpc controller list-tasks --job-id <JOB_ID>
A healthy zephyr job has:
- Coordinator: RUNNING, 1 task running
- Workers: RUNNING, tasks ramping up toward target count
Stage Progress
The coordinator logs stage, completed, in-flight, queued, and worker counts:
uv run iris --config <CONFIG> rpc controller get-task-logs \
--id <COORD_JOB_ID> --max-total-lines 5000 --attempt-id -1 --tail
Large pools can flood the log with pull_task, Started operation,
report_result, registered, and tasks completed; filter those entries.
Coordinator Thread Dump
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.
- 11d ago First seen · 106 lines · 31 tokens per session scan A 6c01ea77cac5
babysit-zephyr is a skill published in the GitHub repository marin-community/marin (3,593 stars, last pushed today), licensed Apache-2.0. It adds 31 tokens to every session and 1,044 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
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
local-ai-agents
Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…
next-cache-components-adoption
Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…
insight-error-page
Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…
next-cache-components-optimizer
Drive a Next.js route to instant navigation by setting up an agentic loop, under Cache Components / PPR, on initial load (hard navigation) and client-side navigation (soft navigation). Encode the goal as a failing @next/playwright instant() e2e and work it to green, one verified route at a time; the shipped test then…
next-partial-prefetching-adoption
Turn on Partial Prefetching in a Next.js app and work through the insights it surfaces. Use when the user wants to enable or adopt Partial Prefetching, flip the partialPrefetching flag, opt routes in with export const prefetch = 'partial', audit Link prefetch={true} behavior, preserve existing prefetched UI with…