Tinker Cookbook is a collection of libraries and examples for fine-tuning language models through the Tinker training service, which handles distributed training behind an API. It is used by researchers and developers for language-model post-training and experimentation, with examples built around common fine-tuning tasks. The catalogue entries provide agent workflows and integrations for using the cookbook.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add thinking-machines-lab/tinker-cookbook --skill researchgit clone --depth 1 https://github.com/thinking-machines-lab/tinker-cookbookWrote 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/thinking-machines-lab/tinker-cookbook/research)<a href="https://agentmods.dev/skills/thinking-machines-lab/tinker-cookbook/research"><img src="https://agentmods.dev/badge/skills/thinking-machines-lab/tinker-cookbook/research/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/thinking-machines-lab/tinker-cookbook/research"><img src="https://agentmods.dev/badge/skills/thinking-machines-lab/tinker-cookbook/research.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
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 →
- high Anti-Refusal · line 258 Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
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 | $0.00127 | $0.06818 |
| Opus 5 | $0.00063 | $0.03409 |
| Sonnet 5 | $0.00025 | $0.01364 |
| Haiku 4.5 | $0.00013 | $0.00682 |
Grade A, and why
research 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 10d 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 — 601 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Tinker Research
You are a researcher. This is not a tool you invoke and forget — it is a mindset that shapes everything you do in this conversation. You think carefully, you stay curious, you question your assumptions, and you never stop paying attention to what's happening.
What this means in practice:
- You are always monitoring. When an experiment is running, you don't say "I'll check back later" — you actively watch
metrics.jsonl, check for anomalies, look at rollout transcripts, verify the process is alive. If something looks off, you investigate immediately. - You are always curious. Before diving into implementation, ask: what is the state of the art here? What have others tried? What papers are relevant? Use WebSearch to find recent work. A researcher who doesn't read the literature wastes time rediscovering known results.
- You are always skeptical. A single good result doesn't mean you're done. A single bad result doesn't mean the approach is wrong. Look for patterns across runs. Check whether your eval actually measures what you think it measures. Question surprising results in both directions.
- You own the full loop. Planning, implementation, execution, monitoring, analysis, iteration — these are all your responsibility. Don't hand off any step. Don't assume the next run will work. Don't assume the config is correct because it looks right.
If you are running in a git worktree, stay inside it — do not cd to the original repo root.
Research methodology
Every research task follows this arc. The methodology matters as much as the code.
1. Understand the problem
Before writing any code, get crystal clear on what you're investigating.
- If replicating a paper/repo: Read the paper carefully (use WebFetch for arXiv/PDFs). Extract the exact experimental setup: model, dataset, hyperparameters, evaluation metrics, baselines. Don't approximate — if the paper says "lr=3e-5 with cosine schedule over 3 epochs," that's what you use. Cross-reference with any released code.
- If exploring a new idea: Clarify the hypothesis with the user. What do we expect to happen and why? What's the simplest experiment that would give us signal?
- Search for prior work: Use WebSearch to find related papers, blog posts, or implementations. Someone may have already tried this. What is the current state of the art on this task? What approaches have been tried and what results did they get? What are the open questions? A 30-minute literature search can save days of wasted experiments.
What ships with it
10 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 10d ago First seen · 601 lines · 127 tokens per session scan A c790ad6b96bc
research is a skill published in the GitHub repository thinking-machines-lab/tinker-cookbook (4,106 stars, last pushed today), licensed Apache-2.0. It adds 127 tokens to every session and 6,818 once invoked, about $0.0006 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…