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 gvkhosla/pi-tinker --skill tinker-inklinggit clone --depth 1 https://github.com/gvkhosla/pi-tinkerWrote 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/gvkhosla/pi-tinker/tinker-inkling)<a href="https://agentmods.dev/skills/gvkhosla/pi-tinker/tinker-inkling"><img src="https://agentmods.dev/badge/skills/gvkhosla/pi-tinker/tinker-inkling/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/gvkhosla/pi-tinker/tinker-inkling"><img src="https://agentmods.dev/badge/skills/gvkhosla/pi-tinker/tinker-inkling.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 | $0.00066 | $0.00702 |
| Opus 5 | $0.00033 | $0.00351 |
| Sonnet 5 | $0.00013 | $0.00140 |
| Haiku 4.5 | $0.00007 | $0.00070 |
Grade A, and why
tinker-inkling 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 9d 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 — 57 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Inkling in Pi
Thinking Machines' models for Tinker. Same renderer (tml_v0), tokenizer, and effort interface.
| Id | Size | When |
|---|---|---|
thinkingmachines/Inkling-Small |
276B / 12B active, 64K ctx | Default. Coding, grading, synthetic data, cheaper runs. |
thinkingmachines/Inkling-Small:peft:262144 |
small, 256K ctx | Efficient long-context inference/training. |
thinkingmachines/Inkling |
975B / 41B active, 64K ctx | Only if the user asks for full Inkling. |
thinkingmachines/Inkling:peft:262144 |
full, 256K ctx | Full long-context inference/training. |
Both are post-trained starting points, not base models.
Setup
uv pip install -U tinker-cookbook
Needs Python 3.11+, Tinker SDK 0.23+, torch>=2.10. tml-renderers is included. There is no [inkling] extra.
/tinker inkling
/model
/tinker inkling sweep --prompt "representative task" --efforts low,medium,high,xhigh --yes
Never tokenizer.encode() a chat prompt. Never hardcode the renderer name — use model_info.get_recommended_renderer_name.
Effort
Finite scalar in [0.0, 1.0). Renderer inserts the effort system message. Set it at sampling and when building training data. Default if omitted is 0.9 (high) — make that deliberate.
| none | minimal | low | medium | high | xhigh |
|---|---|---|---|---|---|
| 0.0 | 0.1 | 0.2 | 0.7 | 0.9 | 0.99 |
- Same effort for train data, baseline eval, and checkpoint eval.
- Effort and
max_tokensare independent. High effort can need 16k+. - Eval at
temperature=1.0. Lowering temperature is not a substitute for lowering effort. effort=0.0conditions toward no reasoning; it is not a hard off switch.
Serving
Stay on Tinker. Cookbook has no Inkling merge/PEFT path. /tinker deploy writes API clients and a SERVING.md that says so. Do not send Inkling to HTDYM or vLLM.
Training
Cookbook get_lr("thinkingmachines/Inkling") raises NotImplementedError. Calibrate LR. Watch entropy. Prefer Small unless quality on the user's eval requires full Inkling.
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.
- 9d ago First seen · 57 lines · 66 tokens per session scan A 14c91c3597e6
tinker-inkling is a skill published in the GitHub repository gvkhosla/pi-tinker (24 stars, last pushed 12d ago), licensed Apache-2.0. It adds 66 tokens to every session and 702 once invoked, about $0.0003 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
fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
Training
Guides fine-tuning and post-training work with explicit data, objective, hardware, and rollback assumptions.
fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
model_finetuning
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.