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 PKU-YuanGroup/OpenAI4S --skill bioprobenchgit clone --depth 1 https://github.com/PKU-YuanGroup/OpenAI4SWrote 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/pku-yuangroup/openai4s/bioprobench)<a href="https://agentmods.dev/skills/pku-yuangroup/openai4s/bioprobench"><img src="https://agentmods.dev/badge/skills/pku-yuangroup/openai4s/bioprobench/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/pku-yuangroup/openai4s/bioprobench"><img src="https://agentmods.dev/badge/skills/pku-yuangroup/openai4s/bioprobench.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 | $0.00046 | $0.02444 |
| Opus 5 | $0.00023 | $0.01222 |
| Sonnet 5 | $0.00009 | $0.00489 |
| Haiku 4.5 | $0.00005 | $0.00244 |
Grade A, and why
bioprobench 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 7d 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 — 224 lines — stays where its author put it; the contents beside it link to each section on GitHub.
BioProBench — protocol understanding and reasoning
Biological protocols are where a plausible-sounding answer becomes a failed experiment: a wrong dosage, a swapped step, an unflagged hazard. BioProBench scores a model on five tasks over real wet-lab protocols, roughly 5,000 instances in the full release.
| Task | What it measures | Metrics returned |
|---|---|---|
PQA |
Protocol question answering — reagents, dosages, parameters | Accuracy, Brier_Score, Failed_Rate |
ORD |
Step ordering — reconstructing procedural sequence | Exact_Match, Kendall_Tau, Failed_Rate |
ERR |
Error correction — is this modified step valid | accuracy, precision, recall, f1, failed_rate |
GEN |
Protocol generation — synthesising steps | BLEU, METEOR, ROUGE-L, KW_F1, Step_Recall, Redundancy_Penalty, Failed_Rate |
REA-ERR |
Error reasoning, graded by an LLM judge | Consistency, Failure_Rate, Total, Failed, Total_Items, Judged, Unjudged, Coverage |
Metric key casing differs per task — ERR returns lowercase keys, the rest are
capitalised. Read them off the table above rather than guessing.
The input contract is the thing that bites
run_bioprobench_eval does not take a plain model-output file and compare
it against a separate answer key. It takes one file that already has the
ground truth merged into each record alongside the model's response. The
upstream inference scripts produce exactly this, by adding a
generated_response key to each benchmark record in place.
Hand it a file containing only model outputs and it does not raise: every
record simply fails to parse and the metrics come back at zero. The envelope
says so — status is "failed" when nothing scored and "partial" when some
records dropped out — but still check Failed_Rate on every run. A rate of
1.0 means the input contract was violated, not that the model scored zero.
Required keys per record, per task:
| Task | Model output key | Ground-truth key(s) |
|---|---|---|
PQA |
generated_response |
answer |
ORD |
generated_response |
wrong_steps, correct_steps |
ERR |
generated_response |
is_correct (true/false, 1/0, or "true"/"false") |
GEN |
generated_response |
output (string, or list of reference steps) |
REA-ERR |
LLM_judge |
none — the judgment text is itself the signal |
What ships with it
12 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.
- data/README_zh.md 1.1 KB
- data/README.md 1.1 KB
- data/sample_pqa_output.json 962 B
- kernel.py 30 KB runs code
- README_zh.md 4.3 KB
- README.md 4.3 KB
- Scripts/generate_response_local.py 3.8 KB runs code
- Scripts/generate_response.py 4.2 KB runs code
- Scripts/LLM-as-a-judge_for_REA-ERR.py 6.1 KB runs code
- Scripts/prompt_format.py 6.9 KB runs code
- Scripts/README_zh.md 1.7 KB
- Scripts/README.md 1.8 KB
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.
- 7d ago Changed · +4 lines 56ddbe9da8c2
- 11d ago First seen · 220 lines · 46 tokens per session scan A 2469266101de
bioprobench is a skill published in the GitHub repository PKU-YuanGroup/OpenAI4S (403 stars, last pushed today), licensed MIT. It adds 46 tokens to every session and 2,444 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
histolab
Lightweight WSI tile extraction and preprocessing. Use for basic slide processing tissue detection, tile extraction, stain normalization for H&E images. Best for simple pipelines, dataset preparation, quick tile-based analysis. For advanced spatial proteomics, multiplexed imaging, or deep learning pipelines use pathml.
pyhealth
Comprehensive healthcare AI toolkit for developing, testing, and deploying machine learning models with clinical data. This skill should be used when working with electronic health records (EHR), clinical prediction tasks (mortality, readmission, drug recommendation), medical coding systems (ICD, NDC, ATC)…
deepchem
Molecular ML with diverse featurizers and pre-built datasets. Use for property prediction (ADMET, toxicity) with traditional ML or GNNs when you want extensive featurization options and MoleculeNet benchmarks. Best for quick experiments with pre-trained models, diverse molecular representations. For graph-first…
torch-geometric
Graph Neural Networks (PyG). Node/graph classification, link prediction, GCN, GAT, GraphSAGE, heterogeneous graphs, molecular property prediction, for geometric deep learning.
zarr-python
Chunked N-D arrays for cloud storage. Compressed arrays, parallel I/O, S3/GCS integration, NumPy/Dask/Xarray compatible, for large-scale scientific computing pipelines.
alphafold-database
Access AlphaFold 200M+ AI-predicted protein structures. Retrieve structures by UniProt ID, download PDB/mmCIF files, analyze confidence metrics (pLDDT, PAE), for drug discovery and structural biology.