Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/topprismdata/cultivating-ml-agentnpx agentmods add skills/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submissionWrote 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/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submission)<a href="https://agentmods.dev/skills/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submission"><img src="https://agentmods.dev/badge/skills/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submission/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/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submission"><img src="https://agentmods.dev/badge/skills/topprismdata/cultivating-ml-agent/ptcg-bc-large-model-submission.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00207 | $0.01630 |
| Opus 5 | $0.00103 | $0.00815 |
| Sonnet 5 | $0.00041 | $0.00326 |
| Haiku 4.5 | $0.00021 | $0.00163 |
Grade A, and why
ptcg-bc-large-model-submission scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 138 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Pokémon TCG — BC Large Model Submission Pipeline
The Core Lesson (validated)
The single biggest jump (165 → 600) came from using the right checkpoint.
Training produces checkpoints_bc_large/best.pth (66 MB, 16.6M params).
Submissions require model_bc.pth (v6 path). Forgetting to copy best.pth
→ submission/model_bc.pth is the #1 reason scores stay low after training.
| Submission | Source | LB |
|---|---|---|
| v6 | stale small-model checkpoint | 165.6 |
| v7 | checkpoints_bc_large/best.pth |
600.0 |
Pipeline (validated end-to-end)
Step 1: Scale BC data via Kaggle dataset snapshots
Episodes are published daily as kaggle/pokemon-tcg-ai-battle-episodes-YYYY-MM-DD.
Each snapshot is ~5000 games (not 200 — that was the pitfall of per-file
downloading). Use whole-snapshot download (one zip per day), not per-file:
for date in 2026-06-20 2026-06-22 ...; do
kaggle datasets download kaggle/pokemon-tcg-ai-battle-episodes-$date -p bc_zips
unzip bc_zips/...zip -d bc_data/$date
done
# Total: 80K+ games → ~6.4M (state, action) pairs
Step 2: Stream extract to part files (avoids OOM)
extract_bc_data.py must NOT accumulate the full sample list in memory before
saving. With 6.4M samples at ~10KB each = 64 GB → instant OOM. Write part
files every 50K samples, then del part_data after each:
PART_SIZE = 50_000
for i, rf in enumerate(replay_files):
samples = extract_from_replay(rf)
batch.extend(samples)
if len(batch) >= PART_SIZE:
torch.save(batch, f"bc_dataset_part_{part_idx}.pt")
part_idx += 1; batch = []
torch.save({"n_parts": part_idx, "total": total_samples}, "bc_dataset_meta.pt")
Step 3: Stream-sampling training (avoids OOM at load)
train_bc_large.py loading 129 part files → extend() into one list = OOM.
Sample per-part to quota (50K / 129 ≈ 400 per part), del part_data after each:
per_part_quota = MAX_SAMPLES // len(part_files)
for pf in part_files:
part_data = torch.load(pf, weights_only=False)
if len(part_data) > per_part_quota:
idx = np.random.choice(len(part_data), per_part_quota, replace=False)
samples.extend([part_data[j] for j in idx])
else:
samples.extend(part_data)
del part_data # critical: release memory
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 · 138 lines · 207 tokens per session scan A 9841e32e8a63
ptcg-bc-large-model-submission is a skill published in the GitHub repository topprismdata/cultivating-ml-agent (5 stars, last pushed 15d ago), licensed MIT. It adds 207 tokens to every session and 1,630 once invoked, about $0.0010 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 2026-08-31.
Other skills, from other repositories
huggingface-hub
Hugging Face Hub CLI (hf) — search, download, and upload models and datasets, manage repos, query datasets with SQL, deploy inference endpoints, manage Spaces and buckets.
tensorboard
Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit.
mlflow
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform.
datachain-knowledge
Use whenever datasets, cloud storage buckets, or data pipelines are mentioned — creating, saving, querying, listing, exploring, deleting, or processing data in S3, GCS, Azure Blob, or local storage. Also use when running any script that may create datasets as a side effect. Maintains a knowledge base at dc-knowledge/…
install-openviking-memory
Install and configure the OpenViking long-term memory plugin for OpenClaw via natural conversation. Once installed, the plugin automatically captures facts from chats and recalls relevant context before each reply (auto-capture + auto-recall, cross-session). Covers prerequisites, install through OpenClaw's plugin…
pufferlib
Version-aware guidance for PufferLib reinforcement-learning environments, vectorization, policies, PuffeRL training, evaluation, and safe checkpoint review. Use when adapting Gymnasium/PettingZoo environments to published PufferLib 3.0.0 or working with the redesigned native 4.0 source line.