autoresearch
Autonomous LLM pretraining research loop on single GPU using Karpathy's autoresearch framework. Use when the user wants to: (1) start an autonomous LLM training experiment session, (2) run overnight hyperparameter/architecture search on a single-GPU nanochat setup, (3) iterate on train.py modific...
Install this skill
or
95/100
Security score
The autoresearch skill was audited on Jul 29, 2026 and we found 1 security issue across 1 threat category. Review the findings below before installing.
Categories Tested
Security Issues
medium line 54
Access to home directory dotfiles
SourceSKILL.md
| 52 | 2. Create branch: `git checkout -b autoresearch/<tag>` |
| 53 | 3. Read `README.md`, `prepare.py`, `train.py` for full context |
| 54 | 4. Verify `~/.cache/autoresearch/` has data shards and tokenizer. If not, tell human to run `uv run prepare.py`. |
| 55 | 5. Create `results.tsv` with header: `commit\tval_bpb\tmemory_gb\tstatus\tdescription` |
| 56 | 6. First run always establishes the baseline — run `train.py` as-is. |
Scanned on Jul 29, 2026
View Security Dashboard