| name | tenacious-publication-packager |
| description | Package Tenacious-Bench outputs into public, reviewer-friendly artifacts. Use when an agent is preparing a Hugging Face dataset or model card, datasheet, README, technical blog draft, publication checklist, executive memo support files, or community-engagement materials for this project. |
Tenacious Publication Packager
Overview
Ship public artifacts that are clear, reproducible, and not embarrassing a day later.
Read references/public-artifact-checklist.md before finalizing anything that will leave the repo.
Workflow
1. Start with reproducibility
A stranger should be able to understand what the artifact is, how it was built, and how to reproduce the headline result.
2. Package the benchmark, not just the files
Include:
- what the benchmark measures
- how the partitions work
- what the evaluator expects
- what the baseline is
- what the known limitations are
3. Keep the public story honest
The blog post, README, model card, and memo should all agree on the same core facts.
4. Check the boring details
Verify license, attribution, seed handling, sealed-slice rules, and artifact links before publishing.
5. Write for an external reader
Assume the reader does not know the cohort context. Name the problem, the method, the result, and the caveats plainly.
Output Shape
Prefer concise, linked artifacts over narrative sprawl:
- README with quickstart
- datasheet with real substance
- model card where applicable
- blog draft with evidence-backed claims
- community artifact text that points to the benchmark gap and contribution
Practical Defaults
- Public artifact quality is part of the grade.
- A smaller, cleaner publication beats a sprawling under-explained one.
- Treat consistency across artifacts as a release requirement.
