allenai

Show the user the agent's work on a research project and save iterations on the user's behalf. Scaffold rendering and deploy infrastructure (Quarto today, GitHub Pages, dev container), show the rendered output, save iterations. Doesn't handle research execution (use `asta-flows`).

Install or update the `asta` CLI to the version pinned by this plugin. Invoke whenever `asta` is not on PATH (e.g. `command not found` for `asta`) or when a subcommand/flag appears missing — that signals a stale install needing an update.

Converse with the user to identify the next most productive units of work for the current project, and record them in project.md under Pending Work. Use when the user wants to plan what to work on next, or when there is no project.md yet.

Execute the approved plan in work/<slug>/README.md (status ready), write outputs into work/<slug>/data/, record what happened in the Results section, then hand off to review-work. Use when there is a ready work item.

Write a detailed, executable plan for a single unit of work, then get it approved by review-plan. Updates work/<slug>/README.md with concrete instructions and sets status to ready (or needs-input if review repeatedly rejects).

Critic skill. Given a plan in work/<slug>/README.md (status pending-review), check that the steps are concrete, the tools are appropriate, and the plan is realistic. Write a verdict and feedback back into the README.

Critic skill. Given an executed work item (status pending-assessment), assess whether the Goal was actually accomplished. Identify the root cause when it was not (incorrect assumption, technical blocker, missing data). Updates work/<slug>/README.md Assessment section.

Router skill that advances the project autonomously. Examines project.md Pending Work, dispatches to do-work for ready items or plan-work otherwise, and loops until no actionable work remains. Use when the user says "keep going", "work on the project", "run the next step".

Runnable evaluation template scripts for ML tasks. Match task_type to template, adapt CONFIG, run.

Use this skill for Hugging Face Dataset Viewer API workflows that fetch subset/split metadata, paginate rows, search text, apply filters, download parquet URLs, and read size or statistics.

Create and manage datasets on Hugging Face Hub. Supports initializing repos, defining configs/system prompts, streaming row updates, and SQL-based dataset querying/transformation. Designed to work alongside HF MCP server for comprehensive dataset workflows.

Add and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval. Works with the model-index metadata format.

Run a VLA model evaluation against a simulation benchmark. Use this skill whenever the user wants to evaluate, benchmark, test, or run a model on a sim environment — even if they say it casually like 'try OpenVLA on LIBERO' or 'get me CALVIN scores'. Covers the full workflow: serving the model, launching the benchmark, sharding for speed, merging results, and interpreting output.

Add a new VLA model server to the evaluation harness. Use this skill whenever the user wants to integrate, create, or add a new model — e.g. 'add OpenVLA server', 'integrate RT-2', 'hook up my model', 'write a model server'. Also use when they ask how model servers work or want to understand the server interface.

Add a new simulation benchmark to the VLA evaluation harness. Use this skill whenever the user wants to integrate, create, or add a new benchmark or simulation environment — e.g. 'add ManiSkill3', 'integrate OmniGibson', 'hook up a new sim'. Also use when they ask how benchmarks are structured or want to understand the benchmark interface.

Update the body of a GitHub pull request. Use when the user asks to update, edit, or modify a PR description/body.

Monitor Beaker experiments until completion. Use when the user asks to monitor, check, or track a Beaker experiment.

Local document metadata index for scientific documents

Local document metadata index for scientific documents

This skill should be used when the user asks to "create a verification script", "write a test training run", "make a quick training job", "verify my change with a short run", "launch a smoke test", or wants a small Beaker job to validate that a feature works end-to-end on GPUs. Also triggers when the user mentions creating a modified 190M script to test a specific behavior.

This skill should be used when the user asks to "start a task", "run a verifier", "verify my solution", "work on a Harbor task", "pier start", "pier verify", or needs to interactively work on Harbor benchmark tasks.