Skip to content
Get packs

Super Research (Autonomous Research)

Runs long, disciplined research with logged, comparable, auditable results across eight modes

Source Download skill
Repository
XiaomiMiMo/MiMo-Code
Stars
13,743
Licence
MIT
Files
15 files

What it does

Once installed, the agent works like a long-running researcher instead of giving a one-shot answer. It agrees on a plan with you, records a baseline, runs many small experiments, analyses, or source checks, and logs every attempt including failures in a tab-separated file so you can audit the work later. It picks one of eight modes from your request, such as optimizing a metric, surveying literature, analysing a dataset, comparing options, debugging a regression, ablation studies, reproducing a paper, or writing a paper with citation checks.

Great for

  • Developers who want an agent to run overnight experiments and hill-climb a measurable metric
  • Researchers or students who need a literature survey or dataset analysis with traceable evidence
  • Teams choosing between libraries, models, or prompts using a fair, fixed benchmark

Skip it if

  • Quick factual questions: the skill adds heavy process and is overkill for a one-line answer
  • Unattended runs on production systems or shared repos: the loop keeps going without checking in and can edit code or run commands for hours, so it needs sandboxing and a clear stopping condition
  • Final academic claims without human review: the skill itself says the human decides later, and citation checks only cover what the free APIs return

Try asking

  • “Optimize the validation loss of this training script overnight and keep a log of every experiment”
  • “Survey the recent literature on retrieval-augmented generation and summarize the evidence”
  • “Compare these three text-cleaning libraries under the same test cases and recommend one”
  • “Check the citations in my draft paper for fabricated references”
Step-by-step guide: how to use Super Research (Autonomous Research)Install, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 15 original files, the MIT licence and an attribution note.

    super-research.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip super-research.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip super-research.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Give a clear goal and stopping condition in your first message so the agent's contract is accurate before it runs autonomously
  • Review the TSV log and the contract afterwards rather than trusting the summary alone
  • Run long experiments inside a sandbox or a git branch so reverts and edits are easy to undo

What's inside

  • evals/evals.json
  • evals/toy_bench/candidates/cleaner_a.py
  • evals/toy_bench/candidates/cleaner_b.py
  • evals/toy_bench/candidates/cleaner_c.py
  • evals/toy_bench/README.md
  • evals/toy_bench/run_case.sh
  • evals/toy_dataset/dataset.csv
  • evals/toy_dataset/README.md
  • README.en.md
  • README.es.md
  • README.fr.md
  • README.ja.md
  • README.md
  • README.ru.md
  • SKILL.md

Redistributed unmodified from XiaomiMiMo/MiMo-Code under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.