Skip to content
Get packs

Arbor Hypothesis Tree Refinement

Runs organized multi-experiment optimization of an artifact with dev/test evaluators and a persistent hypothesis tree.

Source Download skill
Repository
K-Dense-AI/scientific-agent-skills
Stars
55,780
Licence
MIT
Files
6 files

What it does

When installed, the agent treats a measurable improvement task (such as tuning a model, an agent harness, a data pipeline, or a prompt) as a long run of many experiments. It records hypotheses, results and lessons in a persistent JSON tree, tests each idea in an isolated Git worktree, and only accepts a change when it also improves on a separate held-out test evaluation. A bundled Python script keeps the bookkeeping consistent, while the agent makes the scientific judgments.

Great for

  • ML engineers running repeated training or recipe experiments with a scoring command
  • Developers optimizing an agent harness or prompt against an automatic benchmark
  • Researchers who need an auditable record of which experiments were tried and why

Skip it if

  • One-off fixes or single answers, where the setup overhead is not worth it
  • Open-ended brainstorming with no evaluator; the skill itself points to other skills for that
  • Tasks where no clean held-out test set exists: results should then be labelled development-only, and the skill's claims of overfitting protection do not apply

Try asking

  • “Use Arbor to improve my model's validation loss over about 12 experiments, with a held-out test split”
  • “Run an Arbor optimization loop on my search agent's BrowseComp accuracy”
  • “Set up a hypothesis tree to tune my data filtering pipeline against downstream model scores”
Step-by-step guide: how to use Arbor Hypothesis Tree RefinementInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 6 original files, the MIT licence and an attribution note.

    arbor.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

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

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip arbor.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

  • Freeze the evaluator, split and metric before search, and keep a final untouched evaluation set for the reported claim.
  • Start with a small budget (10 to 20 cycles) and review the recorded evidence before extending the run.
  • Keep raw evaluator output, seeds and split IDs alongside the .arbor directory so results can be audited.

What's inside

  • references/arbor-upstream.md
  • references/executor-brief.md
  • references/htr-methodology.md
  • references/report-template.md
  • scripts/tree.py
  • SKILL.md

Redistributed unmodified from K-Dense-AI/scientific-agent-skills 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.