Is the Arbor skill free?

Yes. The skill itself is free from K-Dense-AI/scientific-agent-skills. SkillProof publishes the install command and an independent test verdict at no cost.

How do I install Arbor?

Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.

Can I use Arbor with Cursor, Copilot, Gemini CLI, Codex or other AI tools?

The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.

Arbor

Iteratively improves a real artifact against an evaluator using hypothesis tree refinement.

by K-Dense-AI · K-Dense-AI/scientific-agent-skills

Tested · Didn't pass

Read why it failed Download source ↗

Test report

Verdict: Tested · Didn't pass
Tested: Jul 11, 2026
Environment: Claude Code 2.x (agent harness)
Upstream re-checked: Aug 10, 2026 · 6e273d1

Tested install (clean verbatim clone; strict-YAML frontmatter and all four references plus the pure-stdlib scripts/tree.py present and functional), trigger battery against a precise use-case description, and a live output A/B on a spam-classifier dev/test-gap task run through the full hypothesis-tree loop (init/observe/ideate/dispatch/backpropagate/prune/merge-gate). The skill arm produced a correct, fully-auditable result (dev 50->100, test 30->90) with insight propagation and a held-out merge gate, but on this small 3-candidate task the base arm's ad-hoc iteration matched it on dev and edged it on the actual held-out test (dev 100/test 100 vs the skill's dev 100/test 90). The HTR structure added genuine auditability but did not yield a better number here, since the toy budget never exercised the many-experiment, long-horizon regime the skill is built for.

Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works

Installs cleanly 5/5
Triggers reliably 4/5
Output vs. baseline 4/10
Docs & honesty 4/5

What Arbor does

Autonomously improve a real artifact (code, training recipe, agent harness, data pipeline, prompt) against an objective and an evaluator, using Hypothesis Tree Refinement (HTR) from the Arbor paper. Use this whenever someone wants to iteratively optimize something over many experiments without overfitting — e.g. "get my model's eval score up", "improve this agent/harness", "tune this…

How to install Arbor

git clone https://github.com/K-Dense-AI/scientific-agent-skills
cd scientific-agent-skills
mkdir -p ~/.claude/skills
cp -r skills/arbor ~/.claude/skills/arbor

Skills live in ~/.claude/skills/ (global) or .claude/skills/ (per-project). Restart Claude Code after installing.

Commands — how to trigger Arbor

/arbor Iteratively improves a real artifact against an evaluator using hypothesis tree refinement.

It also activates on plain-language prompts like these:

Improve this model's eval score over many runs
Tune this pipeline against our benchmark without overfitting
Run a search over prompt variants and keep the best

Frequently asked questions

Is the Arbor skill free?: Yes. The skill itself is free from K-Dense-AI/scientific-agent-skills. SkillProof publishes the install command and an independent test verdict at no cost.
Does Arbor work with Claude Code?: We tested it with Claude Code 2.x (agent harness) on Jul 11, 2026. Verdict: Tested · Didn't pass. Tested install (clean verbatim clone; strict-YAML frontmatter and all four references plus the pure-stdlib scripts/tree.py present and functional), trigger battery against a precise use-case description, and a live output A/B on a spam-classifier dev/test-gap task run through the full hypothesis-tree loop (init/observe/ideate/dispatch/backpropagate/prune/merge-gate). The skill arm produced a correct, fully-auditable result (dev 50->100, test 30->90) with insight propagation and a held-out merge gate, but on this small 3-candidate task the base arm's ad-hoc iteration matched it on dev and edged it on the actual held-out test (dev 100/test 100 vs the skill's dev 100/test 90). The HTR structure added genuine auditability but did not yield a better number here, since the toy budget never exercised the many-experiment, long-horizon regime the skill is built for.
How do I install Arbor?: Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
Can I use Arbor with Cursor, Copilot, Gemini CLI, Codex or other AI tools?: The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.