CtrlK
BlogDocsLog inGet started
Tessl Logo

arbor

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 pipeline", "beat the baseline on this benchmark", "run a search over approaches and keep the best", "do an MLE-bench / Kaggle-style optimization", or any long-horizon "make this artifact better and don't just memorize the dev set" task. Trigger it even when the user doesn't say "Arbor" or "hypothesis tree" but describes repeated experiment-and-evaluate loops, branching exploration of competing ideas, or worries about a dev/test gap. Runs Claude itself as the coordinator with subagent executors in isolated git worktrees; for the standalone `arbor` CLI tool see references/arbor-upstream.md.

77

Quality

96%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

92%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is a well-structured, executable playbook: a clearly sequenced six-step loop with a real validation gate, concrete tree.py commands throughout, and a clean one-level-deep reference structure. Its only weakness is mild motivational verbosity that restates the procedure in a separate Principles section.

Suggestions

Tighten or fold the 'Principles that make this work' section into the six steps — it largely restates them — to remove motivational padding and recover tokens.

Trim rhetorical framing like 'This is the heart of HTR; do not collapse it into ad-hoc editing' and 'This is the step that makes the tree more than a log' that add length without procedural content.

Condense the multi-paragraph citation block; keep the fetch-and-cite instruction but drop the repeated restatements of the version rule.

DimensionReasoningScore

Conciseness

The body is mostly efficient and largely covers HTR mechanics that Claude does not already know (a 2026 paper), but it carries motivational prose ('This is the step that makes the tree more than a log', 'This is the heart of HTR; do not collapse it into ad-hoc editing') and a 'Principles' section that restates the six steps — minor over-explanation that could be trimmed.

4 / 5

Actionability

Provides copy-paste-ready CLI invocations for every phase — `tree.py init`, `observe`, `add-node`, `set-evidence`, `propagate`, `prune`, `merge`, `set-status`, `status` — with real flags and example values, plus a pointer to the executor-brief template.

5 / 5

Workflow Clarity

The six-step coordinator loop (Observe → Ideate → Select → Dispatch → Backpropagate → Decide) is explicitly sequenced with a validation checkpoint — the held-out merge gate admits a change only on E_test, run in a fresh worktree, with explicit error-recovery guidance when the gate rejects it.

5 / 5

Progressive Disclosure

The body is an overview with a dedicated 'Reference files' section signaling one-level-deep pointers (htr-methodology.md, executor-brief.md, report-template.md, arbor-upstream.md), each annotated with what it contains and when to read it; all referenced files exist in the bundle.

5 / 5

Total

19

/

20

Passed

Description

100%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

This is an exemplar description: it states what the skill does concretely, when to trigger it with natural user language, and how it differs from neighboring skills, all in third person. It is dense yet every phrase earns its place.

DimensionReasoningScore

Specificity

Names concrete artifact types ('code, training recipe, agent harness, data pipeline, prompt') and concrete mechanisms ('Runs Claude itself as the coordinator with subagent executors in isolated git worktrees'), giving comprehensive coverage of specific actions rather than vague language.

5 / 5

Completeness

Explicitly answers 'what' (autonomously improve an artifact against an evaluator via HTR with a coordinator/subagent/worktree/merge-gate design) and 'when' ('Use this whenever someone wants to iteratively optimize... Trigger it even when the user doesn't say "Arbor"...'), with concrete trigger phrases.

5 / 5

Trigger Term Quality

Packed with natural user phrases — 'get my model's eval score up', 'improve this agent/harness', 'tune this pipeline', 'beat the baseline on this benchmark', 'do an MLE-bench / Kaggle-style optimization' — plus synonyms ('repeated experiment-and-evaluate loops', 'branching exploration', 'dev/test gap').

5 / 5

Distinctiveness Conflict Risk

Occupies a clear niche (iterative artifact optimization under an evaluator with a held-out dev/test split) and even names adjacent skills to use instead ('hypothesis-generation', 'scientific-brainstorming'), minimizing wrong-skill triggers.

5 / 5

Total

20

/

20

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository
K-Dense-AI/scientific-agent-skills
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.