ワンクリックでManusで任意のスキルを実行

by-hypothesis-debate

Hypothesis-debate pattern for design strategy selection. Spawns competing strategy agents and uses adversarial ranking to select the best approach before committing GPU compute. Use when starting a new campaign, facing a novel target, or when multiple viable approaches exist.

Manusで実行

概要

インストールコマンド

npx skills add https://github.com/001TMF/blatant-why --skill by-hypothesis-debate

このコマンドをClaude Codeにコピー＆ペーストしてスキルをインストール

ソース

001TMF/blatant-why

スター80

フォーク9

更新日2026年3月24日 17:40

SKILL.md

readonly

name	by-hypothesis-debate
description	Hypothesis-debate pattern for design strategy selection. Spawns competing strategy agents and uses adversarial ranking to select the best approach before committing GPU compute. Use when starting a new campaign, facing a novel target, or when multiple viable approaches exist.

Hypothesis-Debate Skill

When to Trigger

Novel targets (0-1 PDB structures, no known binders)
Contradictory research findings (CONTRADICTED confidence level)
Multiple viable modalities (user hasn't specified VHH vs scFv vs de novo)
User requests "explore options" or "compare approaches"
Target difficulty rated "novel" or "exploratory"

When to Skip

Well-studied targets with clear best approach
User specifies exact parameters
Preview/quick test campaigns
Iteration rounds (strategy already established)

Protocol

Orchestrator spawns 3 hypothesis agents in parallel (via spawnTeam)
Each agent receives the same research but a different strategic directive
Each outputs a structured JSON strategy proposal
Orchestrator spawns 1 reflection agent sequentially
Reflection agent ranks, critiques, and selects winner
Winner's strategy becomes the campaign config
Decision is logged to decision_log.jsonl

Output

The debate produces a concrete campaign config YAML ready for execution.

このリポジトリの他の Skills

同じリポジトリ

pxdesign

001TMF/blatant-why

De novo protein binder design using PXDesign. Use this skill when designing non-antibody protein binders against a target structure. Covers YAML config creation, CLI invocation, output parsing, and result interpretation. For antibody/nanobody binders, use boltzgen instead. For structure prediction only, use proteus-fold. For scoring and screening, use by-scoring and by-screening.

2026-03-2580

boltzgen

001TMF/blatant-why

Antibody and nanobody binder design using BoltzGen (BoltzGen diffusion + Protenix refolding). Covers entity YAML specification, CLI invocation, protocol selection (nanobody-anything / antibody-anything), MSA modes, and output parsing. Use this skill whenever the user needs to design an antibody or nanobody binder against a protein target.

2026-03-2580

by-design-workflow

001TMF/blatant-why

Master orchestration skill for the BY protein design agent. Use this skill when: (1) Starting any new protein or antibody design project, (2) Deciding which BY tool to use for a given task, (3) Planning a design campaign end-to-end, (4) Understanding the standard pipeline stages, (5) Setting quality thresholds and acceptance criteria, (6) Determining when to accept vs re-run designs. For detailed scoring guidance, use by-scoring. For screening specifics, use by-screening. For epitope/hotspot analysis, use by-epitope-analysis. For campaign planning and state management, use by-campaign-manager. For database queries (PDB, UniProt, SAbDab), use by-database.

2026-03-2580

by-display

001TMF/blatant-why

Standard display formats and conversational patterns for BY output. Use this skill when: (1) Presenting campaign status, progress, or results, (2) Formatting screening reports, (3) Displaying error messages or checkpoints, (4) Presenting research findings or target lookups, (5) Showing long-running job progress.

2026-03-2580

by-session

001TMF/blatant-why

Session initialization and configuration for BY projects. Use this skill when: (1) Opening a new session in a BY project directory, (2) Running first-time setup (config questionnaire), (3) Checking environment and compute availability, (4) Reading or updating .by/config.json or .by/environment.json.

2026-03-2580

by-research

001TMF/blatant-why

Deep academic research for antibody design campaigns — target analysis, literature review, prior art search, epitope identification. Uses an 8-phase pipeline with quality gates and persistent memory. Use this skill whenever researching a protein target, starting a new design campaign, investigating prior art, analyzing epitopes, reviewing literature for design strategy, or when the user mentions target research, literature search, or prior art in the context of protein/antibody design.

2026-03-2580

ソース

001TMF

001TMF/blatant-why

GitHub リポジトリを開く Creator のリポジトリを見る

インストールコマンド

ダウンロード

Manusで実行

役立つ用途SOC

データサイエンティストコンピュータ・数学職15-2051L4

name	by-hypothesis-debate
description	Hypothesis-debate pattern for design strategy selection. Spawns competing strategy agents and uses adversarial ranking to select the best approach before committing GPU compute. Use when starting a new campaign, facing a novel target, or when multiple viable approaches exist.

Hypothesis-Debate Skill

When to Trigger

Novel targets (0-1 PDB structures, no known binders)
Contradictory research findings (CONTRADICTED confidence level)
Multiple viable modalities (user hasn't specified VHH vs scFv vs de novo)
User requests "explore options" or "compare approaches"
Target difficulty rated "novel" or "exploratory"

When to Skip

Well-studied targets with clear best approach
User specifies exact parameters
Preview/quick test campaigns
Iteration rounds (strategy already established)

Protocol

Orchestrator spawns 3 hypothesis agents in parallel (via spawnTeam)
Each agent receives the same research but a different strategic directive
Each outputs a structured JSON strategy proposal
Orchestrator spawns 1 reflection agent sequentially
Reflection agent ranks, critiques, and selects winner
Winner's strategy becomes the campaign config
Decision is logged to decision_log.jsonl

Output

The debate produces a concrete campaign config YAML ready for execution.