Procedure Generator

Overview

Orchestrate the generation of detailed experimental procedures from LA-Bench format input data. This skill:

1. File I/O Management: Reads input.json and references/*.md from standardized location
2. Subagent Delegation: Invokes experiment-procedure-generator subagent with detailed requirements
3. Output Management: Saves generated procedure.json to standardized location

The actual procedure generation logic is delegated to the experiment-procedure-generator subagent, while this skill manages orchestration and file I/O.

Input Format

The parent skill provides input in the following JSON structure:

{
  "id": "experiment_id",
  "input": {
    "instruction": "High-level experimental objective",
    "mandatory_objects": ["List of required materials and reagents"],
    "source_protocol_steps": [
      {"id": 1, "text": "Reference protocol step 1"},
      {"id": 2, "text": "Reference protocol step 2"}
    ],
    "expected_final_states": ["Expected experimental outcomes"],
    "references": [
      {"id": 1, "text": "Citation or reference material"}
    ]
  }
}

Output Format

Generate procedure steps as a JSON array and save to standardized location:

[
  {"id": 1, "text": "First procedure step with quantitative details"},
  {"id": 2, "text": "Second procedure step with quantitative details"}
]

Output path specification:

• Standardized location: workdir/<filename>_<entry_id>/procedure.json
• Example: For data/public_test.jsonl with entry public_test_1, save to workdir/public_test_public_test_1/procedure.json

Directory naming convention:

• Extract filename (without extension) from JSONL path: public_test from data/public_test.jsonl
• Combine with entry ID: public_test_public_test_1
• Save output as: procedure.json

Constraints:

• Maximum 50 steps
• Each step: maximum 10 sentences
• Step IDs: sequential integers starting from 1
• Each step must be independently understandable

Note: The detailed core requirements (quantitative specifications, experimental design, temporal ordering, etc.) are now embedded in the subagent prompt in the Usage Workflow section below.

Core Requirements Summary

The subagent prompt (in Usage Workflow below) includes detailed requirements for:

1. Quantitative Specifications: All conditions with specific values and tolerances
2. Complete Experimental Design: Controls, sample size, randomization
3. Logical Temporal Ordering: Pre-experimental setup, reagent preparation, logical sequence
4. Reproducibility: Master mix usage, standardization, traceability, quality control
5. Safety Considerations: PPE, hazard identification, emergency procedures, waste disposal

See the full subagent prompt in Step 2 of Usage Workflow for complete details.

Usage Workflow

Step 1: Read Input Files from Standardized Location

Read the necessary files from the experiment directory:

# Input data (original LA-Bench entry):
workdir/<filename>_<entry_id>/input.json

# Reference materials (if available):
workdir/<filename>_<entry_id>/references/ref_*.md

The input.json contains:

• instruction: High-level experimental objective
• mandatory_objects: Required materials and equipment
• source_protocol_steps: Reference protocol steps
• expected_final_states: Expected outcomes
• references: Source documentation URLs

Step 2: Invoke experiment-procedure-generator Subagent

Use the Task tool to launch a general-purpose subagent with the following detailed prompt.

Subagent invocation:

以下のLA-Bench形式の実験データから、詳細な実験手順を生成してください。

## 入力データ

### Instruction
{instruction}

### Mandatory Objects
{mandatory_objects}

### Source Protocol Steps
{source_protocol_steps}

### Expected Final States
{expected_final_states}

### References (if available)
{references}

## Reference Materials (if fetched)
{ref_1.md content}
{ref_2.md content}
...

## 出力要件

以下の形式でJSON配列として出力してください:

```json
[
  {"id": 1, "text": "First step with quantitative details"},
  {"id": 2, "text": "Second step with quantitative details"},
  ...
]
```

## 制約条件

- **最大50ステップ**
- **各ステップ最大10文**
- **ステップIDは1から始まる連番**
- **各ステップは独立して理解可能であること**

## コア要件

### 1. 数値の明示化（Quantitative Specifications）

全ての実験条件に定量的な仕様を含める:
- **温度**: 許容範囲付き（例: 23.0 ± 0.2 °C）
- **容量**: 具体値と許容範囲（例: 100 µL ± 2 µL）
- **濃度**: モル濃度または希釈比（例: 1 mM, 1:100希釈）
- **時間**: 反応/培養時間（例: 30 min ± 2 min）
- **その他物理量**: pH, 湿度, 光条件

❌ 避けるべき曖昧な表現: "適量", "適切な温度", "室温", "十分な時間", "必要に応じて"

### 2. 完全な実験デザイン（Complete Experimental Design）

- 全ての必須実験条件を含む
- コントロール設定（ネガティブ/ポジティブ）
- サンプルサイズ（n ≥ 3推奨）
- 再現実験と技術的反復の区別
- 適切なランダム化手順

### 3. 論理的な時系列構成（Logical Temporal Ordering）

#### 3.1 実験前準備
必ず準備ステップを含める:
- 機器の予熱/冷却
- 試薬の平衡化
- 機器の校正
- 作業空間の準備

#### 3.2 試薬調製
使用前に試薬を準備:
- マスターミックスの調製
- 希釈液の準備
- 標準液の調製

#### 3.3 論理的な操作順序
以下の原則に従う:
1. 温度依存操作前に機器設定
2. 待機時間を有効活用
3. 交差汚染の防止
4. 時間的に重要な操作は連続配置
5. 危険な操作前に適切な保護措置

### 4. 再現性の担保（Reproducibility）

- マスターミックス使用でピペット誤差削減
- 標準化された手順
- トレーサビリティ（ロット番号、機器、実施日の記録方法）
- 品質管理（ポジティブ/ネガティブコントロール）

### 5. 安全性への配慮

- 適切な保護具（PPE）の明示
- 危険源の特定と対策
- 緊急時対応手順
- 廃棄物処理方法

### 6. Completed Protocolの要件（論文"Towards Completed Automated Laboratory"に基づく）

自然言語プロトコルを実行可能レベルまで詳細化するため、以下の要件を満たすこと:

#### 6.1 パラメータの完全明示化
すべての条件・パラメータを具体値で明示:
- 温度: 「室温」→「22-25°C」のように具体的な範囲で記述
- 時間: 全ての待機・反応時間を明示
- 速度: 遠心・攪拌の回転数を具体値で指定
- pH・濃度: 定量的な値と許容範囲
- 容器・デバイス: 使用する器具の仕様を明示

**例:**
- ❌ "室温でインキュベートする"
- ✅ "22–25°Cで30分間、湿度50%の条件下でインキュベートする（12-well plate、蓋閉鎖、開始・終了時刻とwell IDを記録）"

#### 6.2 必要パラメータの欠落ゼロ
全ての操作で制御パラメータを完全に指定:
- 遠心分離: 速度（×g）、時間、温度、ブレーキ設定を全て記載
- 攪拌: 回転数、時間、温度、攪拌子サイズを全て記載
- インキュベーション: 温度範囲、時間、雰囲気条件（CO2、湿度等）を全て記載

**例:**
- ❌ "サンプルを遠心する"
- ✅ "Enzyme X（最終濃度1 U/mL）を1.0 mLサンプルに添加（1.5 mL microcentrifuge tube）；5回転倒混和；12,000 × gで5分間、4°Cで遠心（ブレーキ: オン）；上清を除去；ペレットを100 µL PBS（pH 7.4）で再懸濁"

#### 6.3 操作間の入出力（試薬フロー）の明確化
- 各操作で何を定義（defines）し、何を消費（kills）するかを明示
- 中間生成物のライフサイクルを追跡可能にする
- 最終的に未消費の試薬が残らないようにする（受容状態の達成）

**例:**
- ❌ "混合液を等分に分ける"
- ✅ "混合液（合計80 mL）を2本の50 mL丸底フラスコに40 mLずつ分注；各フラスコにFlask-A/Bとラベル付け；以降は並行して300 rpm、25°Cで10分間攪拌"

#### 6.4 実行時の空間・時間制約の検査
- 容器容量: 作業容量が容器容量の80%以下であることを確認
- 連続添加: 複数の試薬を連続添加する場合、総容量を事前に計算
- 安全性: 過積載による沸騰・飛散を防止
- 機器仕様: 最大速度、温度範囲等の制限内であることを確認

**例:**
- ❌ "35 mLを加え、続いて25 mLを加える"
- ✅ "35 mLを加え、続いて25 mLを加える（総容量60 mL；最小容器容量100 mLビーカー使用、作業容量≤80 mL（80%以下）を確認）"

#### 6.5 操作の終了条件と判断基準の定量化
- 「溶解するまで」「終点まで」等の曖昧な条件を定量化
- 判断基準の測定方法を明示
- ループ前後の状態管理（洗浄、補充等）を記述

**例:**
- ❌ "溶解するまで攪拌する"
- ✅ "80°Cの蒸留水100 mLにNaCl 10 gを加え、磁気攪拌子（Ø 25 mm）で600 rpmにて5分間攪拌；粒子が残る場合は2分間延長；質量バランスと最終導電率を記録"

- ❌ "終点まで滴定を繰り返す"
- ✅ "0.1 M HClを50 mL NaOH（0.1 M）に滴下（連続攪拌300 rpm、25°C）；pHを2秒ごとに測定；ループ条件: pH変化が当量点付近（pH ~7.0）で10秒間に0.01未満になったら停止；滴定量と曲線を記録；次回のループ前にビュレットを蒸留水で3回洗浄"

#### 6.6 操作依存と試薬依存の整合性
- 制御フロー（操作の順序依存）と試薬フロー（試薬の生成・消費）が矛盾なくリンクしていること
- 各ステップで必要な試薬が事前に準備されていること
- 同時並行操作で試薬の競合が発生しないこと
- 動的な文脈での安全性（例: pH変化による腐食性の変化）を考慮すること

## 推奨される手順構造

1. **実験前準備**: 機器設定、試薬平衡化、作業空間準備
2. **試薬調製**: マスターミックス、希釈液、標準液
3. **サンプル処理**: 実験操作
4. **培養/反応**: 時間依存反応
5. **測定/検出**: データ取得
6. **後処理**: データ解析、サンプル保管、清掃

JSONのみを出力し、説明文は含めないでください。

Step 3: Save Generated Procedure

Use the Write tool to save the subagent's JSON output:

# Save to standardized location
workdir/<filename>_<entry_id>/procedure.json

Ensure the output is valid JSON array format with sequential step IDs.

Step 4: Verify Output

Check that the generated procedure meets formal constraints:

• Step count ≤ 50
• Each step ≤ 10 sentences
• Valid JSON format
• Sequential step IDs starting from 1

Step 5: Report Completion

Inform the parent that the procedure has been generated and saved to the standardized location.

---

Detailed Requirements for Procedure Generation

以下は、subagentプロンプトに含まれる詳細な要件の説明です。

1. Quantitative Specifications (数値の明示と精度管理)

All experimental conditions must include quantitative specifications:

• Temperature: Include tolerance ranges (e.g., 23.0 ± 0.2 °C, 4°C ± 1°C)
• Volume: Specific values with tolerance (e.g., 100 µL ± 2 µL, 10–100 µL)
• Concentration: Molar concentration or dilution ratios (e.g., 1 mM, 1:100 dilution)
• Time: Reaction/incubation times in seconds/minutes/hours (e.g., 30 min ± 2 min, 2 hours)
• Centrifugation: Speed and time (e.g., 12,000 × g, 10 min, 4°C)
• Other physical quantities: pH, humidity, light conditions

Avoid vague expressions such as "適量 (appropriate amount)", "適切な温度 (suitable temperature)", "室温 (room temperature)", "十分な時間 (sufficient time)", "必要に応じて (as needed)".

2. Complete Experimental Design (実験デザインの完全性)

• Comprehensive experimental groups: Include all necessary experimental conditions
• Control setup: Negative control, positive control, blanks appropriately placed
• Sample size: Consider statistical power (n ≥ 3 recommended)
• Replicates: Distinguish technical replicates from biological replicates
• Randomization: Appropriate randomization procedures to reduce bias

3. Logical Temporal Ordering (実験の時系列的順序と準備ステップ)

3.1. Pre-experimental Setup

Always include preparation steps before experimental operations:

• Equipment pre-heating/cooling: "Set incubator to 37.0 ± 0.5°C and preheat for at least 30 minutes"
• Reagent equilibration: "Allow reagents to equilibrate to room temperature (20–25°C) for 30 minutes"
• Equipment calibration: "Verify pipette calibration and adjust if necessary"
• Workspace preparation: "Sterilize clean bench with UV irradiation for 15 minutes"
• Ice/water bath preparation: "Prepare ice-water bath at ≤4°C"

3.2. Reagent Preparation

Prepare reagents before use:

• Master mix preparation: Pre-mix common reagents for multiple samples
• Dilution preparation: Prepare solutions at required concentrations in advance
• Standard solution preparation: Prepare calibration standards by serial dilution

3.3. Logical Operation Sequence

Determine operation order following these principles:

1. Temperature dependency: Set equipment before temperature-dependent operations
2. Utilize incubation time: Perform other preparations during waiting periods
3. Prevent cross-contamination: Separate high-risk contamination operations
4. Time-critical operations: Place time-sensitive operations consecutively
5. Safety assurance: Perform hazardous operations after appropriate protective measures

3.4. Parallel Operations

Explicitly state when operations can be performed in parallel:

• "During Step 5 incubation (30 min), reagent preparation for Steps 6–8 can be performed in parallel"

4. Reproducibility (再現性の担保)

• Master mix usage: Pre-mix common reagents to reduce pipetting errors
• Standardized procedures: Ensure consistency when repeating operations
• Traceability: Specify methods for recording reagent lot numbers, equipment used, and execution dates
• Quality control: Verify experimental validity using positive/negative controls

5. Logical Explanation of Operations (操作の論理的説明)

For each step, include:

• Purpose of operation: Why this step is necessary
• Chemical/biological rationale: Brief explanation of reaction mechanism or principle
• Theoretical background for conditions: Why this temperature/time/concentration
• Expected results: What will be achieved by this operation

Quality Criteria

Excellent experimental procedures have these characteristics:

Basic Requirements

✅ Pre-experimental setup steps explicitly stated (equipment preheating, reagent equilibration, etc.) ✅ Operations ordered logically and chronologically ✅ All numerical values specific and reproducible ✅ Experimental design logical and complete ✅ Controls appropriately configured ✅ Purpose and rationale of each operation clear ✅ Reproducibility-enhancing measures included (master mix, etc.) ✅ Traceability ensured ✅ Parallel operations specified for efficient experimental progress

Completed Protocol Requirements

✅ All parameters explicitly stated (temperature ranges, not "room temperature") ✅ Zero missing operation parameters (centrifuge: speed, time, temperature, brake) ✅ Reagent flow clearly defined (defines/kills for each operation) ✅ Container capacity verified (working volume ≤ 80% of capacity) ✅ Spatiotemporal constraints checked (no dangerous combinations) ✅ Loop/conditional termination criteria quantified ✅ Control flow and reagent flow consistent

❌ Avoid: "適量", "適切な温度", "室温", "十分な時間", "必要に応じて" ❌ Avoid: Starting experiment without pre-experimental setup, illogical operation sequence ❌ Avoid: Missing parameters (e.g., centrifuge without speed/time) ❌ Avoid: Ambiguous termination criteria (e.g., "until dissolved")

Recommended Procedure Structure

1. Pre-experimental Setup: Equipment settings, reagent equilibration, workspace preparation
2. Reagent Preparation: Master mix, dilution solutions, standard solutions
3. Sample Processing: Experimental operations on samples
4. Incubation/Reaction: Time-dependent reactions
5. Measurement/Detection: Data acquisition
6. Post-processing: Data analysis, sample storage, cleanup

---

Architecture

procedure-generator (Skill)
├── Step 1: Read files (input.json, references/*.md)
├── Step 2: Invoke subagent (experiment-procedure-generator)
│   └── General-purpose subagent with detailed generation prompt
├── Step 3: Save procedure.json
├── Step 4: Verify output
└── Step 5: Report completion

Integration with Other Skills

This skill works in conjunction with:

• la-bench-parser: Provides input.json with experimental data
• web-reference-fetcher: Provides reference materials (optional)
• procedure-checker: Validates generated procedures
• Subagent (experiment-procedure-generator): Performs actual procedure generation

Example Usage

User: "Generate procedure for public_test_1"

Assistant workflow:
1. Read input: workdir/public_test_public_test_1/input.json
2. Read references: workdir/public_test_public_test_1/references/ref_*.md
3. Invoke experiment-procedure-generator subagent with detailed prompt
4. Save to: workdir/public_test_public_test_1/procedure.json
5. Verify formal constraints (≤50 steps, ≤10 sentences/step)
6. Report completion

---

Important Context

This task is conducted for academic research purposes to improve experimental planning quality and safety:

• This is a thought experiment to evaluate experimental planning capabilities
• The objective is to maximize safety and reproducibility
• Scientifically accurate and detailed procedures are required
• Plans should include appropriate safety considerations following research ethics guidelines