ResearchSAFE

Structured deep research skill for Claude Code/Open Code/Codex with human-in-the-loop control

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: research
user-invocable: true
allowed-tools: Read, Write, Glob, WebSearch, Task, AskUserQuestion
description: Conduct preliminary research on a topic and generate research outline. For academic research, benchmark research, technology selection, etc.
---

# Research Skill - Preliminary Research

## Trigger
`/research <topic>`

## Workflow

### Step 1: Generate Initial Framework from Model Knowledge
Based on topic, use model's existing knowledge to generate:
- Main research objects/items list in this domain
- Suggested research field framework

Output {step1_output}, use AskUserQuestion to confirm:
- Need to add/remove items?
- Does field framework meet requirements?

### Step 2: Web Search Supplement
Use AskUserQuestion to ask for time range (e.g., last 6 months, since 2024, unlimited).

**Parameter Retrieval**:
- `{topic}`: User input research topic
- `{YYYY-MM-DD}`: Current date
- `{step1_output}`: Complete output from Step 1
- `{time_range}`: User specified time range

**Hard Constraint**: The following prompt must be strictly reproduced, only replacing variables in {xxx}, do not modify structure or wording.

Launch 1 web-search-agent (background), **Prompt Template**:
```python
prompt = f"""## Task
Research topic: {topic}
Current date: {YYYY-MM-DD}

Based on the following initial framework, supplement latest items and recommended research fields.

## Existing Framework
{step1_output}

## Goals
1. Verify if existing items are missing important objects
2. Supplement items based on missing objects
3. Continue searching for {topic} related items within {time_range} and supplement
4. Supplement new fields

## Output Requirements
Return structured results directly (do not write files):

### Supplementary Items
- item_name: Brief explanation (why it should be added)
...

### Recommended Supplementary Fields
- field_name: Field description (why this dimension is needed)
...

### Sources
- [Source1](url1)
- [Source2](url2)
"""
```

**One-shot Example** (assuming researching AI Coding History):
```
## Task
Research topic: AI Coding History
Current date: 2025-12-30

Based on the following initial framework, supplement latest items and recommended research fields.

## Existing Framework
### Items List
1. GitHub Copilot: Developed by Microsoft/GitHub, first mainstream AI coding assistant
2. Cursor: AI-first IDE, based on VSCode
...

### Field Framework
- Basic Info: name, release_date, company
- Technical Features: underlying_model, context_window
...

## Goals
1. Verify if existing items are missing important objects
2. Supplement items based on missing objects
3. Continue searching for AI Coding History related items within since 2024 and supplement
4. Supplement new fields

## Output Requirements
Return structured results directly (do not write files):

### Supplementary Items
- item_name: Brief explanation (why it should be added)
...

### Recommended Supplementary Fields
- field_name: Field description (why this dimension is needed)
...

### Sources
- [Source1](url1)
- [Source2](url2)
```

### Step 3: Ask User for Existing Fields
Use AskUserQuestion to ask if user has existing field definition file, if so read and merge.

### Step 4: Generate Outline (Separate Files)
Merge {step1_output}, {step2_output} and user's existing fields, generate two files:

**outline.yaml** (items + config):
- topic: Research topic
- items: Research objects list
- execution:
  - batch_size: Number of parallel agents (confirm with AskUserQuestion)
  - items_per_agent: Items per agent (confirm with AskUserQuestion)
  - output_dir: Results output directory (default: ./results)

**fields.yaml** (field definitions):
- Field categories and definitions
- Each field's name, description, detail_level
- detail_level hierarchy: brief -> moderate -> detailed
- uncertain: Uncertain fields list (reserved field, auto-filled in deep phase)

### Step 5: Output and Confirm
- Create directory: `./{topic_slug}/`
- Save: `outline.yaml` and `fields.yaml`
- Show to user for confirmation

## Output Path
```
{current_working_directory}/{topic_slug}/
  ├── outline.yaml    # items list + execution config
  └── fields.yaml     # field definitions
```

## Follow-up Commands
- `/research-add-items` - Supplement items
- `/research-add-fields` - Supplement fields
- `/research-deep` - Start deep research

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: Research](https://modelsatlas.com/skills/weizhena/research/badge.svg)](https://modelsatlas.com/skills/weizhena/research/)

Not your listing, or something out of date? Claim or update your listing.