meta-pytorch/skill-writer

TorchRec Skill Writer

This Skill helps you create well-structured Agent Skills for Claude Code specifically for the TorchRec project.

When to use this Skill

Use this Skill when:

• Creating a new Agent Skill for TorchRec
• Writing or updating SKILL.md files
• Designing skill structure and frontmatter
• Converting existing TorchRec workflows into Skills

Instructions

Step 1: Determine Skill scope

First, understand what the Skill should do:

1. Ask clarifying questions:

   • What specific TorchRec capability should this Skill provide?
   • When should Claude use this Skill?
   • What tools or resources does it need?

2. Keep it focused: One Skill = one capability

   • Good: "sharding-optimizer", "embedding-config-validator"
   • Too broad: "distributed-training", "model-tools"

Step 2: Choose Skill location

TorchRec Skills should be placed in:

fbcode/torchrec/.claude/skills/<skill-name>/SKILL.md

Step 3: Create Skill structure

Create the directory and files:

mkdir -p fbcode/torchrec/.claude/skills/skill-name

For multi-file Skills:

skill-name/
├── SKILL.md (required)
├── reference.md (optional)
├── examples.md (optional)
└── templates/ (optional)

Step 4: Write SKILL.md frontmatter

Create YAML frontmatter with required fields:

---
name: skill-name
description: Brief description of what this does and when to use it
---

Field requirements:

• name:

  • Lowercase letters, numbers, hyphens only
  • Max 64 characters
  • Must match directory name
  • Good: sharding-optimizer, kjt-validator
  • Bad: Sharding_Optimizer, KJT Validator!

• description:

  • Max 1024 characters
  • Include BOTH what it does AND when to use it
  • Use specific trigger words users would say
  • Mention TorchRec concepts (embeddings, sharding, KJT, etc.)

Optional frontmatter fields:

• allowed-tools: Restrict tool access (comma-separated list)

  allowed-tools: Read, Grep, Glob
  

• argument-hint: Hint for expected arguments

  argument-hint: [feature or task description]
  

Step 5: Write effective descriptions

The description is critical for Claude to discover your Skill.

Formula: [What it does] + [When to use it] + [TorchRec keywords]

Examples:

✅ Good:

description: Optimize sharding plans for TorchRec embedding tables. Use when configuring DistributedModelParallel, analyzing sharding strategies, or tuning embedding performance.

✅ Good:

description: Validate KeyedJaggedTensor (KJT) configurations and debug sparse tensor issues. Use when working with KJT, debugging embedding lookups, or validating feature configurations.

❌ Too vague:

description: Helps with TorchRec
description: For distributed training

Step 6: Structure the Skill content

Use clear Markdown sections:

# Skill Name

Brief overview of what this Skill does for TorchRec.

## Quick start

Provide a simple example to get started immediately.

## Instructions

Step-by-step guidance for Claude:
1. First step with clear action
2. Second step with expected outcome
3. Handle edge cases

## TorchRec-Specific Patterns

Document TorchRec-specific patterns and conventions.

## Examples

Show concrete usage examples with TorchRec code.

## Best practices

- Key conventions to follow
- Common pitfalls to avoid
- When to use vs. not use

## Files to Reference

List important TorchRec files for context:
- `torchrec/distributed/` - Distributed training code
- `torchrec/modules/` - Core modules

Step 7: Validate the Skill

Check these requirements:

✅ File structure:

• [ ] SKILL.md exists in correct location
• [ ] Directory name matches frontmatter name

✅ YAML frontmatter:

• [ ] Opening --- on line 1
• [ ] Closing --- before content
• [ ] Valid YAML (no tabs, correct indentation)
• [ ] name follows naming rules
• [ ] description is specific and < 1024 chars

✅ Content quality:

• [ ] Clear instructions for Claude
• [ ] TorchRec-specific examples provided
• [ ] Edge cases handled
• [ ] References to relevant TorchRec code

TorchRec Skill Ideas

Here are some useful Skills to consider creating:

| Skill Name | Purpose | |------------|---------| | sharding-optimizer | Analyze and optimize embedding sharding plans | | kjt-validator | Validate KeyedJaggedTensor configurations | | distributed-debug | Debug distributed training issues | | embedding-benchmark | Benchmark embedding performance | | migration-helper | Help migrate to newer TorchRec APIs |

Example: Complete TorchRec Skill

---
name: sharding-analyzer
description: Analyze TorchRec sharding plans and suggest optimizations. Use when reviewing ShardingPlan, DistributedModelParallel configuration, or optimizing embedding distribution across devices.
---

# Sharding Analyzer

Analyze TorchRec sharding plans and suggest optimizations for embedding tables.

## Quick start

Run `/sharding-analyzer` on a file containing a ShardingPlan to get optimization suggestions.

## Instructions

1. Read the sharding plan configuration
2. Analyze table sizes and sharding strategies
3. Check for common anti-patterns:
   - Large tables with TABLE_WISE sharding
   - Small tables with ROW_WISE sharding
   - Unbalanced memory distribution
4. Suggest optimizations

## TorchRec Sharding Strategies

| Strategy | Best For | Avoid When |
|----------|----------|------------|
| TABLE_WISE | Small tables, < 1M rows | Large tables |
| ROW_WISE | Large tables, uniform access | Small tables |
| COLUMN_WISE | Wide embeddings, > 256 dim | Narrow embeddings |

## Files to Reference

- `torchrec/distributed/planner/` - Sharding planner
- `torchrec/distributed/sharding/` - Sharding implementations

Output format

When creating a Skill, I will:

1. Ask clarifying questions about scope and requirements
2. Suggest a Skill name and location
3. Create the SKILL.md file with proper frontmatter
4. Include TorchRec-specific instructions and examples
5. Add references to relevant TorchRec code
6. Provide validation checklist