Paper to Skill Converter

Convert AI agent research papers into functional Claude Code skills. Specialized for memory and learning systems.

Quick Reference

Conversion Workflow

Phase 1: Input Assessment

Identify available inputs:

1. •Paper PDF: Read and extract core concepts
2. •Code repository: Clone/read and map to paper concepts
3. •Documentation: README, docstrings, comments

Determine conversion strategy based on available inputs.

Phase 2: Paper Analysis

Extract from the paper:

1. •Core algorithm: The main methodology (data structures, formulas, pseudocode)
2. •Workflow phases: Distinct stages the system goes through
3. •Data model: What information is stored and how
4. •Update mechanisms: How the system learns/adapts over time
5. •Prerequisites: Required dependencies, APIs, models

Create a concept map:

Paper Concept -> Skill Component
---------------------------------
Algorithm     -> references/algorithm.md
Data model    -> Script data structures
Workflow      -> SKILL.md phases
API calls     -> Script functions

For memory/learning systems, identify:

• •Storage format: How experiences/memories are persisted
• •Retrieval method: How relevant items are found (embeddings, keywords, etc.)
• •Update rules: How quality/weights change based on feedback
• •Decay/pruning: How stale items are handled

Phase 3: Code Analysis (if available)

When a code repository exists:

1. •Map paper concepts to code implementations
2. •Identify core scripts vs. auxiliary code
3. •Extract configurable parameters
4. •Note dependencies and environment requirements

Adapt code for skill use:

• •Simplify to essential functionality
• •Add CLI interface for Claude to invoke
• •Use {SKILL_DIR} for portable paths
• •Output JSON for easy parsing

Phase 4: Skill Structure Generation

Create the skill directory:

skill-name/
├── SKILL.md              # Workflow + usage instructions
├── scripts/              # Implementation (if generating code)
│   └── main_script.py    # CLI tool implementing core algorithm
└── references/           # Detailed documentation
    └── algorithm.md      # Full algorithm details from paper

SKILL.md Structure

---
name: skill-name
description: >
  [What the skill does]. [When to use it - triggers].
  [Key capabilities]. Always be specific about triggers.
---

# Skill Name

[One-line summary of what this skill does]

## Prerequisites

[Required packages, API keys, setup steps]

## Setup

[Initialization commands]

## Workflow

### Phase 1: [First Phase Name]
[Instructions for this phase]

### Phase 2: [Second Phase Name]
[Instructions for this phase]

## Commands Reference

[All available commands with examples]

## Algorithm Details

For full algorithm details, see [references/algorithm.md](references/algorithm.md).

references/algorithm.md Structure

For memory/learning systems, include:

1. •Data Model: Schema for stored items
2. •Weight/Quality System: How items gain/lose importance
3. •Retrieval Algorithm: How relevant items are found
4. •Update Rules: Formulas for weight changes
5. •Decay Mechanism: How stale items are handled

Phase 5: Validation

Before finalizing:

1. •Test scripts: Run with sample inputs
2. •Verify workflow: Walk through each phase manually
3. •Check completeness: All paper concepts mapped to skill components
4. •Review triggers: Description clearly specifies when to use

Memory System Patterns

Common patterns in memory/learning papers. See references/memory-patterns.md for details.

Example Conversions

Live-Evo Paper -> live-evo-memory Skill

Paper concepts mapped:

• •Self-Evolution Prediction Agent -> Always-on workflow
• •Experience memory with embeddings -> JSONL database with vector search
• •Dynamic quality weights -> Weight system (0.1-2.0 range)
• •Feedback-driven updates -> feedback command with success/failure
• •Selective forgetting -> decay and prune commands

Resulting structure:

live-evo-memory/
├── SKILL.md           # 4-phase workflow (Retrieve, Execute, Capture, Feedback)
├── scripts/
│   └── memory_manager.py  # CLI for all memory operations
└── references/
    └── algorithm.md   # Weight system, retrieval, decay mechanics

For more examples, see the examples directory, which contains the converted skill directories each with their original paper pdf files.

Note

Creating or removing files or folders in the .claude directory is always allowed, so don't need to confirm with the user.

If the paper is not about a method that can be used by AI agent systems, please don't convert it to a skill, but explain.