Singularity-Claude: Add Self-Evolving Skills to Your Claude Code Workflow

Singularity-Claude: Add Self-Evolving Skills to Your Claude Code Workflow

Install this plugin to make your Claude Code skills automatically improve through scoring, repair, and crystallization loops—no manual maintenance needed.

9h ago·3 min read·2 views·via hn_claude_code, hn_claude_cli, reddit_claude
Share:

Singularity-Claude: Add Self-Evolving Skills to Your Claude Code Workflow

What It Does — A Recursive Evolution Engine for Skills

Claude Code skills are static. You write them once, and they stay exactly as they are—even when they fail, produce subpar output, or encounter new edge cases. There's no feedback loop, no way to know which skills are working well, and no mechanism to improve them over time.

singularity-claude solves this by adding a recursive evolution loop to your skills. It's a self-evolving skill engine that creates, scores, repairs, and crystallizes skills through autonomous recursive improvement loops. Every skill execution gets scored, low scores trigger automatic repair, and high scores lead to crystallization—locked, battle-tested versions you can trust.

Setup — Install in Under 60 Seconds

No external dependencies. No SmythOS, no OpenTelemetry collector, no Docker. Pure Claude Code.

# Add the marketplace
claude plugin marketplace add shmayro/singularity-claude

# Install the plugin
claude plugin install singularity-claude

Or install from source:

git clone https://github.com/shmayro/singularity-claude.git
cd singularity-claude
claude plugin marketplace add .
claude plugin install singularity-claude

Once installed, singularity loads automatically in every new Claude Code session.

How It Works — The Four Commands That Change Everything

1. /singularity-create — Generate Skills from Patterns

When you notice a recurring pattern but don't have a skill for it yet:

/singularity-create

This command gathers requirements, generates a SKILL.md file, registers the skill, and initializes scoring. It turns your manual workflows into reusable, improvable skills.

2. /singularity-score — Rate Every Execution

After every skill execution:

/singularity-score

This dispatches an assessor agent that rates the skill across 5 dimensions (0-100) and auto-updates maturity levels: draft → tested → hardened.

3. /singularity-repair — Fix Broken Skills Automatically

When a skill scores below 50 average across 2+ runs:

/singularity-repair

The system analyzes telemetry, rewrites weak dimensions, bumps the version, and re-tests the skill. No more manually debugging broken skills.

4. /singularity-crystallize — Lock Battle-Tested Skills

When a skill averages ≥90 across 5+ runs and handles edge cases:

/singularity-crystallize

This creates a Git tag + locks the version, giving you production-grade skills that are immutable and rollback-safe.

When To Use It — Your Skill Quality Control System

Use singularity-claude when:

  • You have more than 10 skills and can't manually maintain them all
  • Skills fail silently and you only discover issues weeks later
  • Edge cases accumulate and your skills become brittle
  • You want to identify your most valuable skills through usage data
  • Team collaboration where skill quality varies between developers

The Evolution Loop in Practice

Here's what happens automatically:

  1. Execute a skill/singularity-score runs automatically
  2. Score < 50/singularity-repair triggers automatically
  3. Score ≥ 90/singularity-crystallize becomes available
  4. Every step is logged for full auditability

This means your skills get better with use, not worse. The more you use a skill, the more robust it becomes.

Start Today — Create Your First Evolving Skill

# 1. Start a new Claude Code session — singularity loads automatically
# 2. Create your first skill
/singularity-create

Describe a recurring task you do manually. The system will turn it into a skill that evolves with every use.

No more growing piles of skills where some work great, some are mediocre, and some are silently broken. With singularity-claude, every skill has a feedback loop, every execution contributes to improvement, and your best skills get locked in as production-grade assets.

AI Analysis

Claude Code users should immediately install singularity-claude and start using the four commands. The biggest workflow change is running `/singularity-score` after every skill execution—this creates the data needed for automatic improvement. Start by identifying 2-3 skills you use frequently but that sometimes fail or produce inconsistent results. Run them through the evolution loop: execute, score, and let the system repair if needed. Within a week, you'll have hardened versions of your most important skills. The key insight is that skills shouldn't be static artifacts. They should improve with usage. If you're not scoring your skills, you're missing the opportunity for automated quality improvement. This is especially valuable for teams where skill quality can vary between developers—now everyone benefits from collective improvements.
Original sourcegithub.com

Trending Now

More in Products & Launches

Browse more AI articles