using-singularity

Singularity — Self-Evolving Skill Engine

You have singularity-claude, a system that creates, scores, repairs, and hardens Claude Code skills through recursive improvement cycles.

Decision Flow

When a task arrives:

1. Task arrives → Does a skill exist?
   • Yes → Execute it → Score with /singularity-score → Maturity auto-updates
     • Avg < 50 (2+ runs) → Suggest /singularity-repair
     • Avg ≥ 90, 5+ runs, hardened, edge cases → Suggest /singularity-crystallize
     • Otherwise → Keep using and scoring
   • No → Is this a recurring need?
     • Yes → Create with /singularity-create
     • No → Do manually

Available Commands

Command	Skill	Purpose
/singularity-create	singularity-claude:creating-skills	Build a new skill
/singularity-score	singularity-claude:scoring	Rate a skill execution 0-100
/singularity-review	singularity-claude:reviewing	Health check a skill
/singularity-repair	singularity-claude:repairing	Auto-fix a failing skill
/singularity-crystallize	singularity-claude:crystallizing	Lock a validated version
/singularity-dashboard	singularity-claude:dashboard	Overview of all managed skills

Capability Gap Detection

Watch for these signals during task execution — they indicate a new skill should be created:

1. Repetition across sessions — You're doing the same multi-step procedure you've done before
2. Task failure without skill coverage — No existing skill addresses this capability
3. Generalizable pattern — The procedure would apply beyond this specific task
4. Complex workflow — The task requires 5+ steps that could be encoded

When you detect a gap, suggest: "This looks like a capability gap. Want me to create a skill for this with /singularity-create?"

Data Locations

• Registry: ~/.claude/singularity/registry.json
• Scores: ~/.claude/singularity/scores/<skill-name>.json
• Telemetry: ~/.claude/singularity/telemetry/<skill-name>/
• Config: ~/.claude/singularity/config.json
• Created skills: ~/.claude/skills/<skill-name>/SKILL.md

Maturity Levels

Level	Criteria	Meaning
draft	< 3 executions	New, unproven
tested	3+ runs, avg >= 60	Working but not battle-tested
hardened	5+ runs, avg >= 80, edge cases handled	Reliable
crystallized	Locked via git tag	Production-grade, immutable