OpenClaw/Claude Code skills for AI-assisted development.
AI assistants make the same mistakes repeatedly because they don't learn from failures. They can be told "don't do X" but have no memory to enforce it. Instructions alone aren't enough—they're forgotten between sessions, ignored under pressure, and easily bypassed through rephrasing.
Live Neon Skills provide failure-anchored learning—a system where AI mistakes become constraints that prevent recurrence. When a failure happens enough times (R≥3) and is verified by humans (C≥2), it automatically becomes an enforced rule.
Failure detected → Observation created → Human verified → Constraint generated → Runtime enforced
These skills make AI assistants self-correcting. See agentic/README.md for the full failure-to-constraint lifecycle.
ClawHub (recommended):
# Install individual skills via OpenClaw
openclaw install leegitw/context-verifier
openclaw install leegitw/failure-memory
openclaw install leegitw/constraint-engine
# See agentic/README.md for full installation order
# Skill-distiller uses neon- prefix (base slug was taken)
openclaw install neon-skill-distillerManual (Claude Code users):
# Clone to your Claude Code skills directory
git clone https://github.com/live-neon/skills.git ~/.claude/skills/liveneon
# Or symlink specific skills
ln -s /path/to/skills/skill-distiller ~/.claude/skills/skill-distillerWhich skill should I start with?
| Goal | Start With | Then Add |
|---|---|---|
| Extract principles from text | essence-distiller |
pattern-finder for comparison |
| Prevent AI from repeating mistakes | context-verifier |
failure-memory → constraint-engine |
| Full failure-to-constraint lifecycle | Install in order below | - |
Recommended installation order for agentic skills:
# 1. Foundation (no dependencies)
openclaw install leegitw/context-verifier
# 2. Core (builds on foundation)
openclaw install leegitw/failure-memory
openclaw install leegitw/constraint-engine
# 3. Extended (builds on core)
openclaw install leegitw/safety-checks
openclaw install leegitw/review-orchestrator
openclaw install leegitw/governance
openclaw install leegitw/workflow-toolsPBD skills work standalone — install any one to start extracting principles.
Skills for extracting and synthesizing invariant principles from text.
| Skill | Description |
|---|---|
| essence-distiller | User-friendly principle extraction |
| pbe-extractor | Technical principle extraction |
| pattern-finder | Compare two extractions |
| core-refinery | Synthesize N>=3 extractions |
| golden-master | Track source/derived provenance |
| principle-comparator | Compare principles |
| principle-synthesizer | Synthesize principles |
Skills for failure detection, constraint enforcement, and memory operations. See agentic/README.md for the full lifecycle.
| Skill | Alias | Layer | Description |
|---|---|---|---|
| failure-memory | /fm |
Core | Failure tracking, observations, pattern detection |
| constraint-engine | /ce |
Core | Constraint generation, enforcement, circuit breaker |
| context-verifier | /cv |
Foundation | File hashes, integrity verification |
| review-orchestrator | /ro |
Review | Twin and cognitive review coordination |
| governance | /gov |
Governance | Constraint lifecycle, state management |
| safety-checks | /sc |
Safety | Model pinning, fallbacks, session state |
| workflow-tools | /wt |
Extensions | Loop detection, parallel decisions, MCE |
Consolidated architecture — 7 skills (from 48 granular). See ARCHITECTURE.md for details.
Skills for operating on other skills.
| Skill | Tokens | Description |
|---|---|---|
| skill-distiller | ~400 | Default (formula notation, 89% functionality) |
| compressed | ~975 | Prose variant (88% functionality) |
| oneliner | ~100 | Quick reference (72% functionality) |
Full reference: skill-distiller/SKILL.reference.md (~2,500 tokens, 91%)
Install: openclaw install neon-skill-distiller
Manual:
ln -s /path/to/skills/skill-distiller ~/.claude/skills/skill-distillerSkills for transforming technical work into creative artifacts for reflection and knowledge transfer.
| Skill | Alias | Description |
|---|---|---|
| insight-song | /song |
Create original Suno-ready songs from insights |
| song-remix | /remix |
Transform existing songs (Twin Remix method) |
| visual-concept | /vc |
Transform insights into visual concept guides |
| ted-talk | /ted |
Transform insights into full TED talks |
| side-quests | /sq |
Combo: song + visual + TED talk |
After installation, invoke skills in Claude Code:
/essence-distiller [content]
/pbe-extractor [content]
See ARCHITECTURE.md for how skills work together, including:
- Skill layer diagram (Foundation → Core → Review/Detection → Governance/Safety)
- Failure → Constraint lifecycle
- ClawHub integration points
These skills integrate with ClawHub skills via shared workspace files:
| ClawHub Skill | Integration | What We Provide |
|---|---|---|
self-improving-agent |
Reads our learnings | .learnings/LEARNINGS.md, .learnings/ERRORS.md |
proactive-agent |
Reads our constraints | output/constraints/, SESSION-STATE.md |
VFM system |
Reads constraint metadata | output/constraints/metadata.json |
File format compatibility: See output/VERSION.md for version pinning.
Workspace structure:
.learnings/ # self-improving-agent format
├── LEARNINGS.md # [LRN-YYYYMMDD-XXX] corrections
├── ERRORS.md # [ERR-YYYYMMDD-XXX] failures
└── FEATURE_REQUESTS.md # [FEAT-YYYYMMDD-XXX] requests
output/
├── constraints/ # Constraint storage
├── context-packets/ # File hash packets
└── hooks/ # Hook execution logs
Technical guides for skill implementation:
| Guide | Description |
|---|---|
| Semantic Similarity | LLM-based action classification (required for safety-critical skills) |
| Core Memory Walkthrough | Core memory layer implementation guide |
| Workflow | Description |
|---|---|
| Creating a New Skill | Complete skill creation from validation to publication |
| Skill Publishing | Publishing workflow with security compliance |
| Documentation Update | Process for updating docs when skills/architecture change |
| Phase Completion | Checklist for completing implementation phases |
| Batch File Modification | Bulk file changes with verification |
Implementation history and specifications:
| Directory | Description |
|---|---|
| docs/ | Documentation index |
| docs/patterns/ | Validated patterns (N≥3 evidence) |
| docs/proposals/ | Specifications and proposals |
| docs/plans/ | Implementation plans (phases 1-7) |
| docs/research/ | External research (hooks, learning theory, industry patterns) |
| docs/issues/ | Review findings and issues |
| docs/reviews/ | Code and twin reviews |
| docs/implementation/ | Phase results |
Unified test infrastructure for all skills. See tests/README.md for full documentation.
cd tests && npm install && npm test| Command | Description |
|---|---|
npm test |
Run all tests (PBD + Agentic) |
npm run test:pbd |
Run PBD skill tests only |
npm run test:agentic |
Run Agentic skill tests only |
Docker-based testing with OpenClaw: see docker/README.md.
- Open methodology — Give away the "how"
- Provenance-first — Track where ideas come from
- Failure-anchored — Learn from consequences, not instructions
- Community-driven — Accept contributions
Want to add or update skills? See docs/workflows/skill-publish.md.
For optional ClawHub publishing, copy .env.example to .env and add your token.
See CONTRIBUTING.md for guidelines.
- NEON-SOUL - Soul synthesis using these PBD skills for principle extraction and compression.
- Obviously-Not/patent-skills - Proprietary patent-specific skills (code-patent-scanner, patent-validator). PBD skills were migrated from there to this open-source repo.
MIT License - see LICENSE
Built by twins in Alaska.
This is a submodule of live-neon, the umbrella repository for AI agent infrastructure.
| Related | Link |
|---|---|
| Architecture | docs/vision/README.md |
| automation | Uses skills in agentic workflows |
| garden | Validates skill-extracted patterns via N-count |
| go-pbd | PBD skills wrap the go-pbd CLI |