# 2026 Skills System ## Concept Public extraction of production-tested Claude Code skills from the private PARA vault into a standalone open-source repository ([github.com/ngpestelos/skills](https://github.com/ngpestelos/skills)). 27 skills extracted from a 256-skill private system, normalized to the [Agent Skills specification](https://agentskills.io/specification), and published for use across Claude Code, Cursor, VS Code Copilot, Gemini CLI, and other compatible agents. ## Design Decisions | Decision | Choice | Rationale | |----------|--------|-----------| | Extraction over fork | Extract individual skills, not the whole vault | Private vault has project-specific context (PARA, chaos-app) that doesn't generalize | | Agent Skills spec | Follow agentskills.io specification | Cross-agent compatibility (not locked to Claude Code) | | Symlink installer | `install.sh` creates symlinks to `~/.claude/skills/` | Skills stay in sync with `git pull`, no copy drift | | Five-step optimization before publish | Run five-step-optimizer on each skill before extraction | Private skills are verbose; public skills need to be token-efficient | | MIT license | MIT | Maximum adoption, no friction | ## Published Skills (27) ### By Category | Category | Count | Skills | |----------|-------|--------| | ActiveRecord & Database | 7 | query-optimization, query-performance, eager-loading, corrupted-data, idempotent-create, transaction-boundary, migration-safety | | Rails Controllers & Views | 4 | input-validation, transaction-render, memoization, testing-patterns | | Stimulus & Frontend | 3 | controller-integration, form-redirect, formdata-sync | | ActiveJob | 1 | design-patterns | | Testing & Quality | 5 | test-plan, security-patterns, adversarial-validation, dead-code, cyclomatic-complexity | | Process & Problem-Solving | 2 | root-cause-investigation, five-step-optimizer | | Communication | 1 | response-diversity | | Productivity | 1 | scanned-document-extraction | | Claude Code & Tooling | 3 | command-optimizer, hook-state-cascade, skill-decomposition | ### Extraction Pipeline 1. Identify skill with general applicability (not vault-specific) 2. Run `/five-step-optimizer` to compress (typical: 40-85% reduction) 3. Strip PARA-specific references, vault wikilinks, project context 4. Validate against Agent Skills spec (frontmatter, allowed-tools) 5. Add to repo with `check.sh` pre-commit validation 6. Update README marketplace table ## Repository - **Remote**: [github.com/ngpestelos/skills](https://github.com/ngpestelos/skills) - **Local**: `~/src/skills/` - **First commit**: 2026-03-13 (6 skills) - **Current**: 27 skills, 32 commits ## Related Documents - [[Idea Registry]] — A008 (info product teaching devs Claude Code as Second Brain OS) - [[Skill Decomposition Draft]] — draft article about the methodology - [[Recursive Self-Improvement Claude Code Draft]] — draft article about the broader system - [[20260315 Weekly Reflection]] — "first external knowledge product" assessment ## Status - **2026-03-13**: Public repo created with 6 initial skills. Symlink installer and pre-commit validation added. - **2026-03-14**: Expanded to 20 skills. Added individual plugin install support. - **2026-03-15**: Published to Obsidian Publish as portfolio piece. Referenced in weekly reflection as "first crack in the wall between internal capability and external visibility." - **2026-03-16**: 27 skills published. Referenced in GitLab Plan Knowledge cover letter. Optimized five-step-optimizer via its own methodology.