abstract

Tools for building and evaluating Claude Code skills. Includes modular patterns, quality checks, and plugin validation.

Quick Start

make check          # Install dependencies
make test           # Run quality checks
make install-hooks  # Set up git hooks

Installation

Add to marketplace.json:

{
  "name": "abstract",
  "source": { "source": "url", "url": "https://github.com/athola/abstract.git" },
  "description": "Meta-skills infrastructure - modular design and evaluation",
  "version": "1.4.4",
  "strict": true
}

Claude loads the plugin on startup.

What's Included

Abstract provides skills, commands, and agents for plugin development. methodology-curator surfaces expert frameworks, while modular-skills provides architectural guidance. Use skills-eval to score and improve skill quality.

The /validate-plugin command checks structures against requirements. Specialized agents like plugin-validator and meta-architect assist during review.

Structure

• skills/: Skill implementations.
• scripts/: Validation and analysis tools.
• src/abstract/: Shared Python package.
• shared-modules/: Reusable enforcement patterns for cross-skill reference.
  • iron-law-interlock.md: Hard gate for TDD compliance in creation workflows.
  • enforcement-language.md: Language intensity calibration.
  • anti-rationalization.md: Bypass prevention patterns.
  • trigger-patterns.md: Skill trigger design patterns.
• docs/: Technical documentation, ADRs, and examples.
  • docs/examples/modular-skills/: Implementation examples for modular skill patterns.

Documentation

• Skill Observability & Continual Learning: ../../docs/guides/skill-observability-guide.md - zero-dependency skill tracking with PreToolUse/PostToolUse hooks and stability gap detection
• Skill Assurance Framework: docs/skill-assurance-framework.md - patterns for reliable skill discovery (frontmatter-only triggers, enforcement language, migration guide)
• Migration Guide: docs/migration-guide.md - updating skills to new patterns
• Python Structure: docs/python-structure.md - package organization
• ADRs: docs/adr/ - architecture decisions
• Multi-Plugin Design: docs/multi-plugin-design.md - composition model

Each skill has its own SKILL.md with usage details. Run make status for a project overview.

Security

The CI pipeline runs Bandit, Safety, and Semgrep on each push. Pre-commit hooks catch issues locally.

make security   # Run security scans locally

Development

make format        # Format code
make test          # Run all checks
make security      # Security scans
make clean         # Clean cache
make unit-tests    # Run tests
make test-coverage # Coverage report

Tests validate skill discoverability and structure.

Stewardship

Ways to leave this plugin better than you found it:

• An opportunity to improve skill evaluation rubrics with concrete before/after examples from real sessions
• Hook stability tests could cover more edge cases, especially around Python 3.9 compatibility boundaries
• Shared modules like iron-law-interlock.md would benefit from inline examples showing correct and incorrect usage
• Skill trigger patterns in trigger-patterns.md have room for additional real-world discovery scenarios

See STEWARDSHIP.md for the full stewardship principles guiding this project.