Skills Framework
Relevant source files
The following files were used as context for generating this wiki page:
The Skills Framework is the canonical workflow and knowledge surface for the Everything Claude Code (ECC) ecosystem. Skills are modular, versioned knowledge units that provide the AI with domain-specific patterns, best practices, and actionable workflows. They are designed to be loaded dynamically based on the current context of the development session to ensure the AI follows project-specific or language-specific conventions without bloating the global context window.
The framework emphasizes an "Agent-First" and "Skill-First" approach, where skills/ is the primary location for new workflow contributions. While legacy slash commands exist, the long-term architectural direction prioritizes skills as the primary interface for domain knowledge and automated behaviors.
Sources: AGENTS.md:7-14, AGENTS.md:92-95, CONTRIBUTING.md:43-45
Skill Architecture and Structure
A skill is defined as a directory within the skills/ root containing a mandatory SKILL.md file. This file uses YAML frontmatter to define its identity and behavioral triggers.
Standard File Structure
skills/
└── your-skill-name/
├── SKILL.md # Core logic and frontmatter
└── references/ # (Optional) Detailed documentation
└── summary.md
Sources: CONTRIBUTING.md:49-55, skills/coding-standards/SKILL.md:1-20
Skill Metadata (Frontmatter)
All skills must declare specific frontmatter fields to be valid within the ECC ecosystem.
| Field | Requirement | Description |
|---|---|---|
name | Mandatory | Must match the directory name exactly. |
description | Mandatory | Concise string (under 130 chars) used for auto-activation and help menus. |
origin | Mandatory | Source of the skill (e.g., ECC, auto-extracted). |
version | Optional | Semantic versioning for the skill logic. |
Sources: scripts/ci/validate-skills.js:145-165, commands/skill-create.md:59-65, CONTRIBUTING.md:102-103
The Lifecycle of a Skill
The following diagram illustrates how skills are created, validated, and utilized by the AI.
The framework utilizes automated scripts to ensure structural integrity, such as checking for non-empty files and valid YAML scalar formats for descriptions to prevent breaking flat-table renderers.
Sources: scripts/ci/validate-skills.js:15-28, commands/skill-create.md:20-25
Core Framework Components
1. Skill Generation (/skill-create)
The /skill-create command automates the extraction of coding patterns from local git history. It identifies recurring workflows, naming conventions, and architecture patterns to generate new SKILL.md files.
- Gather Data: Analyzes commit frequency, file co-changes, and message patterns.
- Generate Instincts: Optionally creates instincts for the
continuous-learning-v2system, which can later evolve into full skills.
Sources: commands/skill-create.md:7-40
2. Validation and CI
To maintain quality, the framework employs a validation suite (validate-skills.js) that enforces strict rules:
- Directories must contain a
SKILL.md. - Frontmatter must not use literal block scalars (
|) for descriptions as they break external parsers. - Names must be lowercase and hyphenated.
Sources: scripts/ci/validate-skills.js:90-110, CONTRIBUTING.md:258-260
3. Inventory and Auditing (/skill-stocktake)
The stocktake system audits existing skills to prevent "context rot" and redundancy. It categorizes skills into global (~/.claude/skills/) and project-level (.claude/skills/) scopes.
Categorization of Skills
Skills are organized into logical domains to help the AI's selection process.
| Category | Description | Examples |
|---|---|---|
| Language Standards | Idioms and best practices for specific languages. | coding-standards, golang-patterns |
| Framework Patterns | Deep dives into specific technology stacks. | django-patterns, react-patterns |
| Workflow | Step-by-step processes for development tasks. | tdd-workflow, agentic-engineering |
| Domain Knowledge | Specialized knowledge like security or API design. | security-review, api-design |
Sources: CONTRIBUTING.md:77-85, skills/coding-standards/SKILL.md:1-5
Workflow Integration
When a skill is activated, it typically follows a three-step internal workflow:
- Scope Confirmation: Understanding the specific request within the context of the skill's domain.
- Canonical Execution: Running the specific steps defined in the skill (e.g., TDD Red-Green-Refactor).
- Reporting: Providing the results and suggesting subsequent steps.
Sources: skills/coding-standards/SKILL.md:17-20, AGENTS.md:83-87
Conclusion
The Skills Framework provides a structured, scalable way to manage AI knowledge within the Everything Claude Code project. By enforcing strict validation and providing automated tools for creation and auditing, it ensures that the AI remains an accurate and efficient partner in the software development lifecycle.