Extensible Skills: Customization Without Modification
Making AI tools transparent, customizable, and extensible
The Extensible Skills documentation has been split into two documents for clarity:
Formal Specification (Normative)
The Extensible Skills Specification defines:
- Three extensibility categories -- extensible, semi-extensible, not-extensible
- Context loading specification -- Instruction-based skill memories, DCI syntax for other contexts, and graceful degradation
- Detection algorithm -- How categories are determined from SKILL.md content
- Conformance requirements -- What registries must implement
Implementation Guide (Informative)
The Implementation Guide covers:
- Getting started -- For Claude Code users and SpecWeave users
- Architecture -- Instruction-based loading,
.specweave/skill-memories/lookup, and the Reflect system - Real-world examples -- How corrections become persistent preferences
- Skill memory format -- Structured Markdown for customizations
- FAQ -- Common questions and troubleshooting
Quick Reference
| Category | Meaning | Detection |
|---|---|---|
| Extensible | Instruction or injection block referencing skill-memories. Standard, discoverable customization. | Instruction or injection block referencing skill-memories |
| Semi-Extensible | Mentions customization but not through the standard system. | Keyword signals without skill-memories reference |
| Not Extensible | No customization mechanism. Fork to change. | No signals detected |
See Also
- Skills Overview -- Both skill layers at a glance
- Claude Skills Deep Dive -- How skills work under the hood
- Self-Improving Skills -- The Reflect auto-learning system
- Development Guidelines -- Design principles for skill authoring
- Verified Skills Standard -- How skills earn trust through 3-tier security certification
Version: 4.0.0 Authors: Anton Abyzov License: MIT