name: sprint-planner
description: Manages Linear project workflows. Use when user mentions "sprint" or "create tasks".
**Progressive Disclosure** in three levels minimizes token usage:
1. **Frontmatter** — always in system prompt (decides when to trigger)
2. **SKILL.md body** — loaded when relevant (actual instructions)
3. **Linked files** — explored only when needed (detailed references)
## What Soul Spec Does
Soul Spec defines an agent's **identity**:
my-agent/
├── soul.json # Metadata (name, description, tags)
├── SOUL.md # Personality, tone, principles
├── IDENTITY.md # Basic information
└── USER.md # User context
If Skills answer "how to do it," Soul Spec answers "who does it."
## Comparison
| | Skills (SKILL.md) | Soul Spec (soul.json) |
|---|---|---|
| **Purpose** | Workflow knowledge | Persona & identity |
| **Core question** | "How to do it?" | "Who am I?" |
| **Trigger** | On user request | Always active |
| **Multiple** | Many skills at once | One persona |
| **MCP** | Direct support | Indirect (via skills) |
| **Standard** | Anthropic-proprietary | Open spec (LLM-agnostic) |
## Why You Need Both
Anthropic describes Skills with a **kitchen analogy**:
- MCP = professional kitchen (tools, ingredients, equipment)
- Skills = recipes (step-by-step instructions)
Add Soul Spec to complete the picture:
- **Soul = the chef** (experience, style, philosophy)
Same recipe, different chef, different experience. A customer onboarding workflow executed by "friendly, thorough Brad" feels different from "fast, efficient Kira."
## Notable Points from the Guide
**1. Skills API**
- `/v1/skills` endpoint for programmatic management
- `container.skills` parameter in Messages API
- Agent SDK integration
**2. Organization-wide Deployment**
- Admins can deploy skills workspace-wide (shipped Dec 2025)
- Automatic updates, centralized management
**3. Five Patterns**
- Sequential Workflow Orchestration
- Multi-MCP Coordination
- Iterative Refinement
- Context-aware Tool Selection
- Domain-specific Intelligence
**4. Open Standard Declaration**
> "We've published Agent Skills as an open standard. Like MCP, we believe skills should be portable across tools and platforms."
Skills aims for open portability — the same direction Soul Spec has pursued from day one.
## In Practice: Skills + Soul Spec
my-agent/
├── soul.json # Agent identity
├── SOUL.md # Personality and principles
├── IDENTITY.md # Basic info
├── skills/
│ ├── sprint-planner/
│ │ └── SKILL.md # Sprint planning workflow
│ └── code-review/
│ └── SKILL.md # Code review workflow
The Soul defines *who*. Skills define *what*. Together, they form the **complete agent package**.
## What This Means
Anthropic formalizing Skills into a 33-page guide signals a maturing agent ecosystem:
- **MCP** → how agents connect to the world (2024)
- **Skills** → how agents work (2025-2026)
- **Soul Spec** → how agents exist
All three layers aim for open standards. They solve different problems. Not competition — completion.
---
*Explore Soul Spec at [clawsouls.ai](https://clawsouls.ai). A guide for using Skills alongside Soul Spec is coming soon.*
---
*Originally published at [blog.clawsouls.ai](https://blog.clawsouls.ai/posts/anthropic-skills-guide-soul-spec/)*
Top comments (0)