Skills System¶
The Skills System is the "always-on" intelligence layer of the Seer Marketing Agents. While Commands are explicit workflows you trigger manually (like /content-audit), Skills are behavior patterns that activate automatically based on what you are doing.
What Are Skills?¶
Skills define HOW Claude thinks, writes, and approaches tasks. They are modular chunks of knowledge and guidelines that are loaded on-demand to keep the conversation context efficient while ensuring high-quality output.
Key Characteristics:¶
- Auto-activated: Unlike commands, you don't usually type them. Claude decides to use them based on your intent.
- Context-Aware: They activate based on keywords in your prompt, the files you are editing, or the type of task you are performing.
- Modular: Each skill is a self-contained package of standards, methods, or reference materials.
- Token-Efficient: Only the relevant skills are loaded, saving context window space.
How Skills Work¶
The skills system relies on a layered activation engine provided by the core-dependencies plugin.
The Activation Flow¶
- UserPromptSubmit: Before Claude responds to your prompt, this hook fires.
- Fragment Scanning: The engine scans all installed plugins for
skill-rules-fragment.jsonfiles. - Pattern Matching: It matches your prompt against keywords and regex intent patterns, and checks the files you have open.
- Injection: If a match is found, the engine injects a subtle suggestion into the system prompt (e.g., "Consider using the writing-standards skill").
- Loading: Claude recognizes the context and loads the relevant skill content into its active memory.
Skill Architecture¶
Every skill follows a standardized file structure to ensure portability across OpenCode and Claude Code.
plugins/{division}/skills/{skill-name}/
├── SKILL.md # Main skill definition (Quick Reference)
├── skill-rules-fragment.json # Activation triggers and rules
└── resources/ # Detailed methodology and content
├── data-analysis.md
└── strategy.md
1. SKILL.md¶
The primary entry point for the skill. It contains YAML frontmatter and a high-level summary. It should be kept under 500 lines for quick processing.
2. skill-rules-fragment.json¶
Defines when the skill should activate.
{
"skill-name": {
"type": "domain",
"enforcement": "suggest",
"priority": "high",
"promptTriggers": {
"keywords": ["deliverable", "client", "report"],
"intentPatterns": ["(write|create|draft).*(report|deliverable)"]
},
"fileTriggers": {
"pathPatterns": ["**/deliverables/**"]
}
}
}
3. Resources Directory¶
Contains deep-dive documentation, examples, and complex methodologies that are too large for the main SKILL.md. These are loaded only when Claude needs more detail.
Skill Catalog¶
Core Skills (Shared)¶
These live in plugins/core-dependencies/ and are available to all agents.
| Skill | Purpose | Key Triggers |
|---|---|---|
| writing-standards | Ensures Seer voice and client-ready tone. | deliverable, report, email |
| quality-standards | Enforces QA gates and validation rules. | check, validate, edit |
| prompt-engineering | Optimizes prompts for better model performance. | prompt, instructions, agent |
Division-Specific Skills¶
These live within their respective division plugins (e.g., plugins/divisions/seo/).
| Division | Skill | Purpose |
|---|---|---|
| SEO | seo-methods |
SEO scoring, SERP analysis, and strategy. |
| Innovation | skill-seekers |
Tools for creating and optimizing new skills. |
| Innovation | mcp-expertise |
Patterns for building MCP servers. |
Creating Your Own Skills¶
Developers can create new skills by following these steps:
- Location: Place your skill in
plugins/divisions/{division}/skills/{skill-name}/. - Define Rules: Create a
skill-rules-fragment.jsonwith specific keywords and intent patterns. - Write the Skill: Create
SKILL.mdusing the standard template (see Plugin Authoring). - Add Detail: Put detailed methods in the
resources/folder. - Test Activation: Use the
/core-doctorcommand to verify that your skill triggers correctly on your target keywords.
Pro Tip
Keep your SKILL.md focused on How Claude should behave, and put What (data, references, long examples) into the resources/ directory.
Related Documentation¶
- System Architecture: How skills fit into the Hooks → Skills → Commands trifecta.
- Plugin Authoring: Detailed guide on building plugins and skills.
- Plugin Architecture Specification: The full technical spec.