overview.md

Skill System Overview

Skills are the interface between the AI agent and external security tools. Each skill wraps a tool with a uniform execute(params, context) → SkillResult interface.

Architecture

LLM Agent
    │
    ▼
SkillRegistry (manages all skills)
    │
    ├── Built-in Skills (11)  ← Python classes in skills.py
    └── Custom Skills (N)     ← Markdown files in ~/.argus-lite/skills/
         │
         ▼
    SkillResult { success, data, findings, summary }

How Skills Work

1. The LLM agent decides which skill to run based on current context
2. The agent sends JSON: {"action": "scan_nuclei", "input": {"target": "example.com"}}
3. SkillRegistry.execute() finds the skill and calls skill.execute(params, context)
4. The skill runs the underlying tool and returns structured results
5. Results feed back into the agent's context for the next decision

Skill Interface

Every skill implements:

class Skill(ABC):
    @property
    def name(self) -> str: ...       # e.g., "scan_nuclei"

    @property
    def description(self) -> str: ... # e.g., "Scan for known vulnerabilities"

    async def execute(self, params: dict, context: AgentContext) -> SkillResult: ...

    def is_available(self) -> bool: ... # True if underlying tool exists

Two Types of Skills

Built-in Skills

Python classes in src/argus_lite/core/skills.py. Each wraps a specific tool binary. See Built-in Skills.

Custom Markdown Skills

User-defined .md files with YAML frontmatter. Loaded from ~/.argus-lite/skills/ or --skills-dir. See Custom Skills.