Skip to main content
Skills give agents structured, domain-specific behavior without requiring you to wire prompts and tools together every time. Each skill is a directory-based package containing a SKILL.md instruction file, optional Python tool scripts, and metadata — all discovered and loaded automatically by SkillLoader.

​
What a skill contains

Every skill is a directory with the following layout: 
my-skill/
├── SKILL.md          # Instructions + YAML frontmatter metadata
├── scripts/          # Python files exposing callable tool functions
├── resources/        # Templates, reference assets
└── examples/         # Reference implementations
Only SKILL.md is required. The scripts/ directory is optional but necessary if the skill needs to register executable tools.

​
The three layers of a skill

Instructions

Markdown content in SKILL.md that is injected into the agent’s system prompt inside <skills> tags, guiding model behavior for this domain.

Tools

Python functions in scripts/*.py with docstrings and type hints. SkillLoader imports them dynamically and registers their schemas and callables with the agent.

Metadata

YAML frontmatter in SKILL.md parsed into a SkillMetadata object: name, description, version, author, tags, and requires.

​
Built-in skills

Logicore ships two default skills in logicore/skills/defaults/. They are available by name in any agent without any additional setup.

web_research

Structured web research with multi-source verification. Guides the agent through query analysis, search, source cross-referencing, and synthesis with citations.Tags: research, web, search, analysis

code_review

Automated code review covering bugs, security vulnerabilities, performance issues, and code quality. Outputs structured findings rated by severity.Tags: code, review, security, quality

​
Skills vs. tools

ToolsSkills
UnitSingle callable functionBundle of instructions + multiple tools
Prompt impactNoneInjects domain instructions into system prompt
DiscoveryRegistered manuallyAuto-discovered from filesystem
ReusePer-agent wiringLoad by name across any agent
MetadataNonename, version, author, tags, requires
Skills wrap tools. When a skill is loaded, each function discovered in its scripts/ directory is registered as a tool on the agent, alongside the skill’s instructions.

​
Architecture

​
Loading flow

When an agent loads a skill, the following sequence runs:

​
Quick start

from logicore.agents.agent import Agent

agent = Agent(
    llm="ollama",
    tools=True,
    skills=["web_research"]
)

response = await agent.chat("Find and summarize recent updates on AI safety")
print(response)
Use small, focused skills per domain capability. A skill named web_research that does one thing well is easier to maintain and combine than a broad skill trying to cover many domains.

​
Why skills matter

  • Consistency — the same instructions and tool bundle are reused across projects and team members, eliminating prompt drift.
  • Faster setup — load a named skill instead of wiring tools and prompts individually every time.
  • Better quality — domain-specific guidance in SKILL.md helps the model follow the correct workflow for a given task.
  • Governance — SkillMetadata fields (name, version, author, tags, requires) make capabilities auditable and versioned.
  • Composability — multiple focused skills can be combined on a single agent for well-rounded capability.

​
Next steps

Use skills in agents

Load built-in and custom skills at agent creation or dynamically at runtime.

Build custom skills

Create your own SKILL.md package with custom tool functions and register it in your workspace.

Build docs developers (and LLMs) love

Get started for freeTalk to us