Skip to main content
Skills can be loaded when constructing an agent or added later with load_skill / load_skills. The agent resolves string names against the package defaults and your workspace skill directories automatically.

​
Resolution order

When you pass a skill name as a string, the agent resolves it in this order:
  1. Package defaults — logicore/skills/defaults/
  2. Workspace skill directories (when workspace_root is set) — .agent/skills/, _agent/skills/, .agents/skills/, _agents/skills/

​
Loading at initialization

Pass skill names in the skills list when constructing an Agent: 
from logicore.agents.agent import Agent

agent = Agent(
    llm="ollama",
    tools=True,
    workspace_root="/path/to/project",
    skills=["web_research"]
)
Built-in skills like web_research and code_review resolve without a workspace_root: 
agent = Agent(
    llm="ollama",
    tools=True,
    skills=["code_review"]
)

​
Loading a skill object directly

Use SkillLoader.load() to load a skill from a directory path, then pass the resulting object to agent.load_skill(): 
from logicore.skills.loader import SkillLoader
from logicore.agents.agent import Agent

skill = SkillLoader.load("/path/to/project/.agent/skills/release_assistant")

agent = Agent(llm="ollama", tools=True)
if skill:
    agent.load_skill(skill)
SkillLoader.load() returns None if the directory does not contain a valid SKILL.md. Always check the return value before calling agent.load_skill().

​
Loading multiple skills

from logicore.agents.agent import Agent

agent = Agent(
    llm="ollama",
    tools=True,
    workspace_root="/path/to/project",
    skills=["web_research", "code_review"]
)

​
Using skills with different agent types

BasicAgent, SmartAgent, and MCPAgent all share the same underlying skill-loading flow from Agent. The skills constructor parameter and load_skill / load_skills methods work identically across all agent types. 
from logicore.agents.agent_mcp import MCPAgent

agent = MCPAgent(
    provider="ollama",
    mcp_config_path="mcp.json",
    workspace_root="/path/to/project",
    skills=["release_assistant"]
)

​
What happens after loading

When load_skill() runs, the agent does the following:
  1. Deduplication — if a skill with the same name is already loaded, it is skipped.
  2. Tool registration — each tool schema from Skill.tools is appended to self.internal_tools.
  3. Executor registration — each callable from Skill.tool_executors is stored in self.skill_tool_executors.
  4. Tool support flag — supports_tools is set to True if the skill contributes any tools.
  5. Prompt injection — _build_skills_prompt_section() formats all loaded skill instructions and _rebuild_system_prompt_with_tools() appends the block to the system prompt:
<skills>
### Skill: web_research
...
### Skill: code_review
...
</skills>
Once loaded, skill tools participate in the same tool-calling loop as any other registered tool.

​
Inspecting loaded skills

Access the list of currently loaded skills from the skills attribute on an agent instance: 
for skill in agent.skills:
    print(skill.name, "-", skill.description)
    print(f"  Tools: {len(skill.tools)}")
    print(f"  Dir:   {skill.skill_dir}")
The Skill object exposes:
The name field from SKILL.md frontmatter, also available via skill.metadata.name.
The description field from SKILL.md frontmatter, also available via skill.metadata.description.
A list of tool schema dicts in OpenAI function-calling format, one per discovered function in scripts/.
A dict mapping function names to their callables for runtime execution.
The raw instruction markdown from SKILL.md (everything after the frontmatter block).
A pathlib.Path pointing to the skill’s root directory, or None if built from an object directly.

​
Overriding or extending skill tools

To add tools on top of what a skill provides, register additional tools after loading the skill: 
from logicore.agents.agent import Agent

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

# Add a custom tool on top of the skill's built-in tools
def fetch_internal_wiki(query: str) -> str:
    """Search the internal company wiki for a query."""
    # implementation
    return "..."

agent.register_tool(
    name="fetch_internal_wiki",
    func=fetch_internal_wiki,
    description="Search the internal company wiki for a query."
)
Keep skills focused on a single domain. Combining a small number of complementary skills (e.g., web_research + code_review) produces better results than loading many broad skill packages at once.

​
Combining skills effectively

# Well-rounded research-to-report workflow
agent = Agent(
    llm="ollama",
    tools=True,
    workspace_root="/path/to/project",
    skills=[
        "web_research",      # Gather information
        "code_review",       # Analyse any code found
        "release_assistant"  # Custom workspace skill
    ]
)

response = await agent.chat(
    "Research the latest changes in Python 3.13 and review the code in src/"
)
Avoid loading skills with overlapping tool names. If two skills register a function with the same name, the second registration will overwrite the first executor in skill_tool_executors.

Build docs developers (and LLMs) love

Get started for freeTalk to us