Skill management

Reusable operating methods—triggers, procedures, and quality bars—that turn team know-how into agent-ready skills.

Marketing: exemplar.dev/marshal/skill-management .

When to use it

Encode how your team works (when to escalate, what to verify, what “done” means) once, then share across IDE agents, console, and frameworks—without editing IDE-local files for every change.

Where in Console

Author and version skills in Marshal console surfaces; promote with evals when risk is high. Align tone with prompt management.

Use from SDK and CLI

A skill is a folder rooted at SKILL.md (plus optional references/, scripts/, assets/). Prefer skills.install(dest) for runtimes; record.instructions is the markdown body only.


from exemplar_harness import Harness
 
skills = Harness.from_env().skills()
skills.create(
    name="refund-policy",
    instructions="# Refund policy\n\nReturns within 30 days.",
    description="Refund workflow",
    files={"references/policy.md": "# Policy\n\n30-day returns.\n"},
)
skills.install(".agents/skills", names=["refund-policy"])


exemplar skills list
exemplar skills pull refund-policy
exemplar skills install --all --dest .agents/skills
exemplar skills push ./skills/refund-policy

Full patterns: SDK client usage · CLI.

Framework / SDK examples

Install folders for agent runtimes


from exemplar_harness import Harness
 
skills = Harness.from_env().skills()
 
# Materialize all active skills (or pass names=[...])
skills.install(".agents/skills")
 
# IDE / product-specific roots
skills.install(".cursor/skills", names=["refund-policy"])
skills.install(".claude/skills", names=["refund-policy"])


import { Harness } from "@exemplar-dev/exemplar-harness-typescript-sdk";
 
const skills = Harness.fromEnv().skills();
await skills.install(".agents/skills", { names: ["refund-policy"] });
await skills.install(".cursor/skills", { names: ["refund-policy"] });

Agno — pass skill instructions


pip install "exemplar-harness-sdk[agno]"


from agno.agent import Agent
from agno.models.openai import OpenAIChat
from exemplar_harness import Harness
 
harness = Harness.from_env()
skills = harness.skills()
skills.install(".agents/skills", names=["refund-policy"])
 
skill = skills.get("refund-policy")
agent = Agent(
    name="support-bot",
    model=OpenAIChat(id="gpt-4o-mini"),
    instructions=[skill.instructions],  # markdown body; folder install for Agent Skills–aware runtimes
)
agent.run("Can a customer return an item after 20 days?")

OpenAI Agents / chat — inject skill text


from openai import OpenAI
from exemplar_harness import Harness
 
harness = Harness.from_env()
skill = harness.skills().get("refund-policy")
 
client = OpenAI()
client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": skill.instructions},
        {"role": "user", "content": "Can a customer return an item after 20 days?"},
    ],
)


import OpenAI from "openai";
import { Harness } from "@exemplar-dev/exemplar-harness-typescript-sdk";
 
const harness = Harness.fromEnv();
const skill = await harness.skills().get("refund-policy");
const client = new OpenAI();
await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [
    { role: "system", content: skill.instructions },
    { role: "user", content: "Can a customer return an item after 20 days?" },
  ],
});

Claude Agent SDK / CrewAI / Deep Agents

Folder-first install, then point the runtime at the dest (or pass instructions where the framework has no folder loader):

Runtime	Pattern
Claude Agent SDK	`skills.install(".claude/skills", names=[...])` then run Claude with project skills
CrewAI	`Agent(..., skills=[...])` after install — see crewai_skills_demo
Deep Agents	`create_deep_agent(..., skills=[...])` — see deepagents_skills_demo
Google ADK	install folder + ADK skill loader — see google_adk_skills_demo

Runnable sample: platform/skills.py · catalog Live SDK examples.

Bootstrap IDE skills from GitHub


npx github:Exemplar-Dev/exemplar-skills

Guide: Install IDE agent skills.

Related: Prompt management · Evals · Client usage.