Skills

Skills are reusable prompt packages that provide an agent with domain knowledge, procedures, and scripts. VeADK 1.0.1 recommends Google ADK’s SkillToolset for local and cloud skills. The existing Agent(skills=..., skills_mode="local") entry point remains available but is deprecated. The skills_sandbox and aio_sandbox modes are unchanged.

Use local skills

Load a skill directory with ADK’s load_skill_from_dir, then pass a SkillToolset to the VeADK agent:

from google.adk.skills import load_skill_from_dir
from google.adk.tools.skill_toolset import SkillToolset
from veadk import Agent

skill = load_skill_from_dir("/abs/path/to/skills/kb-skill")
agent = Agent(tools=[SkillToolset(skills=[skill])])

ADK loads the skill, provides its prompt to the model, and exposes the list_skills, load_skill, load_skill_resource, and run_skill_script tools. If a skill runs files under scripts/, configure a code executor on the SkillToolset or agent. VeADK does not create a local code executor automatically.

Skill scripts can access the files and runtime exposed by the code executor. Load only trusted skills and restrict the executor to the permissions they need.

Use cloud skills

VeADK 1.0.1 adds VeSkillRegistry, which exposes skills from SkillHub or a Skill Space through the ADK registry interface:

import os

from google.adk.tools.skill_toolset import SkillToolset
from veadk import Agent
from veadk.skills import VeSkillRegistry

registry = VeSkillRegistry(
    skill_source_id=os.environ["AGENTKIT_SKILL_SOURCE_ID"],
)
agent = Agent(tools=[SkillToolset(registry=registry)])

search_skills retrieves the remote skill list, while get_skill downloads and loads a specific version when requested. This approach does not download every skill at initialization.

`VeSkillRegistry` parameter	Type	Default	Description
`skill_source_id`	`str`	Required	One SkillHub or Skill Space identifier; an empty value or comma-separated identifiers are not accepted.
`cache_dir`	`pathlib.Path \| None`	`None`	Local cache for downloaded skill versions; `None` uses the default cache directory.

Skill directory

An ADK local skill has its own directory. It must contain SKILL.md and can also include references, assets, and scripts:

skills/
└── kb-skill/
    ├── SKILL.md
    ├── references/
    ├── assets/
    └── scripts/

The SKILL.md frontmatter must define name and description, and the directory name should match name:

skills/kb-skill/SKILL.md

---
name: kb-skill
description: Query the knowledge base and compose an answer.
---

Follow these steps to query the knowledge base and compose an answer...

Legacy entry-point compatibility

The legacy entry point supports these modes:

Mode	Description
`local`	Deprecated. VeADK loads and executes skills from a local skills root; migrate local use cases to ADK `SkillToolset`.
`skills_sandbox`	Loads skills from a cloud Skill Space and executes them in a sandbox through `execute_skills`.
`aio_sandbox`	Executes skills in an AgentKit All-in-one tool runtime.

If skills_mode is unset, VeADK selects a mode from the runtime environment: local runs use local; an AgentKit tool runtime uses skills_sandbox or aio_sandbox according to its tool type. Legacy local example:

from veadk import Agent

agent = Agent(
    skills=["/abs/path/to/skills"],
    skills_mode="local",
    enable_skills_checklist=True,
    enable_dynamic_load_skills=True,
)

Legacy sandbox example:

import os

from veadk import Agent
from veadk.tools.builtin_tools.execute_skills import execute_skills

agent = Agent(
    skills=[os.environ["AGENTKIT_SKILL_SPACE_ID"]],
    skills_mode="skills_sandbox",
    tools=[execute_skills],
)

Legacy parameters

Parameter	Type	Default	Description
`skills`	`list[str]`	`[]`	Local skills-root paths or cloud Skill Space identifiers. The local-loading entry point is deprecated.
`skills_mode`	`"local" \| "skills_sandbox" \| "aio_sandbox" \| None`	`None`	Execution mode; when `None`, VeADK selects it from the runtime environment.
`enable_skills_checklist`	`bool`	`False`	Tracks checklist items for skills loaded through the legacy entry point.
`enable_dynamic_load_skills`	`bool`	`False`	Checks and reloads legacy skills while the agent is running.

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Use local skills

Use cloud skills

Skill directory

Legacy entry-point compatibility

Legacy parameters

​Use local skills

​Use cloud skills

​Skill directory

​Legacy entry-point compatibility

​Legacy parameters

Use local skills

Use cloud skills

Skill directory

Legacy entry-point compatibility

Legacy parameters