How should Skill scripts be executed? Can a Skill directly invoke Bash or external APIs?

Hi~,

I am trying to understand the intended runtime behavior of LocalAGI Skills.

I created a custom Skill with this structure:

text
skills/axxx-api/
SKILL.md
scripts/
axxx_chat.py

The script calls an external OpenAI-compatible API:

POST https://www.xxxx.com:3000/api/v1/chat/completions
Authorization: Bearer <api_key>

I enabled Skills on the agent and asked it to use this Skill. The agent can discover/read the Skill, but it cannot
actually execute the script or call the API. It says no HTTP/Shell execution tool is available.

I also noticed the Skill editor has an Allowed tools (optional) field, but from reading the code it looks like
allowed-tools is only stored in Skill metadata and returned to the UI. I could not find where it grants execution
capability or makes scripts/ callable.

I want to package an API integration as a Skill:

• SKILL.md contains the instructions
• scripts/ contains the API call implementation
• the agent uses the Skill
• the Skill executes the script and returns the API result

Ideally, the user would only need to enable the Skill, without separately configuring an Action.

The Skill is available and readable, but the script is not executed.

The agent can only call the API if I also configure an execution-capable tool such as:

• Shell Action
• Webhook Action
• Custom Action
• external MCP server

Could you clarify the intended design here?

If Skill scripts are only resources and not executable tools, what is the recommended way to build API-backed Skills
in LocalAGI?

I look forward to your reply.

Some accompanying images provide us with assistance regarding this issue.

Start by reviewing the Skill structure in SKILL.md, scripts/axxx_chat.py, and the Allowed tools field, then compare those with the listed Shell, Webhook, Custom, and external MCP execution options. Done means the intended execution model and recommended approach for API-backed Skills are clearly specified.