Tools

Code & skills

Run code in an isolated sandbox, and load and run the skills an agent carries.

Tool	What it does
`code_execution`	Runs a short Python or Node script the model writes, in a sandbox with no network and CPU, memory and time limits.
`load_skill`	Reads a skill's instructions when the agent needs them: first the body of its SKILL.md and a list of its files, then any one file on demand.
`run_skill_script`	Runs one of a skill's declared entrypoint scripts in the sandbox, with arguments.
`propose_skill`	Drafts a reusable SKILL.md from the conversation as a proposal in the Skill Workshop.

`code_execution`

Runs a short Python or Node script the model writes, in a sandbox with no network and CPU, memory and time limits. Files written to /out come back as downloadable artifacts.

How an agent gets it. Declare it in the agent's tools: list.

numpy, pandas, matplotlib and pillow are available; package managers are not.
Files the run already produced (fetched images, generated images) are in /work, and a workspace's document templates in /work/templates/.
See Secure code execution.

Arguments

Argument	Type	Required	Meaning
`code`	string	yes	Program source to execute.
`language`	one of `python`, `node`

`load_skill`

Reads a skill's instructions when the agent needs them: first the body of its SKILL.md and a list of its files, then any one file on demand. Skills cost nothing until they are used.

How an agent gets it. Added automatically when the agent has at least one skill attached.

Arguments

Argument	Type	Required	Meaning
`slug`	string	yes
`path`	string		Optional bundled file to read (e.g. references/style.md).

`run_skill_script`

Runs one of a skill's declared entrypoint scripts in the sandbox, with arguments.

How an agent gets it. Added automatically when the agent has at least one skill attached.

Only entrypoints the skill declares can run, and only if every tool in the skill's allowed_tools is also granted to the agent.

Arguments

Argument	Type	Required	Meaning
`skill`	string	yes
`entrypoint`	string	yes	Declared entrypoint path.
`args`	array of string

`propose_skill`

Drafts a reusable SKILL.md from the conversation as a proposal in the Skill Workshop. It is never applied automatically; a person reviews, revises and applies it.

How an agent gets it. Added automatically on chat and automation runs.

Pass proposal_id to revise an existing draft.

Arguments

Argument	Type	Required
`title`	string	yes
`draft_md`	string	yes
`proposal_id`	string