Manage platform and local skills

skill manages skills and spaces on AgentKit. skills generates wrappers, uploads and downloads packages, and publishes paired real and secure skills

Command overview

Command	Description
`skill list`	List platform skills.
`skill show`	Show skill details.
`skill versions`	List skill versions.
`skill spaces`	List skill spaces.
`skill delete`	Delete a platform skill.
`skills secure`	Generate secure Skill wrappers from a local skills directory.
`skills upload`	Upload local or GitHub skill packages
`skills download`	Download a skill version
`skills download-space`	Download a skill space
`skills publish-secure`	Publish paired real and secure skills

skill list

List skills.

Flag / Argument	Description	Default
`-r, --region <region>`	Region for the selected cloud provider	Current cloud environment
`-p, --project <name>`	Project name	default
`--json`	Output raw JSON	`false`

agentkit skill list

skill show

Show a skill’s details.

Flag / Argument	Description	Default
`<id>`	Skill id (required)	None
`-r, --region <region>`	Region for the selected cloud provider	Current cloud environment
`--json`	Output raw JSON	`false`

agentkit skill show sk-xxxxxxxx

skill versions

List a skill’s versions.

Flag / Argument	Description	Default
`<id>`	Skill id (required)	None
`-r, --region <region>`	Region for the selected cloud provider	Current cloud environment
`--json`	Output raw JSON	`false`

agentkit skill versions sk-xxxxxxxx

skill spaces

List skill spaces.

Flag / Argument	Description	Default
`-r, --region <region>`	Region for the selected cloud provider	Current cloud environment
`--json`	Output raw JSON	`false`

agentkit skill spaces

skill delete

Delete a skill.

Flag / Argument	Description	Default
`<id>`	Skill id (required)	None
`-r, --region <region>`	Region for the selected cloud provider	Current cloud environment
`-y, --yes`	Skip the confirmation prompt	`false`

Before deletion, use skill show to verify the ID and confirm existing spaces and agents no longer need the skill. Deletion affects subsequent use, and --yes skips confirmation

agentkit skill delete sk-xxxxxxxx --yes

skills secure

Generate secure Skill wrappers in bulk from a local skills directory. The source directory should contain one or more child directories, each with a SKILL.md whose frontmatter includes at least name and description. The command creates a same-name output directory for each skill and keeps only the invocation instructions needed by a managed agent; it does not copy the real Skill prompt, references, scripts, dependencies, or credentials.

Flag / Argument	Description	Default
`<sourceSkillsDir>`	Local skills root directory; each skill subdirectory should contain `SKILL.md`.	Required
`-o, --output <dir>`	Output directory.	Sibling `<sourceSkillsDir>-secure`
`--force`	Overwrite existing generated files.	`false`
`--json`	Output raw JSON.	`false`

For example, create skills/report and save this complete minimal skill file

skills/report/SKILL.md

---
name: report
description: Summarize supplied project notes
---
Read the supplied notes and return a summary of completed work and open questions.

The following commands generate local wrappers without uploading skills or creating spaces. See skills publish-secure for binding the real skills to a sandbox

agentkit skills secure ./skills

agentkit skills secure ./skills --output ./skills-secure --force

The generated wrappers are intended for managed agents that only need to submit a task, poll progress, and show the result. The real Skill still runs inside the AgentKit Skill Sandbox.

skills upload

Upload skill packages to the selected platform. Each SKILL.md must contain name and description

Uploading sends skill instructions, scripts, and resources to the destination and may create TOS storage resources. Review the directory for .env files, credentials, and other files that should not be uploaded

Flag / argument	Description	Default
`<skillsDir>`	Directory with SKILL.md, a parent of skill directories, or a GitHub directory URL	Required
`--source <source>`	Upload destination: `ark_ma` or `agentkit`	`SKILL_SOURCE` → `ark_ma`
`--api-key <key>`	Ark API key for `ark_ma`	`ARK_API_KEY`
`--endpoint <url>`	Ark skill upload endpoint	`https://ark.cn-beijing.volces.com/api/v3/skills`
`--skill-space-id <ids>`	AgentKit skill space IDs, comma-separated	`SKILL_SPACE_ID`
`-r, --region <region>`	AgentKit region	Provider / configuration
`-p, --project <name>`	TOS project for AgentKit uploads	`default`
`--display-title <title>`	Display title for a single Ark skill	—
`--dry-run`	Validate and show the upload plan without sending files	`false`
`--json`	Output JSON results	`false`

agentkit skills upload ./skills --source agentkit --skill-space-id ss-example --dry-run
agentkit skills upload ./skills --source agentkit --skill-space-id ss-example
agentkit skills upload ./skills/report --source ark_ma

When no AgentKit space is specified, interactive terminals offer a space selector. Non-interactive environments require --skill-space-id or SKILL_SPACE_ID. Dry runs do not require a space or create remote resources. Each skill is limited to 500 files, 25 MiB per file, and 50 MiB for both total file size and the ZIP package GitHub sources use https://github.com/<owner>/<repo>/tree/<ref>/<path> and require Git. Ark uses ARK_API_KEY; AgentKit uses cloud credentials for the selected Volcengine or BytePlus account

skills download

Download a skill package and extract it locally

Extraction overwrites files with matching names without confirmation. Use a new output directory

Flag / argument	Description	Default
`<id>`	Skill ID	Required
`-o, --output <dir>`	Extraction directory	`./<id>`
`--source <source>`	Only `agentkit` is supported	`SKILL_SOURCE` → `agentkit`
`--version <version>`	Version to download	Current platform version
`-r, --region <region>`	AgentKit region	Provider / configuration
`-p, --project <name>`	AgentKit project	—
`--json`	Output JSON results	`false`

agentkit skills download s-example --output ./downloaded-skill --source agentkit

Requires local unzip. ZIP packages are limited to 50 MiB and unsafe extraction paths are rejected. Downloads from ark_ma are not supported

skills download-space

Download every skill at the version bound to the space, using a subdirectory for each skill ID

Extraction overwrites files with matching names without confirmation. Use a new output directory

Flag / argument	Description	Default
`<skillSpaceId>`	Skill space ID	Required
`-o, --output <dir>`	Extraction directory	`./<skillSpaceId>`
`--source <source>`	Only `agentkit` is supported	`SKILL_SOURCE` → `agentkit`
`-r, --region <region>`	AgentKit region	Provider / configuration
`-p, --project <name>`	AgentKit project	—
`--json`	Output JSON results	`false`

agentkit skills download-space ss-example --output ./downloaded-space --source agentkit

Requires local unzip. ZIP packages are limited to 50 MiB and unsafe extraction paths are rejected. Downloads from ark_ma are not supported

skills publish-secure

Publish real skills and portable secure wrappers into two newly created skill spaces, then read both spaces to verify the pairs. Wrappers contain names, descriptions, and remote invocation scripts without real skill instructions or resources

Publication creates two spaces and uploads packages. The real space contains the full skills; configure access to each space appropriately. Publication can partially succeed. Inspect the returned space IDs and per-item results before retrying; another invocation creates new spaces

Flag / argument	Description	Default
`<skillsDir>`	Local skill directory or GitHub directory URL	Required
`--space-name-prefix <name>`	Name prefix for the two new spaces	Derived from source directory
`-o, --output-dir <dir>`	Retain generated secure skills locally	Not retained
`-r, --region <region>`	AgentKit region	Provider / configuration
`-p, --project <name>`	AgentKit project	`default`
`--force`	Overwrite generated files in the output directory	`false`
`--dry-run`	Validate and show the plan without creating spaces or uploading skills	`false`
`-y, --yes`	Skip publication confirmation; required non-interactively	`false`
`--json`	Output JSON results	`false`

agentkit skills publish-secure ./skills --dry-run
agentkit skills publish-secure ./skills --space-name-prefix reports --output-dir ./secure-skills --yes

Space names are <prefix>_real_<suffix> and <prefix>_secure_<suffix>, with a random suffix. Interactive terminals show a plan and request confirmation. JSON output and CI require explicit --yes Before executing a wrapper, create or select a Skill sandbox tool bound to the real space. Install agentkit-sdk-python>=0.8,<0.9 in the client runtime and provide credentials through the SDK default credential chain. Set AGENTKIT_TOOL_ID to that tool ID. AGENTKIT_REGION defaults to cn-beijing; set it explicitly for other regions

Getting Started

AI-Friendly

Command reference

Manage platform and local skills

Command overview

skill list

skill show

skill versions

skill spaces

skill delete

skills secure

skills upload

skills download

skills download-space

skills publish-secure

​Command overview

​skill list

​skill show

​skill versions

​skill spaces

​skill delete

​skills secure

​skills upload

​skills download

​skills download-space

​skills publish-secure

Command overview

skill list

skill show

skill versions

skill spaces

skill delete

skills secure

skills upload

skills download

skills download-space

skills publish-secure