Skip to main content

ParsedSkillFrontmatter

@a16njs/models

@a16njs/models / ParsedSkillFrontmatter

Interface: ParsedSkillFrontmatter

Defined in: agentskills-io.ts:15

Parsed frontmatter from an AgentSkills.io SKILL.md file. This is the VERBATIM AgentSkills.io format, NOT the IR format.

Extends​

Properties​

name​

name: string

Defined in: agentskills-io.ts:17

Skill name (required)

description​

description: string

Defined in: agentskills-io.ts:19

Skill description for activation matching (required)

resources?​

optional resources?: string[]

Defined in: agentskills-io.ts:21

Resource file paths relative to skill directory (optional)

disableModelInvocation?​

optional disableModelInvocation?: boolean

Defined in: agentskills-io.ts:23

If true, only invoked via /name (optional)

license?​

optional license?: string

Defined in: types.ts:61

License name or a reference to a bundled license file. Provenance only.

Inherited from​

AgentSkillSpecFields.license

compatibility?​

optional compatibility?: string

Defined in: types.ts:63

Environment requirements, e.g. Requires Python 3.14+ and uv. Documentation only.

Inherited from​

AgentSkillSpecFields.compatibility

specMetadata?​

optional specMetadata?: Record<string, string>

Defined in: types.ts:71

The AgentSkills.io spec's metadata field: author-authored client properties, persisted to disk under the key metadata.

Distinct from AgentCustomization.metadata, which is transient and is never serialized.

Inherited from​

AgentSkillSpecFields.specMetadata

allowedTools?​

optional allowedTools?: string

Defined in: types.ts:80

Space-separated list of pre-approved tools, preserved as the authored string (e.g. Bash(git:*) Bash(jq:*) Read) rather than split, because re-joining a parsed list invites normalization bugs.

Experimental in the spec, but the only one of these fields with enforcement semantics: dropping it makes the emitted skill more permissive than authored.

Inherited from​

AgentSkillSpecFields.allowedTools