Agent Skills specification
references/agent-skills-specificationConfirmed by a person, 2026-10-13
Written 2026-10-13 by noelmcloughlin (a person). Last checked 2026-10-13 by noelmcloughlin (a person).
The full record
- Checks
- 2026-09-09 by
noelmcloughlin(a person) - 2026-09-26 by
noelmcloughlin(a person) - 2026-10-12 by
ktl-librarian(automation) - 2026-10-13 by
noelmcloughlin(a person)
- 2026-09-09 by
- Review by
- 2027-09-26
- Sources
- Pointed to by
- gh skill (GitHub CLI) (defined by)
- Knowledge Trust Ladder in a skills repository (related to)
- ktl-curator skill (defined by)
- ktl-docent skill (defined by)
- ktl-librarian skill (defined by)
- ktl-prose skill (defined by)
- ktl-sidecar skill (defined by)
- ktl-website skill (defined by)
- Open Skills CLI (npx skills) (defined by)
- Repository validation (references)
- Graph
- Open this concept in the knowledge graph
- IRI
https://w3id.org/ktl/knowledge/references/agent-skills-specification- Licence
- CC BY 4.0
- Files
- MarkdownJSON-LDSource on GitHubHistory on GitHub
Overview
Section titled “Overview”Defines the unit this repository distributes. Each skill is a directory with a
SKILL.md whose frontmatter carries name (matching the directory)
and description; supporting files may sit alongside it and travel with the
directory when installed.
Two consequences shape the layout here: the description is always resident in
the calling agent’s context, while the body loads
only on trigger - which is why each SKILL.md is a lean router and detail
lives in references/. The spec’s ~100 tokens covers name and description together, and is a startup estimate, not a rule. This repo only enforces 1024 characters (validate-repository.sh). gh skill publish --dry-run validates conformance in
CI on every pull request.