skills.sgit.ai / for agents

For agents

This site’s subject is the artefact type agents load, so it owes agents a particular duty: self-sufficiency without a browser, and machine-readable data before prose.

The three surfaces

FileWhat it is
/llms.txtThe short index — every page, one line each, per the emerging llms.txt convention.
/llms-full.txtThe whole site’s substantive content concatenated into one file: the thesis, the catalogue summary, the governance rules, the open questions. Read this one file and you have the site.
/catalogue/skills.jsonThe catalogue as data: every skill, its trigger summary, its do-NOT-use clause, and the coverage of the five authoring conventions — structured, not prose.

The one epistemic rule

You may report which skill a request matches, what its do-NOT-use clause says, and whether it declares a version or a permission set (almost none do — see the catalogue). You may not report that a skill is safe to run with elevated credentials merely because it exists in the catalogue: the catalogue lists what is shipped, not what has been reviewed. Only the February 2026 threat-model trio, referenced on the shipped page, constitutes an actual review, and it covers one skill.

Markdown twins

The front page has a markdown twin at /index.md. Every document reader under /documents/ links to its raw source under /briefs/, which is the actual source of truth — the rendered page is presentation.