What
Publish an Agentic Resource Discovery (ARD) ai-catalog.json for Git-Ape so ARD-aware AI agents and discovery services can automatically find and invoke Git-Ape's skills, plus a human-readable ARD reference page in the Docusaurus docs.
Implemented by #220.
Why
ARD is an open, DNS-like discovery protocol (developed by a working group including Microsoft, Google, GitHub, and others) that sits before invocation: an agent queries a well-known catalog endpoint to learn what skills exist, reads their descriptions, and then invokes them through their own native mechanism. Publishing an ai-catalog.json makes Git-Ape's skills discoverable by any ARD crawler or enterprise discovery service without a bespoke integration, and gives us a single authoritative, machine-readable source of the skill set that stays in sync with .github/skills/.
Scope
ai-catalog.json — ARD v1.0 envelope (specVersion, host, entries[], collections[]) listing all 15 skills in .github/skills/, served as a static asset by the docs site.- ARD reference page under Docs → Reference.
- Sidebar entry for the new page.
Acceptance criteria
Open items from review
Tracked on #220. Key points: the catalog must live where the Pages build actually publishes it (website/static/.well-known/, not repo-root docs/), and the reference page + PR description must match the shipped catalog (the collections federation was removed, so the "48 skills" / SaaS + AWS claims are stale).
What
Publish an Agentic Resource Discovery (ARD)
ai-catalog.jsonfor Git-Ape so ARD-aware AI agents and discovery services can automatically find and invoke Git-Ape's skills, plus a human-readable ARD reference page in the Docusaurus docs.Implemented by #220.
Why
ARD is an open, DNS-like discovery protocol (developed by a working group including Microsoft, Google, GitHub, and others) that sits before invocation: an agent queries a well-known catalog endpoint to learn what skills exist, reads their descriptions, and then invokes them through their own native mechanism. Publishing an
ai-catalog.jsonmakes Git-Ape's skills discoverable by any ARD crawler or enterprise discovery service without a bespoke integration, and gives us a single authoritative, machine-readable source of the skill set that stays in sync with.github/skills/.Scope
ai-catalog.json— ARD v1.0 envelope (specVersion,host,entries[],collections[]) listing all 15 skills in.github/skills/, served as a static asset by the docs site.Acceptance criteria
https://azure.github.io/git-ape/.well-known/ai-catalog.jsonreturns HTTP 200 with valid JSON after merge (the file must be part of the published Pages build)..github/skills/(15 skills) and the reference doc.Open items from review
Tracked on #220. Key points: the catalog must live where the Pages build actually publishes it (
website/static/.well-known/, not repo-rootdocs/), and the reference page + PR description must match the shipped catalog (thecollectionsfederation was removed, so the "48 skills" / SaaS + AWS claims are stale).