feat(skill): add skill discovery from URLs via well-known RFC (#12423)

Co-authored-by: Frank <[email protected]>

packages/opencode/test/skill/discovery.test.ts

@@ -0,0 +1,60 @@
+import { describe, test, expect } from "bun:test"
+import { Discovery } from "../../src/skill/discovery"
+import path from "path"
+
+const CLOUDFLARE_SKILLS_URL = "https://developers.cloudflare.com/.well-known/skills/"
+
+describe("Discovery.pull", () => {
+  test("downloads skills from cloudflare url", async () => {
+    const dirs = await Discovery.pull(CLOUDFLARE_SKILLS_URL)
+    expect(dirs.length).toBeGreaterThan(0)
+    for (const dir of dirs) {
+      expect(dir).toStartWith(Discovery.dir())
+      const md = path.join(dir, "SKILL.md")
+      expect(await Bun.file(md).exists()).toBe(true)
+    }
+  }, 30_000)
+
+  test("url without trailing slash works", async () => {
+    const dirs = await Discovery.pull(CLOUDFLARE_SKILLS_URL.replace(/\/$/, ""))
+    expect(dirs.length).toBeGreaterThan(0)
+    for (const dir of dirs) {
+      const md = path.join(dir, "SKILL.md")
+      expect(await Bun.file(md).exists()).toBe(true)
+    }
+  }, 30_000)
+
+  test("returns empty array for invalid url", async () => {
+    const dirs = await Discovery.pull("https://example.invalid/.well-known/skills/")
+    expect(dirs).toEqual([])
+  })
+
+  test("returns empty array for non-json response", async () => {
+    const dirs = await Discovery.pull("https://example.com/")
+    expect(dirs).toEqual([])
+  })
+
+  test("downloads reference files alongside SKILL.md", async () => {
+    const dirs = await Discovery.pull(CLOUDFLARE_SKILLS_URL)
+    // find a skill dir that should have reference files (e.g. agents-sdk)
+    const agentsSdk = dirs.find((d) => d.endsWith("/agents-sdk"))
+    if (agentsSdk) {
+      const refs = path.join(agentsSdk, "references")
+      expect(await Bun.file(path.join(agentsSdk, "SKILL.md")).exists()).toBe(true)
+      // agents-sdk has reference files per the index
+      const refDir = await Array.fromAsync(new Bun.Glob("**/*.md").scan({ cwd: refs, onlyFiles: true }))
+      expect(refDir.length).toBeGreaterThan(0)
+    }
+  }, 30_000)
+
+  test("caches downloaded files on second pull", async () => {
+    // first pull to populate cache
+    const first = await Discovery.pull(CLOUDFLARE_SKILLS_URL)
+    expect(first.length).toBeGreaterThan(0)
+
+    // second pull should return same results from cache
+    const second = await Discovery.pull(CLOUDFLARE_SKILLS_URL)
+    expect(second.length).toBe(first.length)
+    expect(second.sort()).toEqual(first.sort())
+  }, 60_000)
+})