fix(jobindex-search): reject non-jobindex detail URLs instead of fetching them verbatim (#447) (#448)

detail fetched any http(s) input verbatim with no host check and, when
the path didn't match, silently used the whole input URL as the job id -
a non-posting page came back as a well-formed fake posting with exit 0.
buildUrl now requires a jobindex.dk host and a /jobannonce/<id> path,
rebuilds the fetch URL from the extracted id, and exits 1 with the
stderr-JSON BAD_ID contract otherwise; bare ids stay permissive
slash-free tokens per the jobnet precedent. Eight cases in the new
detail-input.test.ts; the five rejection/canonicalization cases fail
against the verbatim unguarded extraction.

.agents/skills/jobindex-search/cli/src/commands/detail.ts

@@ -60,23 +60,38 @@ function stripTags(html: string): string {
 }
 
 /**
- * Extract job ID from URL or return as-is if already an ID
+ * Parse a detail invocation's <id|url> into a canonical fetch target, or null.
+ *
+ * This is the gate between a stored (untrusted) URL and a network fetch, so it
+ * must never trust the raw string: the previous version fetched any http(s)
+ * URL verbatim and, when the path didn't match, used the whole input URL as
+ * the id - a non-posting page (a redirect target, a look-alike host, the
+ * homepage) came back as a well-formed fake posting with exit 0 (#447). A URL
+ * input now needs a jobindex.dk host (apex or subdomain) and a
+ * /jobannonce/<id> path, and the fetch URL is rebuilt from the extracted id -
+ * the canonical short form the bare-id path always used. A bare id stays a
+ * permissive scheme- and slash-free token (the jobnet precedent): the server
+ * 404s unknowns loudly, which is the honest failure. Exported for tests.
  */
-function extractIdFromUrl(url: string): string {
+export function buildUrl(idOrUrl: string): { url: string; id: string } | null {
+  const trimmed = idOrUrl.trim()
+  if (/^https?:\/\//i.test(trimmed)) {
+    let host: string
+    try {
+      host = new URL(trimmed).hostname.toLowerCase()
+    } catch {
+      return null
+    }
+    if (host !== "jobindex.dk" && !host.endsWith(".jobindex.dk")) return null
     // Match IDs like h1647303, r13677312, etc.
-  const match = url.match(/\/jobannonce\/([a-zA-Z]\d+)/)
-  if (match) return match[1]
-  return url
+    const match = trimmed.match(/\/jobannonce\/([a-zA-Z]\d+)/)
+    if (!match) return null
+    return { url: `${BASE_URL}/jobannonce/${match[1]}`, id: match[1] }
   }
-
-function buildUrl(idOrUrl: string): { url: string; id: string } {
-  if (idOrUrl.startsWith("http")) {
-    const id = extractIdFromUrl(idOrUrl)
-    return { url: idOrUrl, id }
+  if (/^[a-zA-Z0-9_-]+$/.test(trimmed)) {
+    return { url: `${BASE_URL}/jobannonce/${trimmed}`, id: trimmed }
   }
-  // It's a bare ID
-  const url = `${BASE_URL}/jobannonce/${idOrUrl}`
-  return { url, id: idOrUrl }
+  return null
 }
 
 const DANISH_MONTHS: Record<string, string> = {
@@ -262,7 +277,15 @@ export const detail = defineCommand({
       process.exit(1)
     }
 
-    const { url, id } = buildUrl(idArg)
+    const parsed = buildUrl(idArg)
+    if (!parsed) {
+      writeError(
+        `Could not parse a jobindex job id or jobannonce URL from "${idArg}"`,
+        "BAD_ID",
+      )
+      process.exit(1)
+    }
+    const { url, id } = parsed
 
     try {
       const html = await htmlFetch(url)

.agents/skills/jobindex-search/cli/tests/detail-input.test.ts

@@ -0,0 +1,52 @@
+import { describe, expect, test } from "bun:test";
+import { buildUrl } from "../src/commands/detail";
+
+// buildUrl is the gate between a stored (untrusted) URL and a network fetch.
+// It must yield a canonical jobindex fetch target or null (-> BAD_ID) - never
+// the raw input. The unguarded version fetched any http(s) URL verbatim and,
+// when the path didn't match, used the whole input URL as the id, so a
+// non-posting page came back as a well-formed fake posting with exit 0 (#447).
+
+describe("jobindex detail input parsing", () => {
+  test("canonical URL with title slug", () => {
+    expect(buildUrl("https://www.jobindex.dk/jobannonce/h1647303/senior-data-engineer")).toEqual({
+      url: "https://www.jobindex.dk/jobannonce/h1647303",
+      id: "h1647303",
+    });
+  });
+
+  test("trailing slash and query string variants", () => {
+    expect(buildUrl("https://www.jobindex.dk/jobannonce/r13677312/")?.id).toBe("r13677312");
+    expect(buildUrl("https://www.jobindex.dk/jobannonce/h1647303?utm_source=x")?.id).toBe("h1647303");
+  });
+
+  test("jobindex subdomains and bare apex are accepted", () => {
+    expect(buildUrl("https://it.jobindex.dk/jobannonce/h1647303")?.id).toBe("h1647303");
+    expect(buildUrl("https://jobindex.dk/jobannonce/h1647303")?.id).toBe("h1647303");
+  });
+
+  test("a bare id builds the canonical URL (server 404s unknowns loudly)", () => {
+    expect(buildUrl("h1647303")).toEqual({
+      url: "https://www.jobindex.dk/jobannonce/h1647303",
+      id: "h1647303",
+    });
+  });
+
+  test("an off-host URL is rejected, not fetched", () => {
+    expect(buildUrl("https://evil.example/jobannonce/h1647303")).toBeNull();
+  });
+
+  test("look-alike and userinfo hosts are rejected", () => {
+    expect(buildUrl("https://jobindex.dk.evil.example/jobannonce/h1647303")).toBeNull();
+    expect(buildUrl("https://www.jobindex.dk@evil.example/jobannonce/h1647303")).toBeNull();
+  });
+
+  test("an own-host URL without a jobannonce id is rejected (the fake-posting repro)", () => {
+    expect(buildUrl("https://www.jobindex.dk/")).toBeNull();
+  });
+
+  test("garbage bare input is rejected", () => {
+    expect(buildUrl("not a slug!")).toBeNull();
+    expect(buildUrl("ftp://www.jobindex.dk/jobannonce/h1")).toBeNull();
+  });
+});

CHANGELOG.md

@@ -47,6 +47,25 @@ per-file diff commands.
 
 ### Fixed
 
+- **`jobindex-search detail` no longer fetches arbitrary URLs or invents posting-shaped
+  output** (#447) - the command fetched any `http(s)` input verbatim (no host check) and,
+  when the path didn't match its one pattern, silently used the whole input URL as the job
+  id; the only net was "the fetched page has a title", so a non-posting page came back as
+  a well-formed fake posting with exit 0 (demonstrated with jobindex's own homepage:
+  `id` = the URL, `title` = the site's tagline, `description` = navigation chrome). Every
+  other portal CLI rejects unparseable detail input with `BAD_ID` and constructs its fetch
+  URL from the extracted id; jobindex was the one CLI trusting the raw string - and
+  `/scrape`/`/rank` agents feed it stored URLs, so a ghost or redirected URL (the #331
+  class) yielded plausible garbage instead of an error. `buildUrl` now requires a
+  jobindex.dk host (apex or subdomain - look-alike and userinfo tricks rejected via real
+  URL parsing) plus a `/jobannonce/<id>` path, rebuilds the fetch URL from the extracted
+  id (the canonical short form the bare-id path always used), and exits 1 with the
+  stderr-JSON `BAD_ID` contract otherwise; bare ids stay permissive scheme- and
+  slash-free tokens (the jobnet precedent - the server 404s unknowns loudly). Pinned by
+  eight cases in the new `detail-input.test.ts`; the five rejection/canonicalization
+  cases fail against the verbatim unguarded extraction. Complementary to the `/apply`
+  host-check rule proposed in #431, which stays with its proposer.
+
 - **`/outcome` and `/interview` no longer confuse two roles at the same company** (#443)
   (`.claude/commands/outcome.md`, `.claude/commands/interview.md`,
   `tests/test_apply_records_application.py`) - when a tracker row's `cv_file` /