test(apply): derive the host-check's installed portals from the skills tree (#528)

SHIPPED_PORTAL_HOSTS was six names typed into the test and nothing tied
it to anything: a portal added by /add-portal fetches from a host the
copy never heard of, so its postings reached /apply classified
"unverified" while the suite stayed green, and a dropped portal left a
stale name behind.

apply.md rule 1 defines an installed portal as "any configured job portal
in .agents/skills/", so that tree is the source: the host literals in
each *-search/cli/src/**/*.ts (one BASE_URL per shipped CLI, www. dropped
so apex and subdomain both match). Discovery is by installation, not by
the enabled: flag - an installed-but-disabled board is still a legitimate
source for a pasted posting. classify_posting_host's default is the
discovered set, and a new class asserts every installed portal has a
discoverable host that classifies as installed_portal at the apex, under
www. and under a subdomain. The ATS apex list stays the fixed set it is.

Reported by Holo-Eter in #497; shape per the maintainer's note there.

CHANGELOG.md

@@ -15,6 +15,22 @@ per-file diff commands.
 
 ### Fixed
 
+- **The `/apply` host-check test derives its installed-portal hosts from the skills tree
+  instead of a hand-copied set** (#497, `tests/test_apply_host_check.py`) - `SHIPPED_PORTAL_HOSTS`
+  was six names typed into the test and nothing tied it to anything: a portal added by
+  `/add-portal` fetches from a host the copy never heard of, so its postings reached `/apply`
+  classified "unverified" while the suite stayed green, and a dropped portal left a stale
+  name behind. Rule 1 in `apply.md` defines an installed portal as "any configured job portal
+  in `.agents/skills/`", so that tree is now the source: the host literals in each
+  `*-search/cli/src/**/*.ts` (one `BASE_URL` per shipped CLI, `www.` dropped so apex and
+  subdomain both match), discovery by installation rather than by the `enabled:` flag
+  because an installed-but-disabled board is still a legitimate source for a pasted posting.
+  `classify_posting_host`'s default is that discovered set, and a new test class asserts that
+  every installed portal has a discoverable host and that each host classifies as
+  `installed_portal` at the apex, under `www.` and under a subdomain. The ATS apex list stays
+  the fixed set it is; parsing `apply.md`'s example list was deliberately not done, since a
+  parser keyed on prose wording is the next thing to go stale. Reported by Holo-Eter.
+
 - **`/expand` and `/setup` now append behavioral findings to sections `02-behavioral-profile.md`
   actually has** (`.claude/commands/expand.md` Step 5, `.claude/commands/setup.md` Step A7,
   `tests/test_expand_command.py`, `tests/test_setup_command.py`) - both told the model to add

tests/test_apply_host_check.py

@@ -6,6 +6,14 @@ Pins the invariants from the maintainer design in issue #431:
   myworkdayjobs.com (or workday.com), ashbyhq.com, smartrecruiters.com, workable.com.
 - Look-alike attacks (prefixes, suffixes, userinfo tricks) must fail closed.
 - Any other host must be named plainly in the output as unverified.
+
+The installed-portal half is derived, not copied (#497). Rule 1 in apply.md
+defines an installed portal as "any configured job portal in `.agents/skills/`",
+so the source of truth is the skills tree: the host literals in each
+`*-search/cli/src/**/*.ts`. A hand-copied set in this file went stale the moment
+a portal was added or dropped, and a new portal CLI fetching from a host the
+copy lacked had its postings classified "unverified" with no test failing.
+The ATS apex list stays the fixed set it is.
 """
 
 import re
@@ -15,6 +23,59 @@ from urllib.parse import urlparse
 
 REPO = Path(__file__).resolve().parent.parent
 APPLY_COMMAND_FILE = REPO / ".claude" / "commands" / "apply.md"
+SKILLS_DIR = REPO / ".agents" / "skills"
+
+# A host literal in a CLI source file. Portal CLIs build every request URL from
+# a BASE_URL / SEARCH_URL constant, so this is where the board's host lives.
+HOST_LITERAL = re.compile(r"https?://([A-Za-z0-9.-]+\.[A-Za-z]{2,})")
+# Hosts a source file may name that are not the portal: schema vocabularies,
+# RFC example domains, loopback. Extend if a CLI ever embeds another one.
+NON_PORTAL_HOSTS = {"schema.org", "www.w3.org", "example.com", "localhost"}
+
+
+def installed_portal_skills() -> list[Path]:
+    """Every `*-search` skill with a CLI source tree, enabled or not.
+
+    Rule 1 says "configured job portal in `.agents/skills/`" - a portal that is
+    installed but `enabled: false` for /scrape is still a legitimate board for
+    a posting the user pastes into /apply, so discovery is by installation.
+    """
+    return sorted(
+        d for d in SKILLS_DIR.glob("*-search") if (d / "cli" / "src").is_dir()
+    )
+
+
+def portal_enabled(skill_dir: Path) -> bool:
+    """The `enabled:` frontmatter flag /scrape honours; a missing key means enabled."""
+    skill_md = skill_dir / "SKILL.md"
+    if not skill_md.is_file():
+        return True
+    match = re.search(r"^enabled:\s*(true|false)\b", skill_md.read_text(encoding="utf-8"), re.M)
+    return match is None or match.group(1) == "true"
+
+
+def portal_hosts(skill_dir: Path) -> set[str]:
+    """The board hosts a portal CLI fetches from, as registrable names.
+
+    A leading `www.` is dropped so the classifier's exact-or-subdomain rule
+    accepts both `jobindex.dk` and `www.jobindex.dk` (the CLI builds its URLs
+    with the latter; postings link to both).
+    """
+    hosts: set[str] = set()
+    for source in (skill_dir / "cli" / "src").rglob("*.ts"):
+        for host in HOST_LITERAL.findall(source.read_text(encoding="utf-8")):
+            host = host.lower()
+            if host in NON_PORTAL_HOSTS:
+                continue
+            hosts.add(host[4:] if host.startswith("www.") else host)
+    return hosts
+
+
+def discover_installed_portal_hosts() -> set[str]:
+    return set().union(*(portal_hosts(d) for d in installed_portal_skills()))
+
+
+INSTALLED_PORTAL_HOSTS = discover_installed_portal_hosts()
 
 KNOWN_ATS_APEXES = {
     "greenhouse.io",
@@ -26,24 +87,21 @@ KNOWN_ATS_APEXES = {
     "workable.com",
 }
 
-SHIPPED_PORTAL_HOSTS = {
-    "jobindex.dk",
-    "linkedin.com",
-    "jobnet.dk",
-    "jobbank.dk",
-    "jobdanmark.dk",
-    "freehire.me",
-}
 
 
-def classify_posting_host(url_str: str, installed_portals: set[str] = SHIPPED_PORTAL_HOSTS) -> tuple[str, str]:
+def classify_posting_host(url_str: str, installed_portals: set[str] | None = None) -> tuple[str, str]:
     """Reference implementation of the host provenance rule in /apply Step 1.
 
+    `installed_portals` defaults to the set discovered from the skills tree, so
+    the classifier can never lag a portal that /add-portal installed.
+
     Returns (tier, host), where tier is one of:
       - 'installed_portal'
       - 'official_ats'
       - 'unverified'
     """
+    if installed_portals is None:
+        installed_portals = INSTALLED_PORTAL_HOSTS
     try:
         parsed = urlparse(url_str)
         host = (parsed.hostname or "").lower().strip()
@@ -102,17 +160,67 @@ class ApplyHostVerificationSpecTests(unittest.TestCase):
             self.assertEqual(tier, "official_ats", f"{url} should classify as official_ats, got {tier}")
 
     def test_classifier_identifies_installed_portal_hosts(self):
+        # A regional subdomain and the bare apex must both pass for every portal;
+        # the derived set is what makes this hold for a portal added by /add-portal.
         urls = [
             "https://www.jobindex.dk/jobannonce/12345",
-            "https://www.linkedin.com/jobs/view/999999",
+            "https://dk.linkedin.com/jobs/view/999999",
             "https://jobnet.dk/find-job/8888",
-            "https://jobbank.dk/job/7777",
-            "https://freehire.me/job/6666",
+            "https://freehire.me/jobs/golang-zensar-2bxu6dxm",
         ]
         for url in urls:
             tier, host = classify_posting_host(url)
             self.assertEqual(tier, "installed_portal", f"{url} should classify as installed_portal, got {tier}")
 
+    def test_classifier_default_is_the_discovered_set_not_a_copy(self):
+        self.assertEqual(classify_posting_host.__defaults__, (None,))
+        self.assertTrue(INSTALLED_PORTAL_HOSTS, "discovery found no portal hosts under .agents/skills/")
+        # An explicit set still overrides, which is how a fork's test can probe a candidate portal.
+        self.assertEqual(
+            classify_posting_host("https://jobs.newboard.test/x", {"newboard.test"})[0], "installed_portal"
+        )
+        self.assertEqual(classify_posting_host("https://jobs.newboard.test/x")[0], "unverified")
+
+
+class InstalledPortalsAreVerifiedTests(unittest.TestCase):
+    """Every installed portal CLI's board host must classify as installed_portal.
+
+    This is the assertion the hand-copied set could not make: a portal added
+    by /add-portal fetches from a host the copy never heard of, and its
+    postings reached /apply flagged "unverified" while the suite stayed green.
+    """
+
+    def test_every_installed_portal_has_a_discoverable_host(self):
+        skills = installed_portal_skills()
+        self.assertTrue(skills, "no *-search skill with a cli/src tree under .agents/skills/")
+        for skill in skills:
+            with self.subTest(portal=skill.name):
+                self.assertTrue(
+                    portal_hosts(skill),
+                    f"{skill.name}/cli/src has no host literal - the classifier cannot verify its postings",
+                )
+
+    def test_every_installed_portals_host_classifies_as_installed_portal(self):
+        for skill in installed_portal_skills():
+            for host in sorted(portal_hosts(skill)):
+                for url in (f"https://{host}/job/1", f"https://www.{host}/job/1", f"https://sub.{host}/job/1"):
+                    with self.subTest(portal=skill.name, url=url, enabled=portal_enabled(skill)):
+                        tier, _ = classify_posting_host(url)
+                        self.assertEqual(tier, "installed_portal", f"{url} classified as {tier}")
+
+    def test_portal_hosts_are_registrable_names_without_www(self):
+        for skill in installed_portal_skills():
+            for host in portal_hosts(skill):
+                with self.subTest(portal=skill.name, host=host):
+                    self.assertFalse(host.startswith("www."), host)
+                    self.assertRegex(host, r"^[a-z0-9.-]+\.[a-z]{2,}$")
+
+    def test_enabled_flag_reads_like_scrape(self):
+        # /scrape: "a missing key means enabled". Both shipped states occur in the tree.
+        states = {skill.name: portal_enabled(skill) for skill in installed_portal_skills()}
+        self.assertTrue(any(states.values()), f"no enabled portal among {states}")
+        self.assertIsInstance(portal_enabled(SKILLS_DIR / "does-not-exist-search"), bool)
+
     def test_classifier_fails_closed_on_look_alikes_and_unverified_hosts(self):
         suspicious = [
             "https://evil-greenhouse.io/job/1",