fleetd #398 follow-up: pin the startup validators with a reflective validateAll()
CI / contract (pull_request) Successful in 1m9s
CI / build (pull_request) Successful in 1m28s

Mutation testing found that deleting a cfg.validateXxx() call from
Fleetd.main left the full suite green: every test called a validator
directly and none exercised main as the caller.

FleetConfig.validateAll() sweeps this class's own public no-arg void
validateXxx() methods by reflection and invokes each in alphabetical
order, so a newly written validator is wired into both callers
(Fleetd.main and ConfigRef.reload) with no second step to forget.
FleetdStartupValidationTest calls the real Fleetd.main with six configs,
each failing exactly one validator.

Recovered by the lead: the worker's agent died mid-turn with this work
uncommitted, and had left the startup call commented out as
'// MUTATION-TEST-TEMP: cfg.validateAll();' from its own mutation run.
I restored the call before committing. Build after restoring:
Tests run: 1491, Failures: 0, BUILD SUCCESS.

NOT covered, and not claimed to be: the five log-only reporters in
main (reportRequiredSecrets, reportGitHostShape, reportMemberTrustModel,
reportMemberCredentialsGap, and reportExhaustedPatternGap on current
main) are not validateXxx() methods, so the sweep does not reach them
and their call sites stay unpinned.
This commit is contained in:
Dai Ha
2026-09-10 08:56:20 +07:00
parent e7b33fe3a0
commit b540a1744b
5 changed files with 578 additions and 30 deletions
+10 -17
View File
@@ -144,23 +144,16 @@ public final class Fleetd {
SubscriptionGuard guard = new SubscriptionGuard(cfg.guard().hostSet());
guard.assertPrimaryClean(System.getenv());
// CB-501: refuse to start if the bind is wider than the auth mode can defend. Under
// loopback-trust, "not a known worker" means "the primary" — sound only because the OS
// refuses remote connections to a loopback socket. This throws rather than warns so the
// dangerous configuration cannot be reached by ignoring a log line.
cfg.validateAuthExposure();
cfg.validateLeadTabPrefixes();
// CB-542: a subscription:true profile whose env: reseats ANTHROPIC_BASE_URL/AUTH_TOKEN would
// reach an unguarded endpoint (the launcher skips SubscriptionGuard for it). Refuse at load.
cfg.validateSubscriptionProfiles();
cfg.validateCharters();
// CB-548: every architect slot must name a configured workers: profile — the strong-model
// backend the future spawn lifecycle would read. A stale reference dies here, not later.
cfg.validateMembers();
// fleetd ticket "central allow-list of usable models": an absent/empty models.allow: keeps
// today's behaviour (no model was ever checked); once configured, a profile naming a model
// outside it dies here, at load, rather than reaching a backend adapter as a free-form string.
cfg.validateModels();
// Every FleetConfig.validateXxx() the operator's config can fail — CB-501's auth-exposure
// check, CB-531's lead-tab-prefix check, CB-542's subscription-profile check, the charter
// and member-slot checks, and the "central allow-list of usable models" check — must run
// here, at load, before anything below opens a socket or spawns a member. fleetd ticket
// "central allow-list of usable models" follow-up: mutation testing found six individual
// calls here with nothing proving any of them still ran (deleting one left the full suite
// green). validateAll() replaces them with the one call that FleetConfigValidateAllTest
// and the Fleetd-startup tests actually pin — see FleetConfig#validateAll's javadoc for
// why a name-by-name list here would have the same defect it replaces.
cfg.validateAll();
Path socket = cfg.herdrSocket() != null && !cfg.herdrSocket().isBlank()
? Path.of(cfg.herdrSocket())
@@ -44,7 +44,8 @@ import java.util.function.Supplier;
* {@code spawnReadyTimeoutMs} / {@code spawnReadyPollMs}, {@code quarantineCooldownSeconds}
* (CB-578 stage B — baked once into the {@code BackendQuarantine} built at startup),
* {@code models:} (fleetd ticket "central allow-list of usable models" — {@link
* FleetConfig#validateModels()} re-runs against the fresh config in {@link #reload()}, so a
* FleetConfig#validateModels()} re-runs against the fresh config in {@link #reload()}
* (via {@link FleetConfig#validateAll()}), so a
* models.allow: edit that would refuse to boot still refuses the reload; a change that
* passes has nothing built at startup to rebuild, so it is reported deferred rather than
* silently accepted with no report at all),
@@ -328,16 +329,12 @@ public final class ConfigRef implements Supplier<FleetConfig> {
fresh = FleetConfig.load(path);
// The same gate startup runs. A config that would have refused to boot must not be able
// to slip in through a reload — that is how a daemon ends up in a state it could never
// have started in, which is the hardest kind to debug.
fresh.validateAuthExposure();
fresh.validateLeadTabPrefixes();
fresh.validateSubscriptionProfiles();
fresh.validateCharters();
fresh.validateMembers();
// fleetd ticket "central allow-list of usable models": same reason as validateMembers
// above — a models.allow: that would have refused to boot must not slip in through a
// reload either.
fresh.validateModels();
// have started in, which is the hardest kind to debug. fleetd ticket "central allow-list
// of usable models" follow-up: this used to be six individual validateXxx() calls, and
// mutation testing found two of the six unpinned here even though startup pinned nothing
// at all — see FleetConfig#validateAll's javadoc for why the fix is one reflective call,
// not a longer hand-maintained list.
fresh.validateAll();
} catch (RuntimeException e) {
String msg = e.getMessage() == null ? e.toString() : e.getMessage();
log.warn("config reload from {} refused, keeping the running config: {}", path, msg);
@@ -450,8 +447,9 @@ public final class ConfigRef implements Supplier<FleetConfig> {
changed.add("idleSleepGuard");
}
// fleetd ticket "central allow-list of usable models": validateModels() runs again in
// reload() above, so a bad edit is already refused as cold-adjacent (the whole reload is
// refused via the catch block, never partially applied). A GOOD edit to the allow-list
// reload() above (via validateAll()), so a bad edit is already refused as cold-adjacent
// (the whole reload is refused via the catch block, never partially applied). A GOOD edit
// to the allow-list
// itself has nothing built at startup to rebuild — it only ever mattered to the validation
// call that already ran — so report it deferred rather than silently swallowing the change.
if (!Objects.equals(old.models(), fresh.models())) {
@@ -16,11 +16,15 @@ import org.slf4j.LoggerFactory;
import java.io.IOException;
import java.io.UncheckedIOException;
import java.lang.reflect.InvocationTargetException;
import java.lang.reflect.Method;
import java.lang.reflect.Modifier;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.List;
@@ -2609,6 +2613,79 @@ public record FleetConfig(
}
}
/**
* Runs every validator this class declares — found by reflection, not by name.
*
* <p>fleetd ticket "central allow-list of usable models", follow-up: mutation testing found
* that although each of the six validators above was well pinned on its own, nothing proved
* either real caller ({@code Fleetd.main} and {@link ConfigRef#reload()}) still
* invoked it — deleting a call site left the full suite green. The fix is not a seventh test
* per caller; a hand-maintained list of six names here would have the exact same defect its
* own javadoc would warn against: the seventh validator someone adds next month has no reason
* to be added to it. So this method does not name any validator. It sweeps {@link
* #getClass()}'s own public, no-argument, {@code void} methods whose name starts with {@code
* "validate"} (excluding itself) and invokes every one it finds, via {@link
* #invokeAllValidators}. A new {@code validateXxx()} method is therefore wired into both
* callers the moment it is written — there is no second step to forget, and so no state in
* which it silently never runs.
*
* <p>{@code Fleetd.main} and {@link ConfigRef#reload()} each call this one method instead of
* the six individually — see the comments at those two call sites for why
* each must run it.
*
* <p>Methods run in a fixed (alphabetical) order, so a config with more than one violation
* always names the same one first, on every run.
*
* @throws IllegalStateException (or whatever unchecked exception a validator itself throws),
* propagated unchanged from the first validator, in that order,
* that finds a problem
*/
public void validateAll() {
invokeAllValidators(this);
}
/**
* The reflective sweep behind {@link #validateAll()}, kept as its own method — taking any
* {@code target}, not just {@code this} — so a test can prove the MECHANISM is generic (it
* would sweep a seventh {@code validateXxx()} method added to any class, not just something
* special-cased to today's six on {@link FleetConfig}) without needing to add a real, unwanted
* seventh validator to this class just to exercise that claim. See {@code
* FleetConfigValidateAllTest} for that proof.
*
* @param target an object whose public, no-argument, {@code void} methods named {@code
* validateXxx} (any name starting with {@code "validate"}, excluding {@code
* validateAll} itself) should all run, in alphabetical-by-name order
*/
static void invokeAllValidators(Object target) {
List<Method> methods = new ArrayList<>();
for (Method m : target.getClass().getMethods()) {
if (Modifier.isPublic(m.getModifiers())
&& m.getParameterCount() == 0
&& m.getReturnType() == void.class
&& m.getName().startsWith("validate")
&& !m.getName().equals("validateAll")) {
methods.add(m);
}
}
methods.sort(Comparator.comparing(Method::getName));
for (Method m : methods) {
try {
m.invoke(target);
} catch (InvocationTargetException e) {
Throwable cause = e.getCause();
if (cause instanceof RuntimeException re) {
throw re;
}
if (cause instanceof Error err) {
throw err;
}
throw new IllegalStateException("validator " + m.getName() + " failed", cause);
} catch (IllegalAccessException e) {
throw new IllegalStateException("cannot invoke validator " + m.getName(), e);
}
}
}
/** True for the loopback addresses and the unspecified-but-local forms we treat as same-host. */
private static boolean isLoopbackBind(String host) {
if (host == null || host.isBlank()) {
@@ -0,0 +1,132 @@
package dev.ltms.fleet;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import java.nio.file.Files;
import java.nio.file.Path;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* Proves the one thing no test proved before this ticket's follow-up: that {@code Fleetd.main}
* ITSELF — not a copy of its logic, not the validator called directly — still refuses to start on
* a bad config. Mutation testing found that deleting {@code cfg.validateAll();} (née six
* individual {@code cfg.validateXxx();} calls) from {@code Fleetd.main} left the full 1478-test
* suite green; every existing test called a validator directly and none exercised {@code
* Fleetd.main} as the caller. See {@code FleetConfigValidateAllTest} for why the fix collapses
* those six calls into one reflective {@link FleetConfig#validateAll()}, and {@code
* ConfigRefTest} for the equivalent proof on the {@link
* dev.ltms.fleet.config.ConfigRef#reload()} path.
*
* <p>This is deliberately the real {@code static void main(String[] args)} — package-private, so
* only a test in this package can call it, which is exactly what makes this proof strong: it is
* not a helper extracted for testability, it is the literal method {@code java -jar fleetd.jar}
* invokes. Every fixture below is otherwise-valid and fails exactly one validator, and — because
* {@link FleetConfig#validateAll()} runs immediately after {@code SubscriptionGuard.
* assertPrimaryClean}, before {@code main} opens the herdr socket, binds Javalin, or touches
* anything else with a real side effect — calling {@code Fleetd.main} with one of these configs is
* safe: it is guaranteed to throw before reaching any of that, precisely because the config is
* deliberately invalid.
*/
class FleetdStartupValidationTest {
private static void assertMainRefuses(Path dir, String fileName, String yaml,
String mustContain) throws Exception {
Path f = dir.resolve(fileName);
Files.writeString(f, yaml);
IllegalStateException e = assertThrows(IllegalStateException.class,
() -> Fleetd.main(new String[]{f.toString()}),
fileName + ": Fleetd.main must refuse this config before doing anything else");
assertTrue(e.getMessage().contains(mustContain),
fileName + ": expected message to contain \"" + mustContain + "\" but was: "
+ e.getMessage());
}
@Test
void mainRefusesANonLoopbackBindWithoutTokenMode(@TempDir Path dir) throws Exception {
assertMainRefuses(dir, "auth-exposure.yaml", """
bind:
host: 0.0.0.0
port: 8765
""", "auth.mode: token");
}
@Test
void mainRefusesALeadTabPrefixCollision(@TempDir Path dir) throws Exception {
assertMainRefuses(dir, "lead-tab-prefixes.yaml", """
bind:
host: 127.0.0.1
port: 8765
fleet:
tabLabel: "lead: {role} {profile}"
leaders:
opus:
tab: "lead: opus"
""", "fleet.tabLabel");
}
@Test
void mainRefusesASubscriptionProfileThatReseatsAnthropicBaseUrl(@TempDir Path dir)
throws Exception {
assertMainRefuses(dir, "subscription-profiles.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
sonnet:
subscription: true
argv: ["ccs", "sonnet"]
env:
ANTHROPIC_BASE_URL: http://anything-not-on-the-allowlist
""", "ANTHROPIC_BASE_URL");
}
@Test
void mainRefusesAnUnknownCharterKey(@TempDir Path dir) throws Exception {
assertMainRefuses(dir, "charters.yaml", """
bind:
host: 127.0.0.1
port: 8765
fleet:
charters:
architetc: text
""", "architetc");
}
@Test
void mainRefusesAnArchitectSlotNamingAnUnconfiguredProfile(@TempDir Path dir) throws Exception {
assertMainRefuses(dir, "members.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
gx10:
baseUrl: http://gx10.gw:8000
fleet:
architects:
lead-designer:
profile: sonnet
""", "lead-designer");
}
@Test
void mainRefusesAProfileNamingAModelOutsideTheAllowList(@TempDir Path dir) throws Exception {
assertMainRefuses(dir, "models.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
sonnet:
baseUrl: http://gx10.gw:8000
model: claude-sonnet-5
rogue:
baseUrl: http://gx11.gw:8000
model: claude-opus-9000
models:
allow:
- model: claude-sonnet-5
""", "rogue");
}
}
@@ -0,0 +1,348 @@
package dev.ltms.fleet.config;
import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.io.TempDir;
import java.lang.reflect.Method;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.ArrayList;
import java.util.List;
import java.util.Set;
import java.util.TreeSet;
import static org.junit.jupiter.api.Assertions.assertDoesNotThrow;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* The gap this class exists to close: mutation testing on the fleetd ticket "central allow-list
* of usable models" found that although {@link FleetConfig#validateModels()}'s own logic was well
* pinned, nothing proved either real caller ({@code Fleetd.main} and {@link ConfigRef#reload()})
* still invoked it — deleting the call site left the full suite green (1478/0/0/0). A follow-up
* measurement (same technique — remove one call site, run the suite, not read the code) found the
* SAME gap for all five of {@link FleetConfig}'s other validators at startup, and for four of the
* six inside {@link ConfigRef#reload()}. This is a class of gap, not one line's mistake: every one
* of those thirteen tests called the validator itself directly, never the real caller that was
* supposed to.
*
* <p>The fix replaces the six individual {@code cfg.validateXxx()} calls at each of the two real
* call sites with one {@link FleetConfig#validateAll()}, which reaches every validator by
* reflection rather than by a hand-maintained list of names. A hand-maintained list of six names
* would have exactly the defect it replaces: the seventh validator someone adds next month has no
* reason to be added to it, and nothing would say so. This class proves TWO separate claims, and
* keeps them separate on purpose:
*
* <ol>
* <li>{@link #theSweepMechanismIsGenericNotHardcodedToFleetConfigsSixNames()} and its neighbours
* prove the reflective sweep itself ({@link FleetConfig#invokeAllValidators}) is a general
* mechanism — it runs whatever public, no-arg, void {@code validateXxx()} methods a class
* happens to declare today, including a class with more of them than {@link FleetConfig}
* has right now. This is the proof that a future, real seventh validator on {@link
* FleetConfig} would be swept automatically, without needing to add a real (unwanted)
* seventh validator just to exercise the claim.</li>
* <li>{@link #validateAllReachesEveryOneOfTodaysSixValidators()} proves {@link
* FleetConfig#validateAll()} itself is wired to that same generic mechanism and genuinely
* reaches each of today's six real validators — reusing the exact minimal failing
* configurations {@code FleetConfigTest} already established for each one directly, so a
* single call to {@code validateAll()} is shown to reproduce every one of those six
* failures.</li>
* </ol>
*
* <p>Together with the direct-{@code Fleetd.main}-invocation tests in {@code
* FleetdStartupValidationTest} (which prove the real startup call site still calls {@code
* validateAll()}) and the {@code ConfigRefTest} reload tests (which prove the same for {@link
* ConfigRef#reload()}), removing {@code cfg.validateAll();} from either real call site, or
* reverting {@link FleetConfig#validateAll()} to a hardcoded list of method calls, now fails a
* test in this module.
*/
class FleetConfigValidateAllTest {
// ── Claim 1: the reflective sweep is a general mechanism, not six names in disguise ──────────
/**
* A throwaway fixture class, unrelated to {@link FleetConfig} in every way except shape: three
* public, no-arg, void methods named {@code validateXxx}. Proves the sweep works on ANY class
* with this shape, not on something special-cased to {@link FleetConfig}.
*/
static class ThreeValidators {
final List<String> ran = new ArrayList<>();
public void validateAlpha() {
ran.add("validateAlpha");
}
public void validateBeta() {
ran.add("validateBeta");
}
public void validateGamma() {
ran.add("validateGamma");
}
}
@Test
void theSweepMechanismIsGenericNotHardcodedToFleetConfigsSixNames() {
ThreeValidators target = new ThreeValidators();
FleetConfig.invokeAllValidators(target);
assertEquals(List.of("validateAlpha", "validateBeta", "validateGamma"), target.ran,
"every validateXxx() method on this unrelated class must run, in alphabetical "
+ "order — the sweep reads the class's own shape, not a name FleetConfig "
+ "happens to know about");
}
/**
* The core of the "self-maintaining" requirement: the exact same class shape as {@link
* ThreeValidators}, plus one more method — standing in for "a developer adds a validator next
* month". Nothing about the sweep changes to pick it up; the new method is invoked purely
* because it exists and matches the shape. This is what makes adding a seventh real validator
* to {@link FleetConfig} safe without touching {@link FleetConfig#validateAll()} or either
* call site — there is no "wire it in" step left to forget.
*/
static class FourValidators {
final List<String> ran = new ArrayList<>();
public void validateAlpha() {
ran.add("validateAlpha");
}
public void validateBeta() {
ran.add("validateBeta");
}
public void validateGamma() {
ran.add("validateGamma");
}
public void validateDelta() {
ran.add("validateDelta");
}
}
@Test
void addingAFourthValidatorMethodGetsSweptWithNoOtherChange() {
FourValidators target = new FourValidators();
FleetConfig.invokeAllValidators(target);
assertEquals(List.of("validateAlpha", "validateBeta", "validateDelta", "validateGamma"),
sorted(target.ran),
"the fourth method must be reached automatically — proving a class can grow the "
+ "set of things it validates with no change to the sweep itself");
}
private static List<String> sorted(List<String> in) {
List<String> copy = new ArrayList<>(in);
copy.sort(String::compareTo);
return copy;
}
@Test
void aFailingValidatorStopsTheSweepAndPropagatesUnchanged() {
class OneFails {
@SuppressWarnings("unused")
public void validateOk() {
// passes
}
public void validateBoom() {
throw new IllegalStateException("refusing to start: boom");
}
}
IllegalStateException e = assertThrows(IllegalStateException.class,
() -> FleetConfig.invokeAllValidators(new OneFails()));
assertEquals("refusing to start: boom", e.getMessage(),
"the real exception must propagate unchanged, not be wrapped or swallowed");
}
/**
* Every rule the sweep's filter applies, proven independently: only public, no-arg, void
* methods whose name starts with {@code "validate"} run, {@code validateAll} itself is
* excluded (so a class that declares one of its own — as {@link FleetConfig} does — cannot
* recurse into itself), and a same-shaped-but-wrongly-named or wrongly-shaped method never
* runs. A reader who "simplifies" the filter in {@link FleetConfig#invokeAllValidators} in a
* way that widens or narrows it breaks one of these.
*/
static class FilterEdgeCases {
final List<String> ran = new ArrayList<>();
public void validateReal() {
ran.add("validateReal");
}
/** Wrong name — must not run. */
public void checkSomething() {
ran.add("checkSomething");
}
/** Wrong shape — takes an argument. */
public void validateWithArg(String ignored) {
ran.add("validateWithArg");
}
/** Wrong shape — returns something. */
public boolean validateReturnsBoolean() {
ran.add("validateReturnsBoolean");
return true;
}
/** Excluded by name on purpose, so the sweep cannot call itself. */
public void validateAll() {
ran.add("validateAll");
}
}
@Test
void onlyPublicNoArgVoidValidateNamedMethodsRun() {
FilterEdgeCases target = new FilterEdgeCases();
FleetConfig.invokeAllValidators(target);
assertEquals(List.of("validateReal"), target.ran,
"checkSomething (wrong name), validateWithArg (wrong shape), "
+ "validateReturnsBoolean (wrong shape), and validateAll (excluded by "
+ "name) must all be skipped");
}
// ── Claim 2: FleetConfig.validateAll() is wired to that mechanism and reaches all six today ──
/**
* Reflectively enumerates {@link FleetConfig}'s own public, no-arg, void {@code validateXxx()}
* methods (excluding {@code validateAll} itself) — the exact same filter {@link
* FleetConfig#invokeAllValidators} applies. This is not the mechanism proof (that is claim 1,
* above, on an unrelated class) — it is a visible denominator: today there are six, named
* here, so a reader adding a seventh sees this assertion name the new count rather than a
* silent pass at the old one.
*/
@Test
void fleetConfigDeclaresExactlyTheseSixValidatorsToday() {
Set<String> names = new TreeSet<>();
for (Method m : FleetConfig.class.getMethods()) {
if (java.lang.reflect.Modifier.isPublic(m.getModifiers())
&& m.getParameterCount() == 0
&& m.getReturnType() == void.class
&& m.getName().startsWith("validate")
&& !m.getName().equals("validateAll")) {
names.add(m.getName());
}
}
assertEquals(new TreeSet<>(Set.of("validateAuthExposure", "validateLeadTabPrefixes",
"validateSubscriptionProfiles", "validateCharters", "validateMembers",
"validateModels")), names,
"FleetConfig's public validate*() methods changed — this is not a failure by "
+ "itself (validateAll() sweeps whatever is here, by construction), it is "
+ "this test's own denominator; update the expected set to match");
}
/** A minimal, otherwise-valid file — same shape FleetConfigTest and ConfigRefTest use. */
private static String minimalValidYaml() {
return """
bind:
host: 127.0.0.1
port: 8765
profiles:
sonnet:
baseUrl: http://gx00.gw:8000
model: sonnet
""";
}
@Test
void aFullyValidConfigPassesValidateAll(@TempDir Path dir) throws Exception {
Path f = dir.resolve("fleetd.yaml");
Files.writeString(f, minimalValidYaml());
assertDoesNotThrow(() -> FleetConfig.load(f).validateAll());
}
/**
* The heart of claim 2: for each of today's six real validators, a minimal file that fails
* ONLY that one — the exact fixtures {@code FleetConfigTest} uses to test each validator
* directly — must also fail through {@link FleetConfig#validateAll()}. If a future edit to
* {@code validateAll()} silently dropped one validator from the sweep (e.g. a typo'd name
* filter), exactly one of these six would start passing when it must not.
*/
@Test
void validateAllReachesEveryOneOfTodaysSixValidators(@TempDir Path dir) throws Exception {
// validateAuthExposure: a non-loopback bind without token mode.
assertValidateAllRefuses(dir, "auth-exposure.yaml", """
bind:
host: 0.0.0.0
port: 8765
""", "auth.mode: token");
// validateLeadTabPrefixes: a fleet-wide tabLabel that starts with a lead's own tabPrefix.
assertValidateAllRefuses(dir, "lead-tab-prefixes.yaml", """
bind:
host: 127.0.0.1
port: 8765
fleet:
tabLabel: "lead: {role} {profile}"
leaders:
opus:
tab: "lead: opus"
""", "fleet.tabLabel");
// validateSubscriptionProfiles: subscription: true with env: reseating ANTHROPIC_BASE_URL.
assertValidateAllRefuses(dir, "subscription-profiles.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
sonnet:
subscription: true
argv: ["ccs", "sonnet"]
env:
ANTHROPIC_BASE_URL: http://anything-not-on-the-allowlist
""", "ANTHROPIC_BASE_URL");
// validateCharters: a charter key that is not a role wire name.
assertValidateAllRefuses(dir, "charters.yaml", """
bind:
host: 127.0.0.1
port: 8765
fleet:
charters:
architetc: text
""", "architetc");
// validateMembers: an architect slot naming an unconfigured profile.
assertValidateAllRefuses(dir, "members.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
gx10:
baseUrl: http://gx10.gw:8000
fleet:
architects:
lead-designer:
profile: sonnet
""", "lead-designer");
// validateModels: a profile naming a model outside the configured allow-list.
assertValidateAllRefuses(dir, "models.yaml", """
bind:
host: 127.0.0.1
port: 8765
profiles:
sonnet:
baseUrl: http://gx10.gw:8000
model: claude-sonnet-5
rogue:
baseUrl: http://gx11.gw:8000
model: claude-opus-9000
models:
allow:
- model: claude-sonnet-5
""", "rogue");
}
private static void assertValidateAllRefuses(Path dir, String fileName, String yaml,
String mustContain) throws Exception {
Path f = dir.resolve(fileName);
Files.writeString(f, yaml);
FleetConfig cfg = FleetConfig.load(f);
IllegalStateException e = assertThrows(IllegalStateException.class, cfg::validateAll,
fileName + ": validateAll() must refuse this config");
assertTrue(e.getMessage().contains(mustContain),
fileName + ": expected message to contain \"" + mustContain + "\" but was: "
+ e.getMessage());
}
}