Compare commits

...

3 Commits

Author SHA1 Message Date
Dai Ha 7d497aa423 fleetd #711: re-key two pane-tab hazard reasons onto the unregistered-pane condition
CI / shell-tests (pull_request) Failing after 7s
CI / contract (pull_request) Successful in 51s
CI / build (pull_request) Failing after 1m39s
Both refusals said a member landing in a lead's or collaborator's
labelled tab would be read back as that identity. CallerResolver
consults the spawned-member roster ahead of every tab map, so a live
registered member is never misread this way. State the real condition
instead: the hazard applies only while the pane is alive and carries
no entry in the spawned-member roster.
2026-10-04 06:54:43 +02:00
Dai Ha 4939d40362 Merge PR #709: fleetd #703 — fleet_list reports collaborators, narrowed by role
CI / shell-tests (push) Failing after 12s
CI / contract (push) Successful in 48s
CI / build (push) Failing after 1m53s
A collaborator could find and message a lead, but a lead had nowhere to read a
collaborator's sessionId, so the channel only worked once the collaborator
spoke first. fleet_list now carries a collaborators array, and
CallerResolver.collaborators() has its first caller outside the resolver.

Visible to the primary, an architect and a collaborator; absent for a worker.
The rule is that the roles which may see a collaborator row are the roles that
can act on it, and a worker cannot send to a collaborator at all. Narrowed in
the payload the way coordinatorVisibleTo already does it, because fleet_list is
gated on READ and a worker holds READ.

Each row carries the registry name and the sessionId only. A collaborator is
never spawned and has no profile, so a lead row's context and seat fields have
no meaning for it.

mvn clean install: exit 0, BUILD SUCCESS, Tests run: 2005, Failures: 0.

Nothing yet pins the handler to collaboratorsVisibleTo, unlike the coordinator
flag, so a literal passed there would regress silently. Tracked in #710.
2026-10-04 06:42:05 +02:00
Dai Ha e7a7711a4e fleetd #703: fleet_list reports a collaborators array so a lead can discover one
CI / shell-tests (pull_request) Failing after 11s
CI / contract (pull_request) Successful in 58s
CI / build (pull_request) Failing after 1m54s
Threads CallerResolver#collaborators() into fleet_list's canonical listFleet
overload and adds a collaborators array (name, sessionId), visible only to
the primary, an architect, and a collaborator -- the same roles Authz grants
SEND to a named peer -- never a worker. Gives CallerResolver#collaborators()
its first real caller outside the resolver itself.
2026-10-04 06:37:48 +02:00
6 changed files with 212 additions and 17 deletions
@@ -2776,9 +2776,10 @@ public record FleetConfig(
});
if (!bad.isEmpty()) {
throw new IllegalStateException("refusing to start: " + String.join("; ", bad)
+ ". Every member labelled that way would be read back as a lead or "
+ "collaborator and granted that identity's authority. Change one of the two "
+ "so member tabs cannot be confused with a lead's or collaborator's tab.");
+ ". A member labelled that way, while its pane carries no entry in the "
+ "spawned-member roster, is read back as a lead or collaborator and granted "
+ "that identity's authority. Change one of the two so member tabs cannot be "
+ "confused with a lead's or collaborator's tab.");
}
List<String> collisions = new ArrayList<>();
@@ -2877,7 +2878,8 @@ public record FleetConfig(
* the focused tab rather than its own, so it can land inside a lead's or collaborator's own
* labelled tab. {@link dev.ltms.fleet.herdr.LeadTabScanner} identifies a lead or collaborator
* purely by that tab's label — it does not exclude the member space — so a member that ends up
* there would be read back as that lead or collaborator and granted that identity's authority.
* there, while its pane carries no entry in the spawned-member roster, is read back as that
* lead or collaborator and granted that identity's authority.
*
* <p>Only an entry with a non-blank {@code tab} is in scope: one with no {@code tab} feeds
* nothing into {@link dev.ltms.fleet.herdr.LeadTabScanner}, so it creates no hazard here.
@@ -2908,8 +2910,9 @@ public record FleetConfig(
}
throw new IllegalStateException("refusing to start: profile(s) " + bad
+ " use placement: pane while fleet.leaders or fleet.collaborators names a tab. A "
+ "pane-placed member can land inside that labelled tab and be read back as the "
+ "lead or collaborator, granted that identity's authority. Set placement: tab for "
+ "pane-placed member can land inside that labelled tab, and while its pane "
+ "carries no entry in the spawned-member roster, it is read back as the lead or "
+ "collaborator and granted that identity's authority. Set placement: tab for "
+ "each named profile, or remove the tab from every fleet.leaders and "
+ "fleet.collaborators entry.");
}
@@ -560,6 +560,7 @@ public final class FleetMcp {
return listFleet(workers, sessions, messages, capacity, healthCoverage, loopHealth, quarantine, outage,
leadSeats, leadContextGauge, leadConfigDirs, callers.leads(),
callerTerminal(exchange),
callers.collaborators(), collaboratorsVisibleTo(principal(exchange)),
new CoordinationSource(leadChannel, peers),
coordinatorVisibleTo(principal(exchange)));
};
@@ -743,6 +744,21 @@ public final class FleetMcp {
return caller.isPrimary();
}
/**
* fleetd #703: who may see {@code fleet_list}'s {@code collaborators} array — a roster of the
* human-opened tabs this daemon recognises as named peers. Visible to exactly the roles that
* may {@link Authz.Action#SEND} to a named peer ({@code Authz.java}'s {@code SEND} case):
* the primary, an architect, and a collaborator (reaching another collaborator or a lead). A
* worker holds {@code READ} but can never {@code SEND} to a collaborator, so listing them to a
* worker would expose which human tabs exist on the host with no use to that caller. Split out
* for the same reason as {@link #coordinatorVisibleTo}: the decision must be unit-testable
* without fabricating an SDK {@code McpSyncServerExchange}, and the handler must call this
* named predicate rather than inlining the check.
*/
static boolean collaboratorsVisibleTo(Principal caller) {
return caller.isPrimary() || caller.isArchitect() || caller.isCollaborator();
}
/** The worker identity resolved from this call's connection, or {@code null} if the primary. */
private static String callerTerminal(McpSyncServerExchange exchange) {
Object v = exchange.transportContext().get(CALLER_TERMINAL);
@@ -1761,7 +1777,8 @@ public final class FleetMcp {
Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, loopHealth, quarantine,
OutageSource.none(), LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
OutageSource.none(), LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false);
}
/** As above, plus fleetd #201 Unit 5 cool-off facts (see {@link OutageSource}). */
@@ -1770,7 +1787,8 @@ public final class FleetMcp {
QuarantineSource quarantine, OutageSource outage,
Map<String, String> leads, String selfTerm) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, CoordinationSource.none(), false);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, CoordinationSource.none(), false);
}
/**
@@ -1787,7 +1805,8 @@ public final class FleetMcp {
QuarantineSource quarantine, Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, OutageSource.none(),
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false);
}
/** As above, plus fleetd #201 Unit 5 cool-off facts (see {@link OutageSource}). */
@@ -1796,7 +1815,8 @@ public final class FleetMcp {
QuarantineSource quarantine, OutageSource outage,
Map<String, String> leads, String selfTerm, CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false);
}
/**
@@ -1818,7 +1838,8 @@ public final class FleetMcp {
LeadSeatSource leadSeats, Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false);
}
/**
@@ -1842,7 +1863,8 @@ public final class FleetMcp {
LeadSeatSource leadSeats, Map<String, String> leads, String selfTerm,
CoordinationSource coordination, boolean callerIsPrimary) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine,
outage, leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, callerIsPrimary);
outage, leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, callerIsPrimary);
}
/**
@@ -1852,6 +1874,13 @@ public final class FleetMcp {
* matter); the one caller that matters for caching, {@code fleet_list}'s MCP handler, passes
* its own single long-lived instance instead (see {@code FleetMcp}'s {@code leadContextGauge}
* field).
*
* @param collaborators terminal_id → collaborator name, live from the resolver
* ({@link dev.ltms.fleet.auth.CallerResolver#collaborators()})
* @param collaboratorsVisible whether this caller may see the {@code collaborators} array (see
* {@link #collaboratorsVisibleTo}); every wrapper overload above
* passes {@code false}, so a test that wants the row must call this
* overload with an explicit {@code true}
*/
static McpSchema.CallToolResult listFleet(PeerLauncher workers, SessionManager sessions, MessageService messages,
CapacitySource capacity, HealthCoverageSource healthCoverage,
@@ -1860,6 +1889,7 @@ public final class FleetMcp {
LeadSeatSource leadSeats, LeadContextGauge contextGauge,
LeadConfigDirSource leadConfigDirs,
Map<String, String> leads, String selfTerm,
Map<String, String> collaborators, boolean collaboratorsVisible,
CoordinationSource coordination, boolean callerIsPrimary) {
try {
Map<String, Agent> live = workers.list().stream()
@@ -1886,6 +1916,16 @@ public final class FleetMcp {
result.put("loopHealth", Map.of(
"statusPoller", loopHealth.statusPoller().get().name(),
"sessionReaper", loopHealth.sessionReaper().get().name()));
// fleetd #703: a collaborator tab is a person's own tab, so the row is assembled and
// included only for the roles that may SEND to a named peer -- gate BEFORE assembling
// it, same reason as the coordinator row just below: the key must be absent for a
// worker, never present-and-empty.
if (collaboratorsVisible) {
result.put("collaborators", collaborators.entrySet().stream()
.sorted(Map.Entry.comparingByValue())
.map(e -> collaboratorRow(e.getKey(), e.getValue()))
.toList());
}
// fleetd #439: coordinator/coordinatorView is lead-to-lead coordination state and must
// never reach a worker or an architect -- gate BEFORE assembling it, not after, so the
// key is absent rather than present-and-empty.
@@ -2197,6 +2237,20 @@ public final class FleetMcp {
return m;
}
/**
* fleetd #703: one row of the {@code collaborators} array — a named peer's registry name and
* the herdr {@code terminal_id} a {@code fleet_send} must target to reach it. No status, no
* context gauge, no profile: a collaborator is never spawned and carries no profile, so those
* fields have no meaning for it, and the scan behind {@code terminal} only reports a tab it
* actually found in herdr, so a tab nobody has open does not appear here at all.
*/
private static Map<String, Object> collaboratorRow(String terminal, String name) {
Map<String, Object> m = new LinkedHashMap<>();
m.put("name", name);
m.put("sessionId", terminal);
return m;
}
/**
* Reads the lead context gauge for one lead. {@code configDir} is this lead's configured
* {@code CLAUDE_CONFIG_DIR} override (see {@link LeadConfigDirSource}), derived from
@@ -2422,7 +2476,12 @@ public final class FleetMcp {
+ "msgId/from/preview, never the full body), and one row per coordinator.peers "
+ "coord-id ('peers': coordId/reachable, plus pending/consumers when reachable) — "
+ "this is peer DISCOVERY for cross-host leads, distinct from the local 'leads' "
+ "array above. It is omitted entirely when no coordinator is configured.",
+ "array above. It is omitted entirely when no coordinator is configured. A "
+ "'collaborators' array, visible only to the primary, an architect, and a "
+ "collaborator (never a worker), reports every other named peer tab this daemon "
+ "recognises: each row is 'name' (its registry name) and 'sessionId' (the "
+ "terminal id a fleet_send targets to reach it). Absent entirely for a worker, "
+ "whatever is configured.",
objectSchema(Map.of(), List.of()));
}
@@ -850,7 +850,8 @@ class FleetConfigTest {
/**
* The hazard this guard closes: a pane-placed member lands inside the focused tab rather than
* its own, so it can land inside a lead's labelled tab and be read back as that lead.
* its own, so it can land inside a lead's labelled tab and, while its pane carries no entry in
* the spawned-member roster, be read back as that lead.
*/
@Test
void aPanePlacedProfileWithALeadTabRefusesToStart(@TempDir Path dir) throws Exception {
@@ -1087,8 +1088,9 @@ class FleetConfigTest {
}
/**
* fleetd #669: a member tabLabel that can render as a configured collaborator tab is the same
* hazard as the lead case above — a member labelled that way is read back as the collaborator.
* A member tabLabel that can render as a configured collaborator tab is the same hazard as the
* lead case above — while its pane carries no entry in the spawned-member roster, a member
* labelled that way is read back as the collaborator.
*/
@Test
void aProfileTabLabelOverrideMatchingACollaboratorTabRefusesToStart(@TempDir Path dir)
@@ -366,6 +366,58 @@ class FleetMcpAuthzTest {
+ "calling, not pass a literal boolean -- found: " + trailing);
}
// --- fleetd #703: who may see fleet_list's collaborators array -------------------------------
/**
* fleetd #703: {@link FleetMcp#collaboratorsVisibleTo} is the whole policy decision for
* {@code fleet_list}'s {@code collaborators} array. Visible to exactly the roles that may
* {@code SEND} to a named peer -- the primary, an architect, and a collaborator itself -- never
* a worker, which holds {@code READ} but can never {@code SEND} to a collaborator, and never an
* anonymous caller.
*/
@Test
void onlyPrimaryArchitectAndCollaboratorMaySeeTheCollaboratorsArray() {
assertTrue(FleetMcp.collaboratorsVisibleTo(PRIMARY), "the primary must see the collaborators array");
assertTrue(FleetMcp.collaboratorsVisibleTo(ARCH_DESIGN), "an architect must see the collaborators array");
assertTrue(FleetMcp.collaboratorsVisibleTo(COLLABORATOR), "a collaborator must see its own peer roster");
assertFalse(FleetMcp.collaboratorsVisibleTo(WORKER_A),
"a worker holds READ but can never SEND to a collaborator, so it must not see the array");
assertFalse(FleetMcp.collaboratorsVisibleTo(ANON), "authenticated as nothing must not see it either");
}
/**
* fleetd #703, same reasoning as {@link #theFleetListHandlerActuallyConsultsCoordinatorVisibleTo}:
* the predicate above can be perfectly correct while the one production call site never asks it.
* This reads {@code FleetMcp.java}'s own source and asserts the {@code fleet_list} handler's
* {@code listFleet(...)} call both threads {@code callers.collaborators()} into the payload and
* asks {@code collaboratorsVisibleTo(principal(exchange))} for the visibility flag, rather than a
* literal boolean or an empty map.
*/
@Test
void theFleetListHandlerActuallyConsultsCollaboratorsVisibleTo() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("listHandler =");
assertTrue(start >= 0, "could not find the fleet_list handler (listHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("stopHandler =", start);
assertTrue(end > start, "could not find the handler declared after listHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does contain a call to listFleet(...) -- if this
// fails, the anchors above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("listFleet("),
"control failed: the scraped listHandler block contains no listFleet( call at all -- "
+ "the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("callers.collaborators()"),
"the fleet_list handler must thread callers.collaborators() into listFleet(...), not an "
+ "empty or literal map -- block: " + handlerBlock);
assertTrue(handlerBlock.contains("collaboratorsVisibleTo(principal(exchange))"),
"the fleet_list handler must ask collaboratorsVisibleTo(principal(exchange)) who is "
+ "calling, not pass a literal boolean -- block: " + handlerBlock);
}
// --- which action each tool hands the gate (fleetd #272) ------------------------------------
/**
@@ -121,6 +121,7 @@ class FleetMcpLeadContextGaugeWiringTest {
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.LoopHealthSource.none(), FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(),
FleetMcp.LeadSeatSource.none(), contextGauge, leadConfigDirs,
Map.of(LEAD_TERMINAL, LEAD_NAME), LEAD_TERMINAL, FleetMcp.CoordinationSource.none(), false);
Map.of(LEAD_TERMINAL, LEAD_NAME), LEAD_TERMINAL, Map.of(), false,
FleetMcp.CoordinationSource.none(), false);
}
}
@@ -10,6 +10,7 @@ import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.AgentStatus;
import dev.ltms.fleet.herdr.FakeHerdr;
import dev.ltms.fleet.inject.LoopWatchdog;
import dev.ltms.fleet.lead.LeadContextGauge;
import dev.ltms.fleet.herdr.PaneLocator;
import dev.ltms.fleet.herdr.WorkspaceControl;
import dev.ltms.fleet.inject.Injector;
@@ -879,6 +880,83 @@ class FleetMcpTest {
assertTrue(out.contains("x".repeat(80) + "…"), "expected an 80-char preview with an ellipsis: " + out);
}
// --- fleetd #703: fleet_list's collaborators array -------------------------------------------
/** Calls the canonical {@code listFleet} overload directly, so a test can set the collaborators
* payload and its visibility independently of a real {@code Principal} / MCP exchange. */
private static McpSchema.CallToolResult listFleetWithCollaborators(FakeHerdr h,
Map<String, String> collaborators, boolean collaboratorsVisible) {
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
return FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.LoopHealthSource.none(), FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(),
FleetMcp.LeadSeatSource.none(), new LeadContextGauge(), FleetMcp.LeadConfigDirSource.none(),
Map.of(), "", collaborators, collaboratorsVisible,
FleetMcp.CoordinationSource.none(), false);
}
/**
* fleetd #703 acceptance A: a visible caller with one configured collaborator gets a
* {@code collaborators} row whose key ({@code CallerResolver.collaborators()}'s
* {@code terminal_id -> name} entry) lands as that row's {@code sessionId}, and {@code leads}/
* {@code members} are unaffected by the new key.
*/
@Test
void listReportsACollaboratorsRowKeyedByTheCollaboratorsTerminalId() {
FakeHerdr h = new FakeHerdr();
String out = textOf(listFleetWithCollaborators(h, Map.of("term_collab", "ops"), true));
assertTrue(out.contains("\"collaborators\":["), out);
assertTrue(out.contains("\"name\":\"ops\""), out);
assertTrue(out.contains("\"sessionId\":\"term_collab\""), out);
assertTrue(out.contains("\"leads\":[]"), out);
assertTrue(out.contains("\"members\":[]"), out);
}
/**
* fleetd #703 acceptance A, control half: with no collaborator configured, the key is absent
* (caller cannot see it) or an empty array (caller can), and {@code leads}/{@code members} are
* unchanged either way.
*/
@Test
void listOmitsOrEmptiesCollaboratorsWhenNoneAreConfigured() {
FakeHerdr h = new FakeHerdr();
String visibleButEmpty = textOf(listFleetWithCollaborators(h, Map.of(), true));
assertTrue(visibleButEmpty.contains("\"collaborators\":[]"), visibleButEmpty);
assertTrue(visibleButEmpty.contains("\"leads\":[]"), visibleButEmpty);
assertTrue(visibleButEmpty.contains("\"members\":[]"), visibleButEmpty);
String notVisible = textOf(listFleetWithCollaborators(h, Map.of(), false));
assertFalse(notVisible.contains("\"collaborators\""), notVisible);
assertTrue(notVisible.contains("\"leads\":[]"), notVisible);
assertTrue(notVisible.contains("\"members\":[]"), notVisible);
}
/**
* fleetd #703 acceptance B: a worker must not see the {@code collaborators} array at all, while
* an architect -- a role that also holds READ, same as a worker -- does see it. Both halves are
* asserted: a test that only checked the worker-hidden half would pass even if the feature were
* never wired up for anyone.
*/
@Test
void listOmitsCollaboratorsForAWorkerAndIncludesThemForAnArchitect() {
FakeHerdr h = new FakeHerdr();
Map<String, String> collaborators = Map.of("term_collab", "ops");
String asWorker = textOf(listFleetWithCollaborators(h, collaborators,
FleetMcp.collaboratorsVisibleTo(Principal.worker("term_w", 1))));
assertFalse(asWorker.contains("\"collaborators\""),
"a worker must not see the collaborators array: " + asWorker);
String asArchitect = textOf(listFleetWithCollaborators(h, collaborators,
FleetMcp.collaboratorsVisibleTo(Principal.architect("design", "term_arch", 2))));
assertTrue(asArchitect.contains("\"collaborators\":["),
"an architect must see the collaborators array: " + asArchitect);
assertTrue(asArchitect.contains("\"sessionId\":\"term_collab\""), asArchitect);
}
/**
* fleetd #421: {@code mailbox.pending} counts only broker-ready messages, so a blocked lead's
* normal, healthy state is {@code "pending": 0} next to a non-empty {@code held[]} — which