hive-subagent-mcp: check subagent models against the swarm's subagent_models

A `start` or `continue` naming a `model` now reads this agent's key in the
`agent-inference` bucket and, when `subagent_models` is set, checks the
model against that list instead of HIVE_AVAILABLE_MODELS (agent.nix
`availableModels`). An unset field, an absent key or an absent bucket
keeps today's check. A read that fails or takes over 2 s falls back to
HIVE_AVAILABLE_MODELS with a warning, so a subagent start never waits on
the queue. A call without a `model` is not checked and does not read.

The read uses the agent's own queue credential through swarm_term's
connect, and the one direct-get subject the agent is granted in that
bucket. The refusal names the list and that it came from the swarm.

Refs #5034 (S5)

docs/swarm/inference.md

@@ -16,8 +16,11 @@ swarmctl agent inference clear iris
 - `set` changes only the flags you pass; the other fields keep what's
   stored. A hive must declare the agent.
 - `--effort` is one of `low`, `medium`, `high`, `xhigh`, `max`.
-- `--subagent-models` takes a comma-separated list. It's stored and shown,
+- `--subagent-models` takes a comma-separated list. It replaces the
-  but agents don't apply it yet.
+  agent's `availableModels` as the models its subagent `start` and
+  `continue` calls may name, read on each call
+  ([`subagent.md`](../tools/subagent.md#which-models-a-subagent-may-run-on)).
+  The web UI's model picker doesn't use it.
 - `get` shows each field, naming the unset ones.
 - `clear` removes every field; the agent goes back to its own values.
 
@@ -65,4 +68,8 @@ The harness (`hive-agent`'s `swarm_inference`) applies a set field to the
 in-memory model or effort without writing `hyperhive-model` or
 `hyperhive-effort`, which keep the agent's own value.
 
+The subagent daemon (`hive-subagent-mcp`'s `swarm_models`) reads
+`subagent_models` with the same direct get, under the same credential, on
+each `start` or `continue` that names a model.
+
 </details>

docs/tools/subagent.md

@@ -307,6 +307,28 @@ agent that needs a stable or non-default port.
 Own systemd unit, defined alongside the other per-agent MCP daemons in
 `nix/agent-modules/mcp.nix`.
 
+## Which models a subagent may run on
+
+The daemon refuses a `start` or `continue` that names a `model` outside the
+agent's list before anything spawns, and the error names the list and where
+it came from. One without a `model` runs on claude's own default, which the
+daemon doesn't check. The list is:
+
+- the swarm's `subagent_models` for this agent, when it sets one
+  ([`inference.md`](../swarm/inference.md)). It replaces the agent's own
+  list, so it can name models that list leaves out.
+- otherwise `services.hyperhive.agent.availableModels`, which reaches the
+  daemon as `HIVE_AVAILABLE_MODELS`. An empty list restricts nothing.
+
+The daemon reads the swarm's value on each such call, so a change applies
+to the next one. When the read fails, or doesn't answer within 2 seconds,
+the daemon logs a warning and checks against `availableModels`. The same
+happens until the daemon's first connection to the swarm queue succeeds.
+The match ignores case. On an ACP agent the names are the agent's own, such
+as `<provider>/<model>` on `opencode`.
+
+The web UI's model picker still offers `availableModels`.
+
 ## Runtime
 
 A subagent runs on its parent agent's runtime

hive-subagent-mcp/Cargo.toml

@@ -29,7 +29,8 @@ schemars.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 # `subagent_term`: the subject each subagent's rows are published on.
-swarm-queue-client.workspace = true
+# `agent-inference`: this agent's `subagent_models`, read with one direct get.
+swarm-queue-client = { workspace = true, features = ["agent-inference"] }
 # The agent's own queue credential, read under its store identity.
 swarm-secret-client.workspace = true
 tokio.workspace = true

hive-subagent-mcp/src/lib.rs

@@ -21,4 +21,5 @@ pub mod mcp_config;
 pub mod paths;
 pub mod role;
 pub mod session;
+pub mod swarm_models;
 pub mod swarm_term;

hive-subagent-mcp/src/main.rs

@@ -56,7 +56,8 @@ async fn main() -> Result<()> {
     let state = Arc::new(
         hive_subagent_mcp::session::State::new(todo_socket, signal_base)
             .on_runtime(runtime)
-            .publishing_to(hive_subagent_mcp::swarm_term::Publisher::spawn()),
+            .publishing_to(hive_subagent_mcp::swarm_term::Publisher::spawn())
+            .reading_models_from(hive_subagent_mcp::swarm_models::Reader::spawn()),
     );
 
     // Serve the MCP tools over streamable-http forever. No background poll

hive-subagent-mcp/src/mcp.rs

@@ -216,7 +216,7 @@ impl SubagentMcp {
             you do not have refuses the whole call rather than spawning without it. See the \
             `base:claude-subagents` skill for when to reach for this."
     )]
-    fn start(&self, Parameters(args): Parameters<StartArgs>) -> String {
+    async fn start(&self, Parameters(args): Parameters<StartArgs>) -> String {
         match session::start(
             &self.state,
             session::StartRequest {
@@ -231,7 +231,9 @@ impl SubagentMcp {
                 max_turns: args.max_turns,
                 report_file: args.report_file,
             },
-        ) {
+        )
+        .await
+        {
             Ok(msg) => msg,
             Err(e) => format!("start error: {e:#}"),
         }

hive-subagent-mcp/src/session.rs

@@ -386,6 +386,9 @@ pub struct State {
     /// Where every subagent's output lines go up to the swarm, when this
     /// agent can publish there ([`State::publishing_to`]).
     term: Option<crate::swarm_term::Publisher>,
+    /// Where the swarm's `subagent_models` for this agent is read from, when
+    /// this agent can read it ([`State::reading_models_from`]).
+    models: Option<crate::swarm_models::Reader>,
 }
 
 /// Which opaque URL segment belongs to which session — the whole of a
@@ -439,6 +442,7 @@ impl State {
             signal_tokens: Mutex::new(SignalTokens::default()),
             runtime: SubagentRuntime::Claude,
             term: None,
+            models: None,
         }
     }
 
@@ -449,6 +453,27 @@ impl State {
         self
     }
 
+    /// Check every named `model` against the swarm's `subagent_models` as
+    /// `models` reads it, where the swarm sets one.
+    #[must_use]
+    pub fn reading_models_from(mut self, models: Option<crate::swarm_models::Reader>) -> Self {
+        self.models = models;
+        self
+    }
+
+    /// Refuse a `model` outside the list in effect for this agent (see
+    /// `check_model`). The swarm key is read only for a named model.
+    async fn allow_model(&self, model: Option<&str>) -> anyhow::Result<()> {
+        if model.is_none() {
+            return Ok(());
+        }
+        let read = match &self.models {
+            Some(reader) => Some(reader.read().await),
+            None => None,
+        };
+        check_model(model, &model_list(read, available_models()))
+    }
+
     /// Run every subagent on `runtime` — the daemon passes the parent
     /// agent's, read from its environment at startup. On ACP this resolves
     /// the harness dir, which only exists inside a container.
@@ -918,28 +943,69 @@ fn available_models() -> Option<String> {
     std::env::var(AVAILABLE_MODELS_ENV).ok()
 }
 
+/// The list a named `model` is checked against.
+#[derive(Debug, PartialEq, Eq)]
+enum ModelList {
+    /// The swarm's `subagent_models` for this agent.
+    Swarm(Vec<String>),
+    /// [`AVAILABLE_MODELS_ENV`]; `None` when the var is absent.
+    Local(Option<String>),
+}
+
+/// The swarm's list when it sets one, else `local`. `read` is `None` when
+/// this daemon has no swarm queue to read from; a failed read is logged and
+/// falls back to `local`.
+fn model_list(read: Option<crate::swarm_models::Read>, local: Option<String>) -> ModelList {
+    use crate::swarm_models::Read;
+    match read {
+        Some(Read::Set(models)) => ModelList::Swarm(models),
+        Some(Read::Failed(error)) => {
+            tracing::warn!(
+                %error,
+                "reading this agent's subagent_models from the swarm failed; checking against \
+                 {AVAILABLE_MODELS_ENV}"
+            );
+            ModelList::Local(local)
+        }
+        Some(Read::Unset) | None => ModelList::Local(local),
+    }
+}
+
 /// Refuse a `model` the operator hasn't made available to this agent.
 ///
 /// Default-open on purpose: this is a safety rail, not a security boundary,
-/// so an absent `allowed` (or one naming nothing) restricts nothing and an
+/// so an absent list (or one naming nothing) restricts nothing and an
 /// agent deployed before this gate existed keeps working unchanged. A `None`
 /// model isn't a choice of model either — it lets claude fall back to its own
 /// default — so there is nothing to check.
 ///
 /// # Errors
 ///
-/// A named model absent from `allowed`; the message names it and every
+/// A named model absent from `list`; the message names it, every permitted
-/// permitted one, since it is all the caller sees.
+/// one and where the list came from, since it is all the caller sees.
-fn check_model(model: Option<&str>, allowed: Option<&str>) -> anyhow::Result<()> {
+fn check_model(model: Option<&str>, list: &ModelList) -> anyhow::Result<()> {
-    let (Some(model), Some(allowed)) = (model, allowed) else {
+    let Some(model) = model else {
         return Ok(());
     };
-    // Same split/trim/skip-empty rules as `ToolGroup::parse_list`.
+    let (allowed, source): (Vec<&str>, &str) = match list {
-    let allowed: Vec<&str> = allowed
+        ModelList::Swarm(models) => (
-        .split(',')
+            models
-        .map(str::trim)
+                .iter()
-        .filter(|token| !token.is_empty())
+                .map(|m| m.trim())
-        .collect();
+                .filter(|m| !m.is_empty())
+                .collect(),
+            "the swarm's subagent_models for this agent",
+        ),
+        ModelList::Local(None) => return Ok(()),
+        // Same split/trim/skip-empty rules as `ToolGroup::parse_list`.
+        ModelList::Local(Some(env)) => (
+            env.split(',')
+                .map(str::trim)
+                .filter(|token| !token.is_empty())
+                .collect(),
+            AVAILABLE_MODELS_ENV,
+        ),
+    };
     if allowed.is_empty()
         || allowed
             .iter()
@@ -948,7 +1014,7 @@ fn check_model(model: Option<&str>, allowed: Option<&str>) -> anyhow::Result<()>
         return Ok(());
     }
     anyhow::bail!(
-        "model `{model}` is not available to this agent — {AVAILABLE_MODELS_ENV} permits: {}",
+        "model `{model}` is not available to this agent — {source} permits: {}",
         allowed.join(", ")
     );
 }
@@ -1207,10 +1273,10 @@ pub struct StartRequest {
 /// case is the only one that can happen *after* commit-to-run, and it's
 /// exactly why nothing is registered in `running` until spawn actually
 /// succeeds.
-pub fn start(state: &Arc<State>, req: StartRequest) -> anyhow::Result<String> {
+pub async fn start(state: &Arc<State>, req: StartRequest) -> anyhow::Result<String> {
     let name = req.name.as_str();
     validate_name(name)?;
-    check_model(req.model.as_deref(), available_models().as_deref())?;
+    state.allow_model(req.model.as_deref()).await?;
     // Before `reserve`, so a role this agent does not have costs the caller
     // an error and nothing else: no name claimed, no session archived, no
     // process spawned.
@@ -1388,7 +1454,7 @@ pub async fn continue_(
     dir: Option<&str>,
 ) -> anyhow::Result<String> {
     validate_name(name)?;
-    check_model(model.as_deref(), available_models().as_deref())?;
+    state.allow_model(model.as_deref()).await?;
     if !state.reserve(name) {
         anyhow::bail!(
             "subagent `{name}` is already running — use `interrupt` first if you meant to redirect it"
@@ -2626,15 +2692,17 @@ mod tests {
         assert!(err.to_string().contains("task instructions"), "{err}");
     }
 
-    #[test]
+    #[tokio::test]
-    fn a_start_naming_an_unknown_role_errors_and_reserves_nothing() {
+    async fn a_start_naming_an_unknown_role_errors_and_reserves_nothing() {
         // The error path end to end: the refusal happens before anything is
         // claimed, so an unknown role costs a message and nothing else —
         // no reservation to release, no archived prior session, no spawn.
         let state = Arc::new(State::new(PathBuf::from("/dev/null"), signal_url()));
         let mut req = start_request("unknown-role-run");
         req.role = Some("no-such-role-ships-anywhere".to_owned());
-        let err = start(&state, req).expect_err("a role this agent lacks must refuse the start");
+        let err = start(&state, req)
+            .await
+            .expect_err("a role this agent lacks must refuse the start");
         assert!(
             err.to_string().contains("no-such-role-ships-anywhere"),
             "the refusal names the role that was asked for: {err}"
@@ -2646,12 +2714,14 @@ mod tests {
         );
     }
 
-    #[test]
+    #[tokio::test]
-    fn a_start_naming_an_unusable_role_refuses_before_touching_the_filesystem() {
+    async fn a_start_naming_an_unusable_role_refuses_before_touching_the_filesystem() {
         let state = Arc::new(State::new(PathBuf::from("/dev/null"), signal_url()));
         let mut req = start_request("bad-role-name");
         req.role = Some("../../etc/passwd".to_owned());
-        let err = start(&state, req).expect_err("a role name that is not an identifier refuses");
+        let err = start(&state, req)
+            .await
+            .expect_err("a role name that is not an identifier refuses");
         assert!(err.to_string().contains("invalid role name"), "{err}");
         assert_eq!(state.occupancy("bad-role-name"), None);
     }
@@ -2809,8 +2879,8 @@ mod tests {
         );
     }
 
-    #[test]
+    #[tokio::test]
-    fn start_does_not_corrupt_the_remembered_dir_when_the_name_is_already_running() {
+    async fn start_does_not_corrupt_the_remembered_dir_when_the_name_is_already_running() {
         // The exact scenario argus caught in review: a rejected concurrent
         // `start` used to commit its `dir` before the reservation check
         // refused it, corrupting what a later bare `continue` would resolve
@@ -2829,7 +2899,8 @@ mod tests {
                 dir: Some("/tmp/rejected".to_owned()),
                 ..start_request("dup")
             },
-        );
+        )
+        .await;
         assert!(
             result.is_err(),
             "a name already reserved must be refused, not spawned"
@@ -4000,8 +4071,8 @@ mod tests {
         );
     }
 
-    #[test]
+    #[tokio::test]
-    fn a_fresh_start_inherits_nothing_from_the_run_before_it() {
+    async fn a_fresh_start_inherits_nothing_from_the_run_before_it() {
         // `start` archives the prior session precisely so this is a fresh
         // start; a leftover stop reason or report path would make the new run
         // report the old one's ending, into the old one's file.
@@ -4009,7 +4080,7 @@ mod tests {
         state.record_stop("n", StopReason::NeedHelp("old block".to_owned()));
         state.set_report_file("n", Some("/tmp/old-report.md"));
         state.reserve("n");
-        let refused = start(&state, start_request("n"));
+        let refused = start(&state, start_request("n")).await;
         assert!(refused.is_err(), "the reserved name is refused as before");
 
         // The same clearing `start` does once it owns the name.
@@ -4139,22 +4210,22 @@ mod tests {
 
     #[test]
     fn an_unset_model_whitelist_allows_any_model() {
-        assert!(check_model(Some("opus"), None).is_ok());
+        assert!(check_model(Some("opus"), &local(None)).is_ok());
     }
 
     #[test]
     fn an_unset_model_whitelist_allows_an_omitted_model() {
-        assert!(check_model(None, None).is_ok());
+        assert!(check_model(None, &local(None)).is_ok());
     }
 
     #[test]
     fn a_listed_model_is_allowed() {
-        assert!(check_model(Some("sonnet"), Some("haiku,sonnet,opus")).is_ok());
+        assert!(check_model(Some("sonnet"), &local(Some("haiku,sonnet,opus"))).is_ok());
     }
 
     #[test]
     fn an_unlisted_model_is_rejected_naming_what_is_permitted() {
-        let err = check_model(Some("opus"), Some("haiku,sonnet"))
+        let err = check_model(Some("opus"), &local(Some("haiku,sonnet")))
             .expect_err("a model outside the list must be refused")
             .to_string();
         assert!(
@@ -4167,7 +4238,7 @@ mod tests {
     #[test]
     fn a_set_whitelist_still_allows_an_omitted_model() {
         assert!(
-            check_model(None, Some("haiku")).is_ok(),
+            check_model(None, &local(Some("haiku"))).is_ok(),
             "omitting `model` lets claude pick its own default — it is not a \
              choice of model, so there is nothing to gate"
         );
@@ -4175,14 +4246,111 @@ mod tests {
 
     #[test]
     fn whitespace_and_empty_entries_are_ignored() {
-        assert!(check_model(Some("sonnet"), Some(" haiku , sonnet ,, ")).is_ok());
+        assert!(check_model(Some("sonnet"), &local(Some(" haiku , sonnet ,, "))).is_ok());
-        assert!(check_model(Some("opus"), Some(" haiku , sonnet ,, ")).is_err());
+        assert!(check_model(Some("opus"), &local(Some(" haiku , sonnet ,, "))).is_err());
         assert!(
-            check_model(Some("opus"), Some("  ,, ")).is_ok(),
+            check_model(Some("opus"), &local(Some("  ,, "))).is_ok(),
             "a var naming nothing restricts nothing — default open"
         );
     }
 
+    fn local(env: Option<&str>) -> ModelList {
+        ModelList::Local(env.map(str::to_owned))
+    }
+
+    fn swarm(models: &[&str]) -> crate::swarm_models::Read {
+        crate::swarm_models::Read::Set(models.iter().map(|m| (*m).to_owned()).collect())
+    }
+
+    /// The swarm's list replaces the local one, so it can widen it as well
+    /// as narrow it.
+    #[test]
+    fn a_swarm_list_replaces_the_local_one() {
+        let env = || Some("haiku,sonnet".to_owned());
+        let list = model_list(Some(swarm(&["opus"])), env());
+        assert_eq!(list, ModelList::Swarm(vec!["opus".to_owned()]));
+        assert!(check_model(Some("opus"), &list).is_ok());
+        assert!(check_model(Some("sonnet"), &list).is_err());
+    }
+
+    #[test]
+    fn without_a_swarm_list_the_local_one_applies() {
+        let env = || Some("haiku,sonnet".to_owned());
+        for read in [None, Some(crate::swarm_models::Read::Unset)] {
+            assert_eq!(model_list(read, env()), local(Some("haiku,sonnet")));
+        }
+        assert_eq!(
+            model_list(Some(crate::swarm_models::Read::Unset), None),
+            local(None)
+        );
+    }
+
+    #[test]
+    fn a_failed_read_falls_back_to_the_local_list_and_warns() {
+        let logs = CapturedLogs::default();
+        let subscriber = tracing_subscriber::fmt()
+            .with_writer(logs.clone())
+            .with_ansi(false)
+            .finish();
+        let list = tracing::subscriber::with_default(subscriber, || {
+            model_list(
+                Some(crate::swarm_models::Read::Failed("no answer".to_owned())),
+                Some("haiku".to_owned()),
+            )
+        });
+        assert_eq!(list, local(Some("haiku")));
+        let logged = logs.text();
+        assert!(
+            logged.contains("WARN") && logged.contains("no answer"),
+            "the fallback is logged as a warning with the cause: {logged}"
+        );
+    }
+
+    #[test]
+    fn a_model_outside_the_swarm_list_is_refused_naming_the_list_and_its_source() {
+        let list = model_list(Some(swarm(&["haiku", "sonnet"])), Some("opus".to_owned()));
+        let err = check_model(Some("opus"), &list)
+            .expect_err("a model outside the swarm's list must be refused")
+            .to_string();
+        assert!(
+            err.contains("opus")
+                && err.contains("haiku, sonnet")
+                && err.contains("swarm's subagent_models"),
+            "{err}"
+        );
+        assert!(
+            check_model(None, &list).is_ok(),
+            "an omitted model is not checked against the swarm's list either"
+        );
+    }
+
+    /// A `tracing` writer whose output a test can read back.
+    #[derive(Clone, Default)]
+    struct CapturedLogs(Arc<Mutex<Vec<u8>>>);
+
+    impl CapturedLogs {
+        fn text(&self) -> String {
+            String::from_utf8_lossy(&self.0.lock().unwrap()).into_owned()
+        }
+    }
+
+    impl std::io::Write for CapturedLogs {
+        fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
+            self.0.lock().unwrap().extend_from_slice(buf);
+            Ok(buf.len())
+        }
+        fn flush(&mut self) -> std::io::Result<()> {
+            Ok(())
+        }
+    }
+
+    impl<'a> tracing_subscriber::fmt::MakeWriter<'a> for CapturedLogs {
+        type Writer = Self;
+        fn make_writer(&'a self) -> Self::Writer {
+            self.clone()
+        }
+    }
+
     /// An ACP agent in plain `sh`, after `hive-runtime`'s own test agent. It
     /// has one session, `s1`, appends every method it is sent to
     /// `<log>.methods` and every `session/prompt` line to `<log>`, and

hive-subagent-mcp/src/swarm_models.rs

@@ -0,0 +1,144 @@
+//! Reads the subagent model list the swarm sets for this agent.
+//!
+//! swarm-controller keeps the agent's settings under its key in the
+//! `agent-inference` KV bucket. Its `subagent_models`, when set, is the list
+//! a `start` or `continue` naming a `model` is checked against, in place of
+//! `HIVE_AVAILABLE_MODELS` (`crate::session::check_model`).
+//!
+//! **Read on every such call**, never cached, with the one subject the
+//! agent's own queue credential is granted in that bucket
+//! (`swarm_queue_client::agent_inference::read_own`). The connection is made
+//! the way [`crate::swarm_term`] makes its own: as the agent, never as the
+//! hive.
+//!
+//! **A subagent start never waits on the queue.** Before the first connect
+//! succeeds, and when a read fails or outlasts `READ_TIMEOUT`, the read is
+//! [`Read::Failed`] and the caller checks against `HIVE_AVAILABLE_MODELS`.
+
+use std::sync::{Arc, OnceLock};
+use std::time::Duration;
+
+use swarm_queue_client::agent_inference::{self, read_own};
+
+use crate::swarm_term::{Target, connect};
+
+/// How long after a failed connect the task tries again.
+const RETRY_AFTER: Duration = Duration::from_mins(1);
+
+/// How long one read of the key may take. A read the queue refuses for
+/// permissions reaches the client as a timeout, so this is also how long a
+/// misgranted agent's `start` waits.
+const READ_TIMEOUT: Duration = Duration::from_secs(2);
+
+/// What one read of this agent's key said about `subagent_models`.
+#[derive(Debug, PartialEq, Eq)]
+pub enum Read {
+    /// The swarm sets the list.
+    Set(Vec<String>),
+    /// The key, the field, or the bucket is absent.
+    Unset,
+    /// Nothing could be read, and why.
+    Failed(String),
+}
+
+/// The handle `start`/`continue` read through.
+pub struct Reader {
+    agent: String,
+    /// Set once the first connect succeeds; async-nats reconnects it after.
+    client: Arc<OnceLock<async_nats::Client>>,
+}
+
+impl Reader {
+    /// Start connecting, or return `None` when this daemon has no queue
+    /// address or no store to read the agent's credential from. Logs which,
+    /// once.
+    #[must_use]
+    pub fn spawn() -> Option<Self> {
+        let target = match Target::from_lookup(|k| std::env::var(k).ok()) {
+            Ok(Some(target)) => target,
+            Ok(None) => {
+                tracing::info!(
+                    "no swarm queue address or no secret store for this agent; subagent models \
+                     are checked against HIVE_AVAILABLE_MODELS only"
+                );
+                return None;
+            }
+            Err(e) => {
+                tracing::warn!(
+                    error = format!("{e:#}"),
+                    "this agent's secret store coordinates are incomplete; subagent models are \
+                     checked against HIVE_AVAILABLE_MODELS only"
+                );
+                return None;
+            }
+        };
+        swarm_queue_client::install_crypto_provider();
+        let reader = Self {
+            agent: target.agent.clone(),
+            client: Arc::new(OnceLock::new()),
+        };
+        tokio::spawn(connect_until_up(
+            Arc::new(target),
+            Arc::clone(&reader.client),
+        ));
+        Some(reader)
+    }
+
+    /// This agent's `subagent_models` as the swarm holds it now.
+    pub async fn read(&self) -> Read {
+        let Some(client) = self.client.get() else {
+            return Read::Failed("not connected to the swarm queue yet".to_owned());
+        };
+        // An unconnected client does not fail a JetStream request, it hangs on it.
+        if let Err(e) = swarm_queue_client::ensure_connected(client) {
+            return Read::Failed(swarm_queue_client::chain(&e));
+        }
+        match tokio::time::timeout(READ_TIMEOUT, read_own(client, &self.agent)).await {
+            Err(_) => Read::Failed(format!("no answer within {READ_TIMEOUT:?}")),
+            Ok(Err(e)) => Read::Failed(e.to_string()),
+            Ok(Ok(agent_inference::Read::NoBucket)) => {
+                tracing::debug!("subagent models: the agent-inference bucket does not answer");
+                Read::Unset
+            }
+            Ok(Ok(agent_inference::Read::Current(current))) => {
+                current.subagent_models.map_or(Read::Unset, Read::Set)
+            }
+        }
+    }
+}
+
+/// Connect, retrying every [`RETRY_AFTER`] until one succeeds.
+async fn connect_until_up(target: Arc<Target>, client: Arc<OnceLock<async_nats::Client>>) {
+    loop {
+        match connect(&target).await {
+            Ok(connected) => {
+                tracing::info!(url = %target.url, "subagent models: connected as this agent");
+                let _already_set = client.set(connected);
+                return;
+            }
+            Err(e) => {
+                tracing::warn!(
+                    error = format!("{e:#}"),
+                    retry_in = ?RETRY_AFTER,
+                    "subagent models: connecting with this agent's own queue credential failed; \
+                     HIVE_AVAILABLE_MODELS applies until a connect succeeds"
+                );
+                tokio::time::sleep(RETRY_AFTER).await;
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn a_reader_not_yet_connected_reads_as_failed() {
+        let reader = Reader {
+            agent: "iris".to_owned(),
+            client: Arc::new(OnceLock::new()),
+        };
+        assert!(matches!(reader.read().await, Read::Failed(_)));
+    }
+}

hive-subagent-mcp/src/swarm_term.rs

@@ -149,9 +149,9 @@ impl Publisher {
 }
 
 /// Where to publish, and what to connect with.
-struct Target {
+pub(crate) struct Target {
-    url: String,
+    pub(crate) url: String,
-    agent: String,
+    pub(crate) agent: String,
     store: StoreSource,
 }
 
@@ -168,7 +168,9 @@ impl Target {
     ///
     /// # Errors
     /// A store address without the agent name or the identity beside it.
-    fn from_lookup(get: impl Fn(&str) -> Option<String>) -> anyhow::Result<Option<Self>> {
+    pub(crate) fn from_lookup(
+        get: impl Fn(&str) -> Option<String>,
+    ) -> anyhow::Result<Option<Self>> {
         let Some(url) = get(ENV_NATS_URL).filter(|v| !v.is_empty()) else {
             return Ok(None);
         };
@@ -225,7 +227,7 @@ impl StoreSource {
 
 /// Connect presenting the agent's own token, read from the store on every
 /// connection attempt so a re-minted secret is picked up on reconnect.
-async fn connect(target: &Arc<Target>) -> anyhow::Result<async_nats::Client> {
+pub(crate) async fn connect(target: &Arc<Target>) -> anyhow::Result<async_nats::Client> {
     let url = target.url.clone();
     let target = Arc::clone(target);
     let agent = target.agent.clone();