feat(#2290): split ensure_runtime — dirs to lifecycle, listeners to mcp_sockets supervisor

- lifecycle::ensure_agent_runtime_dir(name): pure filesystem op, no
  Coordinator dep. Creates /run/hyperhive/agents/<name> without touching
  the MCP listener map.

- workers/mcp_sockets::spawn_poll(coord): 10 s reconcile loop (same shape
  as agent_sockets::spawn_poll). Converges 'agent running => MCP listener
  bound'. First tick is immediate so hive-c0re restarts re-register all
  running agents without waiting a full interval. Fixes the dead-listener-
  after-daemon-restart gap.

- All ensure_runtime() call sites updated:
  - Prebuild/Swap/WriteDropin: Coordinator::agent_dir() (pure, no IO)
  - Reconcile-Start: ensure_agent_runtime_dir + agent_dir (dir may be
    missing after reboot; listener deferred to supervisor)
  - run_create / handle_spawn: ensure_agent_runtime_dir + register_agent
    (eager on first spawn so socket ready before harness first turn)
  - apply_commit / merge_config_pr: ensure_agent_runtime_dir + agent_dir
  - Manager (auto_update): ensure_agent_runtime_dir + agent_dir
    (manager has no MCP listener; socket_server::start_manager owns it)

- ensure_runtime() retained in Coordinator with updated doc pointing at
  the preferred split form. No callers remain outside tests.

hive-c0re/src/actions.rs

@@ -159,7 +159,11 @@ pub async fn run_approval_apply_commit(
     approval_id: i64,
 ) -> Result<()> {
     let approval = fetch_approval_for_worker(coord, approval_id, ApprovalKind::ApplyCommit)?;
-    let agent_dir = coord.ensure_runtime(&approval.agent)?;
+    // Create the bind-mount source dir (first-spawn may not have it yet).
+    // MCP listener registration is deferred to mcp_sockets::spawn_poll
+    // which fires within 10 s of the container coming up.
+    lifecycle::ensure_agent_runtime_dir(&approval.agent)?;
+    let agent_dir = Coordinator::agent_dir(&approval.agent);
     let applied_dir = Coordinator::agent_applied_dir(&approval.agent);
     coord.set_queue_step(queue_entry_id, "apply commit");
     let (result, terminal_tag, is_first_spawn) =
@@ -192,7 +196,8 @@ pub async fn run_approval_merge_config_pr(
     approval_id: i64,
 ) -> Result<()> {
     let approval = fetch_approval_for_worker(coord, approval_id, ApprovalKind::MergeConfigPr)?;
-    let agent_dir = coord.ensure_runtime(&approval.agent)?;
+    lifecycle::ensure_agent_runtime_dir(&approval.agent)?;
+    let agent_dir = Coordinator::agent_dir(&approval.agent);
     let applied_dir = Coordinator::agent_applied_dir(&approval.agent);
     coord.set_queue_step(queue_entry_id, "merge config pr");
     let (result, terminal_tag) =

hive-c0re/src/coordinator.rs

@@ -545,14 +545,11 @@ impl Coordinator {
         }
     }
 
-    /// Assemble the per-agent filesystem paths for `name`. The caller
-    /// must supply `agent_dir` (from `ensure_runtime`) since that
-    /// creates the tmpfs entry on first call. All other paths are
-    /// derived statically from `name`.
-    ///
-    /// ```ignore
-    /// let paths = Coordinator::agent_paths(name, coord.ensure_runtime(name)?);
-    /// ```
+    /// Assemble the per-agent filesystem paths for `name`. `agent_dir`
+    /// is the runtime directory (`/run/hyperhive/agents/<name>`), obtained
+    /// from `Coordinator::agent_dir(name)` (pure path) or from
+    /// `lifecycle::ensure_agent_runtime_dir` + `agent_dir` when the dir
+    /// must be created. All other paths are derived statically from `name`.
     #[must_use]
     pub fn agent_paths(name: &str, agent_dir: PathBuf) -> AgentPaths {
         AgentPaths {
@@ -1453,11 +1450,17 @@ impl Coordinator {
         Self::agent_dir(name).join("mcp.sock")
     }
 
-    /// Ensure a runtime dir + (for sub-agents) per-agent socket exists. For
-    /// the manager, `socket_server::start_manager` owns the socket — just return
-    /// the dir. For sub-agents this is `register_agent` (creates a fresh
-    /// listener bound to `socket_path(name)`). Source directory of the
-    /// `/run/hive/mcp.sock` bind that ends up in `set_nspawn_flags`.
+    /// Ensure a runtime dir + (for sub-agents) per-agent socket exists.
+    ///
+    /// **Prefer the split form:**
+    /// - dir creation → `lifecycle::ensure_agent_runtime_dir(name)`
+    /// - listener registration → `register_agent(name)` (eagerly, on first
+    ///   spawn) or leave it to `mcp_sockets::spawn_poll` (reconcile within
+    ///   10 s, safe for the start + rebuild paths where the container
+    ///   takes longer than that to boot).
+    ///
+    /// This method is kept as a convenience shim for any remaining callers
+    /// that need the combined semantics in one call.
     pub fn ensure_runtime(self: &Arc<Self>, name: &str) -> Result<PathBuf> {
         if name == crate::lifecycle::MANAGER_NAME {
             let dir = Self::agent_dir(name);

hive-c0re/src/job_queue/exec.rs

@@ -97,9 +97,10 @@ async fn run_prebuild(
     relock: bool,
 ) -> Result<NodeOutput> {
     let name = &claim.agent;
-    let agent_dir = coord
-        .ensure_runtime(name)
-        .with_context(|| format!("ensure_runtime {name}"))?;
+    // Prebuild runs while the agent is still up — the runtime dir and
+    // MCP listener already exist. Use the pure path accessor; no need
+    // to re-register the listener (the mcp_sockets supervisor owns that).
+    let agent_dir = Coordinator::agent_dir(name);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(name, agent_dir);
     crate::lifecycle::prepare_rebuild_dirs(name, &paths).await?;
@@ -131,7 +132,9 @@ async fn run_prebuild(
 /// `Reconcile` runs after this node terminal ok *or* fail.
 async fn run_swap(coord: &Arc<Coordinator>, claim: &Claim, ctx: &Ctx<'_>) -> Result<NodeOutput> {
     let name = &claim.agent;
-    let agent_dir = coord.ensure_runtime(name)?;
+    // Swap runs on an already-existing (stopped) container — runtime dir
+    // and listener were created earlier. Pure path accessor suffices.
+    let agent_dir = Coordinator::agent_dir(name);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(name, agent_dir);
     let result =
@@ -178,7 +181,12 @@ async fn run_swap(coord: &Arc<Coordinator>, claim: &Claim, ctx: &Ctx<'_>) -> Res
 /// build+create — no prebuild needed).
 async fn run_create(coord: &Arc<Coordinator>, claim: &Claim, ctx: &Ctx<'_>) -> Result<NodeOutput> {
     let name = &claim.agent;
-    let agent_dir = coord.ensure_runtime(name)?;
+    // First-spawn: create the bind-mount source dir (tmpfs — empty after
+    // reboot). Register the MCP listener eagerly so it's ready when the
+    // tail Reconcile starts the container and the harness connects.
+    crate::lifecycle::ensure_agent_runtime_dir(name)?;
+    coord.register_agent(name)?;
+    let agent_dir = Coordinator::agent_dir(name);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(name, agent_dir);
     ctx.step("nixos-container create");
@@ -254,7 +262,13 @@ async fn run_reconcile(
             // Prebuild/Swap nodes; the bare-Reconcile templates (boot
             // reconcile, plain start/restart) otherwise start with
             // nothing under /run and fail.
-            let agent_dir = coord.ensure_runtime(name)?;
+            //
+            // Dir creation is the pure-filesystem part (no Coordinator
+            // dep). The MCP listener is reconciled by mcp_sockets::spawn_poll
+            // whose first tick fires immediately on daemon start — the
+            // container boot takes longer than the 10 s interval anyway.
+            crate::lifecycle::ensure_agent_runtime_dir(name)?;
+            let agent_dir = Coordinator::agent_dir(name);
             let hive = coord.hive_env();
             let paths = Coordinator::agent_paths(name, agent_dir);
             crate::lifecycle::write_dropins(name, &hive, &paths).await?;
@@ -336,7 +350,10 @@ async fn run_drain(coord: &Arc<Coordinator>, claim: &Claim, ctx: &Ctx<'_>) -> Re
 /// `set_nspawn_flags` + `set_resource_limits` + daemon-reload.
 async fn run_write_dropin(coord: &Arc<Coordinator>, claim: &Claim) -> Result<NodeOutput> {
     let name = &claim.agent;
-    let agent_dir = coord.ensure_runtime(name)?;
+    // write_dropins only needs the path value to build AgentPaths; the
+    // dir doesn't need to exist at this point (created by ensure_runtime
+    // on the upstream Prebuild/Start node).
+    let agent_dir = Coordinator::agent_dir(name);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(name, agent_dir);
     crate::lifecycle::write_dropins(name, &hive, &paths).await?;

hive-c0re/src/lib.rs

@@ -50,6 +50,6 @@ pub use stores::{
     approvals, audit_log, broker, build_logs, db, operator_questions, power, scheduled_prompts,
 };
 pub use workers::{
-    agent_sockets, auto_update, crash_watch, knowledge, reminder_scheduler,
+    agent_sockets, auto_update, crash_watch, knowledge, mcp_sockets, reminder_scheduler,
     scheduled_prompts_worker,
 };

hive-c0re/src/lifecycle/mod.rs

@@ -749,6 +749,23 @@ pub async fn sync_tmpfiles() {
     }
 }
 
+/// Ensure the per-agent runtime directory `/run/hyperhive/agents/<name>`
+/// exists. The directory is also written by `SyncAgentTmpfiles` (run at
+/// boot + spawn/destroy), but explicit creation in start/spawn paths guards
+/// against races where hive-c0re starts a container before tmpfiles.d has
+/// applied the new entry.
+///
+/// Pure filesystem op — no `Coordinator` dependency — so callers that only
+/// need the dir do not have to hold an `Arc<Coordinator>`.
+///
+/// # Errors
+/// Returns an error if `create_dir_all` fails.
+pub fn ensure_agent_runtime_dir(name: &str) -> Result<()> {
+    let dir = std::path::PathBuf::from(format!("/run/hyperhive/agents/{name}"));
+    std::fs::create_dir_all(&dir)
+        .with_context(|| format!("create agent runtime dir {}", dir.display()))
+}
+
 /// Build the per-line callback for `create_container_streaming` /
 /// `update_container_streaming`. Both ops share identical dispatch logic
 /// (stdout → info + `append_stdout`, stderr → warn + `append_stderr`); this

hive-c0re/src/main.rs

@@ -13,8 +13,8 @@ use hive_sh4re::{HostRequest, HostResponse};
 use hive_c0re::coordinator::{Coordinator, HiveEnv, ServeConfig};
 use hive_c0re::{
     agent_sockets, auto_update, broker, client, crash_watch, dashboard, dashboard_events, forge,
-    job_queue, knowledge, matrix, migrate, reminder_scheduler, scheduled_prompts_worker, server,
-    socket_server,
+    job_queue, knowledge, matrix, migrate, mcp_sockets, reminder_scheduler,
+    scheduled_prompts_worker, server, socket_server,
 };
 
 #[derive(Parser)]
@@ -420,6 +420,13 @@ async fn cmd_serve(
     // is one stat per agent per tick.
     // See `docs/gateway.md::Per-agent unix-socket upstream`.
     agent_sockets::spawn_poll();
+    // MCP socket listener reconcile loop: every 10s re-registers any
+    // running agent that lost its host-side MCP listener (e.g. after a
+    // hive-c0re restart cleared /run/hyperhive/agents/). First tick fires
+    // immediately so restarts re-register all running agents without delay.
+    // Decouples listener registration from the start path — start only needs
+    // lifecycle::ensure_agent_runtime_dir; the supervisor converges the rest.
+    mcp_sockets::spawn_poll(coord.clone());
     // Reminder scheduler: drains due reminders + handles
     // file_path payload persistence. See reminder_scheduler.rs.
     reminder_scheduler::spawn(coord.clone());

hive-c0re/src/server.rs

@@ -202,7 +202,12 @@ async fn dispatch(req: &HostRequest, coord: Arc<Coordinator>) -> HostResponse {
 /// registration and notifying the manager on failure.
 async fn handle_spawn(coord: &Arc<Coordinator>, name: &str) -> Result<HostResponse> {
     tracing::info!(%name, "spawn");
-    let agent_dir = coord.ensure_runtime(name)?;
+    // Create the bind-mount source dir (pure filesystem, no Coordinator dep).
+    // MCP listener registration happens eagerly here (not deferred to the
+    // supervisor) so the socket is ready before the harness's first turn.
+    lifecycle::ensure_agent_runtime_dir(name)?;
+    coord.register_agent(name)?;
+    let agent_dir = Coordinator::agent_dir(name);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(name, agent_dir);
     match lifecycle::spawn(name, &hive, &paths).await {

hive-c0re/src/workers/auto_update.rs

@@ -153,7 +153,10 @@ pub async fn ensure_root_agent(coord: &Arc<Coordinator>) -> Result<()> {
         return Ok(());
     }
     tracing::info!("manager container missing — spawning");
-    let runtime = coord.ensure_runtime(MANAGER_NAME)?;
+    lifecycle::ensure_agent_runtime_dir(MANAGER_NAME)?;
+    // Manager has no MCP listener (socket_server::start_manager owns its
+    // socket); just need the dir + path value.
+    let runtime = Coordinator::agent_dir(MANAGER_NAME);
     let hive = coord.hive_env();
     let paths = Coordinator::agent_paths(MANAGER_NAME, runtime);
     lifecycle::spawn(MANAGER_NAME, &hive, &paths).await?;

hive-c0re/src/workers/mcp_sockets.rs

@@ -0,0 +1,59 @@
+//! MCP socket listener reconcile loop.
+//!
+//! Periodically checks that every running agent container has a bound
+//! MCP listener registered in the `Coordinator`. Any agent that is running
+//! but whose listener has gone (e.g. after a hive-c0re restart that cleared
+//! `/run/hyperhive/agents/`) gets re-registered automatically.
+//!
+//! Same shape as `agent_sockets::spawn_poll` — a simple 10 s tick loop that
+//! converges "agent container running ⇒ MCP listener bound". The self-healing
+//! guarantee means the start path no longer needs to call `register_agent`
+//! directly (though spawn still does for eagerness); callers only need
+//! `lifecycle::ensure_agent_runtime_dir` to create the bind-mount source.
+
+use std::sync::Arc;
+
+use crate::coordinator::Coordinator;
+
+/// Spawn the MCP socket listener reconcile loop.
+///
+/// Every 10 s the loop lists running agent containers and calls
+/// `register_agent` for any that lack a bound listener. The first tick fires
+/// immediately so hive-c0re restarts re-register all running agents without
+/// waiting a full interval.
+pub fn spawn_poll(coord: Arc<Coordinator>) {
+    tokio::spawn(async move {
+        let mut interval = tokio::time::interval(std::time::Duration::from_secs(10));
+        loop {
+            interval.tick().await;
+            reconcile_once(&coord).await;
+        }
+    });
+}
+
+async fn reconcile_once(coord: &Arc<Coordinator>) {
+    let running = match crate::lifecycle::list().await {
+        Ok(names) => names,
+        Err(e) => {
+            tracing::debug!(error = ?e, "mcp_sockets poll: failed to list agents");
+            return;
+        }
+    };
+    let registered = coord.list_agents();
+    for container in running {
+        let Some(name) = container.strip_prefix(crate::lifecycle::AGENT_PREFIX) else {
+            continue;
+        };
+        if !registered.contains(&name.to_owned()) {
+            if let Err(e) = coord.register_agent(name) {
+                tracing::warn!(
+                    agent = %name,
+                    error = ?e,
+                    "mcp_sockets poll: register_agent failed"
+                );
+            } else {
+                tracing::debug!(agent = %name, "mcp_sockets poll: registered missing listener");
+            }
+        }
+    }
+}

hive-c0re/src/workers/mod.rs

@@ -1,12 +1,13 @@
 //! Background tasks and periodic sweeps: crash/login watcher, the
 //! reminder and scheduled-prompt delivery loops, boot-time auto-update
-//! reconcile, the agent-sockets.json writer loop, and knowledge-repo
-//! sync. Each submodule is re-exported at the crate root, so
-//! `crate::crash_watch::…` etc. keep working unchanged.
+//! reconcile, the agent-sockets.json writer loop, the MCP socket listener
+//! reconcile loop, and knowledge-repo sync. Each submodule is re-exported
+//! at the crate root, so `crate::crash_watch::…` etc. keep working unchanged.
 
 pub mod agent_sockets;
 pub mod auto_update;
 pub mod crash_watch;
 pub mod knowledge;
+pub mod mcp_sockets;
 pub mod reminder_scheduler;
 pub mod scheduled_prompts_worker;