diff --git a/swarm-controller/src/config_pr.rs b/swarm-controller/src/config_pr.rs
new file mode 100644
index 00000000..a94c7da7
--- /dev/null
+++ b/swarm-controller/src/config_pr.rs
@@ -0,0 +1,234 @@
+//! Swarm-level config-PR status, kept current by both a webhook nudge and a
+//! periodic poll.
+//!
+//! `hive-c0re::forge::config_pr_poll` already scans `agent-configs/*` for
+//! open PRs — but it does that once per hive, to queue that hive's own
+//! `MergeConfigPr` approval, and nothing at swarm level reads the result.
+//! swarm-ui's config-PR panel needs a *swarm*-level answer to "does agent X
+//! have an open config PR" that does not depend on which hive currently
+//! hosts X being reachable.
+//!
+//! Both paths write the same [`ConfigPrCache`]:
+//!
+//! - [`spawn`] — a periodic full rescan, mirroring `hive-c0re`'s own poll
+//!   shape. The backstop: catches anything a missed delivery loses, and is
+//!   what populates the cache before the first delivery ever arrives.
+//! - [`ConfigPrCache::apply_webhook_delivery`] — called from
+//!   `crate::webhook::post_webhook_forge` on a verified `ConfigPr` delivery.
+//!   The low-latency path: a PR opening or closing shows up immediately
+//!   instead of waiting up to `POLL_INTERVAL`.
+//!
+//! Per mara's review call: ship both from the start rather than the poll
+//! alone — the eventual swarm-level replacement for `hive-c0re`'s own
+//! poll+webhook pair needs both anyway, so building only half here would be
+//! work redone rather than work reused.
+
+use std::collections::HashMap;
+use std::sync::{Arc, Mutex};
+
+use serde::Deserialize;
+
+use crate::forge::{Client, ConfigPrStatus};
+
+/// The handful of fields this cache needs out of a Forgejo `pull_request`
+/// webhook payload — not a full typed mirror of the event (Forgejo's own
+/// schema has dozens more), just enough to know which agent, which PR, and
+/// whether it's still open.
+#[derive(Deserialize)]
+pub struct ConfigPrWebhookPayload {
+    pull_request: WebhookPullRequest,
+    repository: WebhookRepository,
+}
+
+#[derive(Deserialize)]
+struct WebhookPullRequest {
+    number: u64,
+    /// Forgejo sends `"open"` or `"closed"` here — merged and
+    /// closed-without-merging are indistinguishable at this field, but this
+    /// cache only ever answers "is there an open PR," so the distinction
+    /// doesn't matter to it.
+    state: String,
+    html_url: Option<String>,
+}
+
+#[derive(Deserialize)]
+struct WebhookRepository {
+    /// The config repo's name IS the agent's name — same convention
+    /// `Client::list_open_config_prs` relies on.
+    name: String,
+}
+
+/// How often to rescan `agent-configs/*`. Matches the interval named in
+/// `hive-c0re::forge::config_pr_poll`'s own doc comment — same org, same
+/// staleness tolerance, no reason for the two to disagree.
+const POLL_INTERVAL: std::time::Duration = std::time::Duration::from_mins(5);
+
+/// The latest full scan, replaced atomically each cycle.
+///
+/// A full replace rather than an incremental merge: `Client::list_open_config_prs`
+/// already returns the complete current set (an agent with no open PR is
+/// simply absent), so merging would need its own stale-entry eviction to
+/// avoid an agent's long-closed PR lingering forever — the same reconcile
+/// problem `hive-c0re`'s poller solves for its approvals. A full replace
+/// sidesteps needing that logic twice: the map at any moment is nothing more
+/// than "the last successful scan's answer."
+pub struct ConfigPrCache(Mutex<HashMap<String, ConfigPrStatus>>);
+
+impl ConfigPrCache {
+    fn new() -> Self {
+        Self(Mutex::new(HashMap::new()))
+    }
+
+    /// `agent`'s open PR, if the last successful scan found one.
+    ///
+    /// Returns `None` both when the agent has no open PR and when no scan
+    /// has completed yet — the caller (`GET /api/agents/<name>/config-pr`)
+    /// treats both as "nothing to show," which is the honest answer for a
+    /// value that's a best-effort cache, not a live read.
+    pub fn get(&self, agent: &str) -> Option<ConfigPrStatus> {
+        self.0
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner)
+            .get(agent)
+            .cloned()
+    }
+
+    /// ⚠️ Can race [`Self::apply_webhook_delivery`]: a scan started before a
+    /// PR opened may finish *after* the webhook already upserted it, and
+    /// this snapshot — taken before that PR existed — will overwrite the
+    /// fresh entry. Self-heals within one `POLL_INTERVAL` (the next scan
+    /// sees the PR), so not worth coordinating against; noted per argus's
+    /// review rather than left implicit.
+    fn replace(&self, scan: HashMap<String, ConfigPrStatus>) {
+        *self
+            .0
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner) = scan;
+    }
+
+    /// Apply one verified `ConfigPr` webhook delivery's raw body to the
+    /// cache — the low-latency counterpart to [`spawn`]'s periodic rescan.
+    ///
+    /// A parse failure is logged and dropped, not propagated: the caller
+    /// (`crate::webhook::post_webhook_forge`) already returned 200 to
+    /// Forgejo (HMAC verification, not payload parsing, is what a retry
+    /// could fix), and the next poll tick reconciles whatever this delivery
+    /// would have changed — same "poll as backstop" property [`spawn`]'s
+    /// doc comment describes, just exercised on the failure path instead of
+    /// the steady-state one.
+    ///
+    /// An `open` PR is upserted unconditionally. A `closed` one is removed
+    /// only if the cached entry's PR number still matches — guards against
+    /// an out-of-order delivery (a stale `closed` for PR #1 arriving after a
+    /// newer `opened` for PR #2 on the same repo) wiping out a genuinely
+    /// current entry.
+    pub fn apply_webhook_delivery(&self, body: &[u8]) {
+        let payload: ConfigPrWebhookPayload = match serde_json::from_slice(body) {
+            Ok(p) => p,
+            Err(e) => {
+                tracing::warn!(
+                    error = %e,
+                    "config-pr webhook: payload did not parse, cache unchanged (next poll reconciles)"
+                );
+                return;
+            }
+        };
+        let mut map = self
+            .0
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        if payload.pull_request.state == "open" {
+            map.insert(
+                payload.repository.name,
+                ConfigPrStatus {
+                    pr_number: payload.pull_request.number,
+                    html_url: payload.pull_request.html_url,
+                },
+            );
+        } else if map
+            .get(&payload.repository.name)
+            .is_some_and(|cached| cached.pr_number == payload.pull_request.number)
+        {
+            map.remove(&payload.repository.name);
+        }
+    }
+}
+
+/// Build an empty cache and spawn the periodic scan that keeps it current.
+///
+/// The scan runs immediately on the first tick (`tokio::time::interval`'s
+/// default), so the cache is populated on startup rather than staying empty
+/// for a full `POLL_INTERVAL` after boot.
+pub fn spawn(client: Arc<Client>) -> Arc<ConfigPrCache> {
+    let cache = Arc::new(ConfigPrCache::new());
+    let cache_for_task = Arc::clone(&cache);
+    tokio::spawn(async move {
+        let mut ticker = tokio::time::interval(POLL_INTERVAL);
+        loop {
+            ticker.tick().await;
+            match client.list_open_config_prs().await {
+                Ok(scan) => cache_for_task.replace(scan),
+                Err(e) => {
+                    tracing::warn!(
+                        error = %format!("{e:#}"),
+                        "config-pr poll: scan failed, cache keeps its last value"
+                    );
+                }
+            }
+        }
+    });
+    cache
+}
+
+#[cfg(test)]
+mod tests {
+    use super::ConfigPrCache;
+
+    fn payload(agent: &str, number: u64, state: &str) -> Vec<u8> {
+        serde_json::json!({
+            "pull_request": { "number": number, "state": state, "html_url": "https://forge.example/pr" },
+            "repository": { "name": agent },
+        })
+        .to_string()
+        .into_bytes()
+    }
+
+    #[test]
+    fn an_open_delivery_upserts_the_entry() {
+        let cache = ConfigPrCache::new();
+        cache.apply_webhook_delivery(&payload("damocles", 5, "open"));
+        let status = cache.get("damocles").expect("entry inserted");
+        assert_eq!(status.pr_number, 5);
+    }
+
+    #[test]
+    fn a_closed_delivery_removes_a_matching_entry() {
+        let cache = ConfigPrCache::new();
+        cache.apply_webhook_delivery(&payload("damocles", 5, "open"));
+        cache.apply_webhook_delivery(&payload("damocles", 5, "closed"));
+        assert!(cache.get("damocles").is_none());
+    }
+
+    #[test]
+    fn a_stale_closed_delivery_does_not_clobber_a_newer_open_pr() {
+        let cache = ConfigPrCache::new();
+        // PR #5 opened, then closed, then a genuinely new PR #6 opens.
+        cache.apply_webhook_delivery(&payload("damocles", 5, "open"));
+        cache.apply_webhook_delivery(&payload("damocles", 6, "open"));
+        // The #5 close event arrives late, after #6 already replaced it.
+        cache.apply_webhook_delivery(&payload("damocles", 5, "closed"));
+        let status = cache.get("damocles").expect("PR #6 must survive");
+        assert_eq!(status.pr_number, 6);
+    }
+
+    #[test]
+    fn a_malformed_payload_leaves_the_cache_unchanged() {
+        let cache = ConfigPrCache::new();
+        cache.apply_webhook_delivery(&payload("damocles", 5, "open"));
+        cache.apply_webhook_delivery(b"not json");
+        let status = cache
+            .get("damocles")
+            .expect("prior entry must survive a bad delivery");
+        assert_eq!(status.pr_number, 5);
+    }
+}
diff --git a/swarm-controller/src/forge.rs b/swarm-controller/src/forge.rs
index d524ccae..d2f8c3e0 100644
--- a/swarm-controller/src/forge.rs
+++ b/swarm-controller/src/forge.rs
@@ -20,14 +20,26 @@ use forgejo_api::structs::{
     AddCollaboratorOption, AddCollaboratorOptionPermission, ChangeFileOperation,
     ChangeFileOperationOperation, ChangeFilesOptions, CreateBranchProtectionOption,
     CreateHookOption, CreateHookOptionConfig, CreateHookOptionType, CreateRepoOption,
-    RepoGetContentsQuery,
+    RepoGetContentsQuery, RepoListPullRequestsQuery, RepoListPullRequestsQueryState,
 };
 use forgejo_api::{ApiErrorKind, Auth, Forgejo, ForgejoError};
 use reqwest::StatusCode;
+use serde::Serialize;
 use std::collections::BTreeMap;
+use utoipa::ToSchema;
 
 use crate::webhook::DeliveryKind;
 
+/// An agent's open config-PR, as [`Client::list_open_config_prs`] reports it
+/// and `GET /api/agents/{name}/config-pr` serves it.
+#[derive(Clone, Debug, PartialEq, Serialize, ToSchema)]
+pub struct ConfigPrStatus {
+    pub pr_number: u64,
+    /// Absent only if Forgejo itself omitted the field — every real PR has
+    /// one; not worth failing the whole scan over.
+    pub html_url: Option<String>,
+}
+
 /// The `operators` team, whitelisted for the merge gate on every repo
 /// this client protects — provisioned by `hive-c0re::forge::repos`
 /// already (`ensure_operators_team`), not re-provisioned here. If that
@@ -351,6 +363,72 @@ impl Client {
         }
     }
 
+    /// Every agent in [`CONFIG_ORG`] with an open config PR, keyed by agent
+    /// name. Mirrors `hive-c0re::forge::config_pr_poll::poll_open_config_prs`'s
+    /// scan shape (list repos in the org, list open PRs per repo) but returns
+    /// data instead of side-effecting an approval queue — this daemon has no
+    /// approval system of its own; it exists so `GET
+    /// /api/agents/{name}/config-pr` has something to answer from,
+    /// independent of any one hive being up.
+    ///
+    /// A single repo's list failing does not fail the whole scan — logged and
+    /// skipped, so one flaky repo can't blank out every other agent's status.
+    pub async fn list_open_config_prs(
+        &self,
+    ) -> Result<std::collections::HashMap<String, ConfigPrStatus>> {
+        let repos = self
+            .api
+            .org_list_repos(CONFIG_ORG)
+            .all()
+            .await
+            .with_context(|| format!("list repos in {CONFIG_ORG}"))?;
+
+        let mut out = std::collections::HashMap::new();
+        for repo in repos {
+            let Some(agent) = repo.name.as_deref() else {
+                continue;
+            };
+            let query = RepoListPullRequestsQuery {
+                state: Some(RepoListPullRequestsQueryState::Open),
+                sort: None,
+                milestone: None,
+                labels: None,
+                poster: None,
+                base: None,
+                head: None,
+            };
+            let prs = match self
+                .api
+                .repo_list_pull_requests(CONFIG_ORG, agent, query)
+                .all()
+                .await
+            {
+                Ok(prs) => prs,
+                Err(e) => {
+                    tracing::debug!(%agent, error = %e, "swarm forge: listing config PRs failed, skipping repo");
+                    continue;
+                }
+            };
+            // Only the first open PR matters for the panel — a config repo
+            // is meant to carry at most one live proposal at a time (the
+            // same assumption `hive-c0re`'s poller and the `MergeConfigPr`
+            // approval flow both make).
+            if let Some(pr) = prs.into_iter().next() {
+                let Some(pr_number) = pr.number.and_then(|n| u64::try_from(n).ok()) else {
+                    continue;
+                };
+                out.insert(
+                    agent.to_owned(),
+                    ConfigPrStatus {
+                        pr_number,
+                        html_url: pr.html_url.map(|u| u.to_string()),
+                    },
+                );
+            }
+        }
+        Ok(out)
+    }
+
     /// Register the swarm-wide hooks against this controller, so a real
     /// forge event reaches [`crate::webhook`] instead of the endpoint only
     /// being reachable by hand.
diff --git a/swarm-controller/src/main.rs b/swarm-controller/src/main.rs
index 703faf65..a9a578aa 100644
--- a/swarm-controller/src/main.rs
+++ b/swarm-controller/src/main.rs
@@ -29,13 +29,18 @@ use std::path::PathBuf;
 use std::sync::{Arc, Mutex};
 
 use anyhow::{Context, Result};
-use axum::{Json, extract::State, routing::get};
+use axum::{
+    Json,
+    extract::{Path, State},
+    routing::get,
+};
 use hive_jobq_wire::GraphWire as _;
 use serde::{Deserialize, Serialize};
 use utoipa::{OpenApi, ToSchema};
 use utoipa_axum::{router::OpenApiRouter, routes};
 
 mod auth;
+mod config_pr;
 mod forge;
 mod status;
 mod webhook;
@@ -352,6 +357,11 @@ struct AppState {
     /// startup, and this way `as_deref()` yields the `&str` the verifier
     /// takes without a second hop through `String`.
     webhook_secret: Option<Arc<str>>,
+    /// Last successful `agent-configs/*` scan, kept current by
+    /// [`config_pr::spawn`]. `None` when this host has no forge configured
+    /// — same "absent means don't ask" shape as `status` and `forge` above,
+    /// not a startup failure.
+    config_prs: Option<Arc<config_pr::ConfigPrCache>>,
     /// The swarm's human display name (`services.hyperhive.swarm.name`),
     /// loaded once at startup (`load_swarm_name`). `None` when the
     /// operator never set it — a swarm without a display name is a
@@ -760,6 +770,40 @@ async fn get_jobq_rollup(State(state): State<AppState>) -> Json<Vec<hive_jobq_wi
     Json(hive_jobq_wire::state_rollup(graph, roots))
 }
 
+/// `agent`'s open config PR, from [`config_pr::ConfigPrCache`] — kept
+/// current by a webhook nudge (near-real-time) backstopped by a periodic
+/// scan (worst case: up to one `config_pr::POLL_INTERVAL` stale if a
+/// delivery was ever missed). Not a live forge read either way, so this
+/// answers even when the forge itself is momentarily unreachable.
+///
+/// `200` with a `null` body means "no open PR" (or "no scan has completed
+/// yet") — not distinguished, for the same reason
+/// [`config_pr::ConfigPrCache::get`] doesn't: both read as "nothing to show"
+/// to the swarm-ui config-PR panel this feeds, and a cache is a best-effort
+/// read, not a source of truth callers should expect to disambiguate
+/// against.
+#[utoipa::path(
+    get,
+    path = "/api/agents/{name}/config-pr",
+    params(("name" = String, Path, description = "agent name")),
+    responses(
+        (status = 200, description = "the agent's open config PR, or null if none / not scanned yet", body = Option<forge::ConfigPrStatus>),
+        (status = 503, description = "no forge is configured on this host", body = String),
+    ),
+    tag = "agents"
+)]
+async fn get_agent_config_pr(
+    State(state): State<AppState>,
+    Path(name): Path<String>,
+) -> Result<Json<Option<forge::ConfigPrStatus>>, StatusUnavailable> {
+    let Some(cache) = state.config_prs.as_ref() else {
+        return Err(StatusUnavailable(
+            "no forge is configured on this host".to_owned(),
+        ));
+    };
+    Ok(Json(cache.get(&name)))
+}
+
 /// The swarm's own public base URL, as the forge must address it.
 ///
 /// Set by `swarm-controller.nix` **only when this host actually serves the
@@ -938,6 +982,8 @@ async fn main() -> Result<()> {
         }
     };
 
+    let config_prs = forge_client.clone().map(config_pr::spawn);
+
     register_swarm_webhooks(forge_client, webhook_secret.clone());
 
     let state = AppState {
@@ -946,6 +992,7 @@ async fn main() -> Result<()> {
         status,
         jobq,
         webhook_secret,
+        config_prs,
         swarm_name: load_swarm_name().map(Arc::from),
     };
 
@@ -957,6 +1004,7 @@ async fn main() -> Result<()> {
         .routes(routes!(get_swarm_info))
         .routes(routes!(get_jobq_graph))
         .routes(routes!(get_jobq_rollup))
+        .routes(routes!(get_agent_config_pr))
         .routes(routes!(create_agent))
         .routes(routes!(webhook::post_webhook_forge))
         .split_for_parts();
@@ -1063,6 +1111,7 @@ mod tests {
             status: None,
             jobq: std::sync::Arc::clone(&sched),
             webhook_secret: None,
+            config_prs: None,
             swarm_name: None,
         };
         (state, sched)
diff --git a/swarm-controller/src/webhook.rs b/swarm-controller/src/webhook.rs
index ce4f9098..5412e6e9 100644
--- a/swarm-controller/src/webhook.rs
+++ b/swarm-controller/src/webhook.rs
@@ -311,10 +311,11 @@ pub(super) fn verify(
 
 /// POST `/webhook/forge/{kind}` — a swarm-wide Forgejo delivery.
 ///
-/// Verifies the HMAC over the raw body. Nothing consumes the delivery yet;
-/// once the swarm→hive channel lands, this daemon **parses** it and emits a
-/// semantic message (*knowledge repo changed*, *deploy agent X at rev Y*) to
-/// the hives that need it — see the module docs.
+/// Verifies the HMAC over the raw body, then acts per `kind`: a `Knowledge`
+/// delivery announces the change to every hive (see
+/// [`announce_knowledge_change`]); a `ConfigPr` delivery updates
+/// `crate::config_pr::ConfigPrCache` immediately (see that module's doc
+/// comment for why both this and a periodic poll write the same cache).
 ///
 /// Returns 200 on an accepted delivery so Forgejo does not retry. A refused
 /// one answers 401 (bad signature) or 503 (this daemon has no secret), and
@@ -367,14 +368,13 @@ pub(super) async fn post_webhook_forge(
         "webhook: verified delivery"
     );
 
-    // Only the knowledge delivery is acted on. Deploy coordination is a
-    // separate concern with its own issue — a hive does not want to hear that a
-    // config PR was opened, it wants to be told when to rebuild from main, and
-    // that is a decision the controller makes after a merge rather than a relay
-    // of this delivery. Written as a condition rather than a match arm holding
-    // an empty body, which would claim this is where that path is handled.
-    if kind == DeliveryKind::Knowledge {
-        announce_knowledge_change(&state).await;
+    match kind {
+        DeliveryKind::Knowledge => announce_knowledge_change(&state).await,
+        DeliveryKind::ConfigPr => {
+            if let Some(cache) = state.config_prs.as_ref() {
+                cache.apply_webhook_delivery(&body);
+            }
+        }
     }
 
     (StatusCode::OK, "ok").into_response()