athena-agent-launcher 0.3.3 → 0.3.4

package/lib/hosted-gateway.mjs

@@ -37,11 +37,12 @@ const TURN_TIMEOUT_MS = 120_000;             // hard cap on a single turn
 const SKEW_MS = 60_000;                       // M2: max clock skew for the HMAC timestamp
 const REPLAY_TTL_MS = 120_000;                // M2: nonce cache window (> skew)
 const DEFAULT_RATE_PER_MIN = 30;              // M6: token-bucket refill
-const DEFAULT_MAX_CONCURRENCY = 1;            // M6: in-flight cap (pilot = 1)
+const DEFAULT_MAX_CONCURRENCY = 1;            // M6: in-flight cap (pilot = 1) — load-bearing for arm custody, do NOT raise
+const BUSY_RETRY_AFTER_MS = 2000;             // DAE-16: hint the poller waits before retrying a busy container
 
-function sendJson(res, status, obj) {
+function sendJson(res, status, obj, headers) {
   const body = JSON.stringify(obj);
-  res.writeHead(status, { "content-type": "application/json; charset=utf-8" });
+  res.writeHead(status, { "content-type": "application/json; charset=utf-8", ...(headers || {}) });
   res.end(body);
 }
 
@@ -292,12 +293,32 @@ export async function startHostedGateway({ hermesBin, env, hermesHome, hostedGat
       // M6 — rate limit, then hard concurrency cap. Fail closed (429), never open.
       if (!takeToken()) {
         logEvent({ event: "rate_limited", request_id: requestId, status: 429 });
-        sendJson(res, 429, { error: "rate limited" });
+        sendJson(
+          res,
+          429,
+          { status: "rate_limited", reason: "rate_limited", message: "Too many requests — try again shortly.", retry_after_ms: BUSY_RETRY_AFTER_MS },
+          { "retry-after": String(Math.ceil(BUSY_RETRY_AFTER_MS / 1000)) },
+        );
         return;
       }
       if (inFlight >= maxConcurrency) {
-        logEvent({ event: "over_capacity", request_id: requestId, status: 429, in_flight: inFlight });
-        sendJson(res, 429, { error: "busy" });
+        // DAE-16 (D3 container half): a turn is already in flight. max_concurrency
+        // stays 1 (load-bearing for arm custody — never raised), so instead of
+        // blocking or a bare 429 we return an IMMEDIATE, TYPED busy signal the
+        // poller (IRIS-24) relays. Contract: HTTP 429 + Retry-After, body
+        // { status:"busy", reason:"turn_in_flight", message, retry_after_ms }.
+        logEvent({ event: "busy", request_id: requestId, status: 429, reason: "turn_in_flight", in_flight: inFlight });
+        sendJson(
+          res,
+          429,
+          {
+            status: "busy",
+            reason: "turn_in_flight",
+            message: "The agent is still processing a previous message — try again shortly.",
+            retry_after_ms: BUSY_RETRY_AFTER_MS,
+          },
+          { "retry-after": String(Math.ceil(BUSY_RETRY_AFTER_MS / 1000)) },
+        );
         return;
       }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "athena-agent-launcher",
-  "version": "0.3.3",
+  "version": "0.3.4",
   "description": "Run an Athena-configured agent locally. Resolves runtime + bindings from the Athena control plane and spawns the correct agent binary (Anthropic SDK, OpenClaw, or Hermes).",
   "type": "module",
   "bin": {