From a3e61e6698e366c37290a12cccd79187cc0445c1 Mon Sep 17 00:00:00 2001
From: Jory Irving <46251616+joryirving@users.noreply.github.com>
Date: Wed, 16 Sep 2026 17:04:51 -0600
Subject: [PATCH] fix(bridge): count pipeline-shaped coders toward
 CODER_AGENT_SLOTS

_coder_agent_name only read spec.coderAgentRef, so pipeline-shaped Workloads
(spec.pipeline, e.g. a retry re-dispatched with an explicit pipeline) named
their coder on the issue-fix step's agentRef and resolved to None. Such
Workloads contributed nothing to _load_by_coder_agent, so their coder never
consumed a slot and a second coder could be dispatched alongside them --
observed as two 'coder' Jobs running at once under a coder:1 cap. Fall back to
the pipeline issue-fix step's agentRef so these Workloads count.
---
 bridge/main.py               | 29 ++++++++++++++++++++++++-----
 tests/test_bridge_runtime.py | 30 ++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+), 5 deletions(-)

diff --git a/bridge/main.py b/bridge/main.py
index 3423262..55b14d5 100644
--- a/bridge/main.py
+++ b/bridge/main.py
@@ -595,14 +595,33 @@ def _list_agentic_tasks_by_workload(
 
 
 def _coder_agent_name(workload: Dict[str, Any]) -> Optional[str]:
-    """Return a Workload's coder ref, or None when it cannot be resolved."""
+    """Return a Workload's coder ref, or None when it cannot be resolved.
+
+    Standard bridge Workloads name the coder at ``spec.coderAgentRef``. Pipeline-
+    shaped Workloads (``spec.pipeline`` — e.g. a retry re-dispatched with an
+    explicit pipeline) carry no top-level ``coderAgentRef`` and name the coder on
+    the ``issue-fix`` step's ``agentRef`` instead. Without the pipeline fallback
+    those Workloads resolve to None and contribute nothing to
+    ``_load_by_coder_agent``, so their coder never consumes a ``CODER_AGENT_SLOTS``
+    slot and a second coder can be dispatched alongside them (observed: a
+    pipeline-shaped coder and a standard coder both running the ``coder`` agent at
+    once under a ``coder:1`` cap).
+    """
     spec = workload.get("spec") or {}
     if not isinstance(spec, dict):
         return None
-    ref = spec.get("coderAgentRef") or {}
-    if not isinstance(ref, dict):
-        return None
-    return ref.get("name")
+    ref = spec.get("coderAgentRef")
+    if isinstance(ref, dict) and ref.get("name"):
+        return ref.get("name")
+    pipeline = spec.get("pipeline")
+    if isinstance(pipeline, list):
+        for step in pipeline:
+            if not isinstance(step, dict) or step.get("kind") != "issue-fix":
+                continue
+            step_ref = step.get("agentRef")
+            if isinstance(step_ref, dict) and step_ref.get("name"):
+                return step_ref.get("name")
+    return None
 
 
 def _active_workloads(
diff --git a/tests/test_bridge_runtime.py b/tests/test_bridge_runtime.py
index 298617b..ee812ea 100644
--- a/tests/test_bridge_runtime.py
+++ b/tests/test_bridge_runtime.py
@@ -115,6 +115,22 @@ def _task(name: str, workload: str, *, kind: str = "issue-fix", phase: str) -> d
     }
 
 
+def _pipeline_workload(name: str, phase: str, *, coder: str) -> dict:
+    # Pipeline-shaped Workload: the coder is named on the issue-fix step's
+    # agentRef, with no top-level coderAgentRef (the shape a retry re-dispatch
+    # produces). See _coder_agent_name's pipeline fallback.
+    return {
+        "metadata": {"name": name, "labels": {"created-by": "dispatch-bridge"}},
+        "spec": {
+            "pipeline": [
+                {"name": "code", "kind": "issue-fix", "agentRef": {"name": coder}},
+                {"name": "review", "kind": "review", "agentRef": {"name": "reviewer"}},
+            ]
+        },
+        "status": {"phase": phase},
+    }
+
+
 def _prfix_workload(name: str, phase: str) -> dict:
     return {
         "metadata": {"name": name, "labels": {"created-by": "dispatch-bridge-prfix"}},
@@ -303,6 +319,20 @@ def test_counts_running_issue_fix(self) -> None:
         )
         assert _load_by_coder_agent(api, "ns") == {"coder-py": 1}
 
+    def test_counts_pipeline_shaped_coder(self) -> None:
+        # A pipeline-shaped Workload names its coder on the issue-fix step, not at
+        # spec.coderAgentRef. It must still count toward coder load, or the coder
+        # slot leaks and a second coder runs alongside it under CODER_AGENT_SLOTS.
+        api = FakeAPI(
+            responses={
+                "list_namespaced_custom_object": [
+                    {"items": [_pipeline_workload("a", "Running", coder="coder-py")]},
+                    {"items": [_task("t1", "a", phase="Running")]},
+                ]
+            }
+        )
+        assert _load_by_coder_agent(api, "ns") == {"coder-py": 1}
+
     def test_terminal_issue_fix_with_running_review_frees_slot(self) -> None:
         # A terminal issue-fix task plus a running review does NOT hold the
         # coder busy: only issue-fix tasks count, and only while non-terminal.