diff --git a/skills/cuopt-numerical-optimization-api-c/SKILL.md b/skills/cuopt-numerical-optimization-api-c/SKILL.md
index b25acc14ba..808c110ce8 100644
--- a/skills/cuopt-numerical-optimization-api-c/SKILL.md
+++ b/skills/cuopt-numerical-optimization-api-c/SKILL.md
@@ -35,6 +35,14 @@ QP uses the same library, include/lib paths, and build pattern as LP/MILP — on
 - **Continuous variables only** — set `CUOPT_CONTINUOUS` for every variable; integer QP is not supported.
 - **Q should be PSD** for a convex problem.
 
+## Dual values (LP / QP)
+
+`cuOptGetDualSolution` (shadow prices) and `cuOptGetReducedCosts` (reduced costs) return values
+for **LP and QP with linear constraints** — the barrier solver is primal-dual, so a quadratic
+objective still yields duals. cuOpt returns **no duals for a problem with quadratic constraints**
+(the returned arrays are filled with `NaN`). See [assets/lp_duals](assets/lp_duals/) for the call
+sequence — it is an LP, but the same calls apply to a QP whose constraints are all linear.
+
 ## Debugging (MPS / C)
 
 **MPS parsing:** Required sections in order: NAME, ROWS, COLUMNS, RHS, (optional) BOUNDS, ENDATA. Integer markers: `'MARKER'`, `'INTORG'`, `'INTEND'`.
diff --git a/skills/cuopt-numerical-optimization-api-c/assets/README.md b/skills/cuopt-numerical-optimization-api-c/assets/README.md
index e354988da1..d9bb715d00 100644
--- a/skills/cuopt-numerical-optimization-api-c/assets/README.md
+++ b/skills/cuopt-numerical-optimization-api-c/assets/README.md
@@ -11,6 +11,11 @@ LP/MILP C API reference implementations. Use as reference when building new appl
 | [milp_production_planning](milp_production_planning/) | MILP | Production planning with resource constraints |
 | [mps_solver](mps_solver/) | LP/MILP | Solve from MPS file via `cuOptReadProblem` |
 
+> **Duals:** `lp_duals` is an LP, but `cuOptGetDualSolution` / `cuOptGetReducedCosts` also return
+> shadow prices and reduced costs for a **QP with linear constraints** (the barrier solver is
+> primal-dual). No duals are returned for problems with **quadratic constraints** — the arrays
+> come back filled with `NaN`.
+
 ## Build and run
 
 Set include and library paths, then build and run.
diff --git a/skills/cuopt-numerical-optimization-api-python/SKILL.md b/skills/cuopt-numerical-optimization-api-python/SKILL.md
index a7460c7c7c..07d7e7f632 100644
--- a/skills/cuopt-numerical-optimization-api-python/SKILL.md
+++ b/skills/cuopt-numerical-optimization-api-python/SKILL.md
@@ -253,12 +253,18 @@ settings.set_parameter("log_to_console", 1)
 | QP rejected with MAXIMIZE | QP only supports MINIMIZE | Negate the objective: minimize `-f(x)` |
 | QP returns non-optimal | Q not PSD or variables badly scaled | Check Q is PSD; rescale variables to similar magnitudes |
 
-## Getting Dual Values (LP only)
+## Getting Dual Values (LP / QP)
+
+Shadow prices (`DualValue`) and reduced costs (`ReducedCost`) are returned for **LP and QP** —
+cuOpt's barrier solver is primal-dual, so a QP with a quadratic objective and **linear**
+constraints returns duals just like an LP. The constraints you read duals from must be linear:
+cuOpt returns **no dual variables for a problem that has any quadratic constraint** (every
+`DualValue`/`ReducedCost` comes back as `NaN`). MILP returns no duals.
 
 ```python
 if problem.Status.name == "Optimal":
-    constraint = problem.getConstraint("resource_a")
-    shadow_price = constraint.DualValue
+    constraint = problem.getConstraint("resource_a")   # linear constraint
+    shadow_price = constraint.DualValue                # NaN if the model has quadratic constraints
     print(f"Shadow price: {shadow_price}")
 ```
 
diff --git a/skills/cuopt-numerical-optimization-api-python/references/qp_examples.md b/skills/cuopt-numerical-optimization-api-python/references/qp_examples.md
index 80b9802dbb..10ceba8ac9 100644
--- a/skills/cuopt-numerical-optimization-api-python/references/qp_examples.md
+++ b/skills/cuopt-numerical-optimization-api-python/references/qp_examples.md
@@ -123,6 +123,36 @@ if problem.Status.name == "Optimal":
     print(f"Objective = {problem.ObjValue:.4f}")
 ```
 
+## Reading Duals from a QP
+
+A QP with **linear** constraints returns shadow prices and reduced costs just like an LP
+(cuOpt's barrier solver is primal-dual). A quadratic *objective* is fine; a quadratic
+*constraint* is not — cuOpt returns **no duals for a problem with quadratic constraints**
+(every `DualValue`/`ReducedCost` is `NaN`), so read duals only when all constraints are linear.
+
+```python
+"""
+minimize    x² + y² + z²
+subject to  x + y + z == 10   (linear)
+The equality's shadow price is the marginal change in the optimal objective
+per unit increase of the right-hand side.
+"""
+from cuopt.linear_programming.problem import Problem, MINIMIZE
+
+problem = Problem("QPDuals")
+x = problem.addVariable(lb=0, name="x")
+y = problem.addVariable(lb=0, name="y")
+z = problem.addVariable(lb=0, name="z")
+problem.setObjective(x*x + y*y + z*z, sense=MINIMIZE)
+problem.addConstraint(x + y + z == 10, name="budget")
+problem.solve()
+
+if problem.Status.name in ["Optimal", "PrimalFeasible"]:
+    for c in problem.getConstraints():
+        # 'budget' dual magnitude ≈ 6.667 (Δobjective per unit of the RHS)
+        print(f"{c.ConstraintName} DualValue = {c.DualValue}")
+```
+
 ## Maximization Workaround
 
 ```python