From b60cfa6787b0b23bac91ce92454f853efae9be2b Mon Sep 17 00:00:00 2001
From: cyxer000 <48980829+cyxer000@users.noreply.github.com>
Date: Wed, 30 Sep 2026 22:58:08 +0800
Subject: [PATCH] agents: omp, Pi, Cline and dsh are no longer told a model
 writes more than its window holds (#338)

models.dev lists some models' output above their context (deepseek-chat:
384000 out, 128000 in), and magpie handed that on as maxTokens. The limit
is now kept within the contextWindow written beside it; a model whose
window isn't known keeps its output as before. Follow-up to the review of
---
 internal/agent/agents.go    |  2 +-
 internal/agent/cline.go     |  2 +-
 internal/agent/dsh.go       |  2 +-
 internal/agent/omp.go       |  2 +-
 internal/agent/route.go     | 11 +++++++++++
 internal/agent/sync_test.go | 24 ++++++++++++++++++++++++
 6 files changed, 39 insertions(+), 4 deletions(-)

diff --git a/internal/agent/agents.go b/internal/agent/agents.go
index 4d4212e1..8c0e1d86 100644
--- a/internal/agent/agents.go
+++ b/internal/agent/agents.go
@@ -292,7 +292,7 @@ func magpieProviderJSONAt(shape, catalog, gw string) any {
 			// without it Pi caps every reply at 16384 tokens, a model
 			// that can write 128K included
 			if m.Output > 0 {
-				e["maxTokens"] = m.Output
+				e["maxTokens"] = maxTokens(m)
 			}
 			ms = append(ms, e)
 		}
diff --git a/internal/agent/cline.go b/internal/agent/cline.go
index 34f1198f..96dd3115 100644
--- a/internal/agent/cline.go
+++ b/internal/agent/cline.go
@@ -375,7 +375,7 @@ func clineModels(model string) map[string]any {
 			e["contextWindow"] = m.Context
 		}
 		if m.Output > 0 {
-			e["maxTokens"] = m.Output
+			e["maxTokens"] = maxTokens(m)
 		}
 		ms[m.ID] = e
 	}
diff --git a/internal/agent/dsh.go b/internal/agent/dsh.go
index c2cf81e3..1fa9f7d2 100644
--- a/internal/agent/dsh.go
+++ b/internal/agent/dsh.go
@@ -405,7 +405,7 @@ func dshProviderLines(modern bool, effort string) []string {
 			lines = append(lines, fmt.Sprintf("        contextWindow: %d", m.Context))
 		}
 		if m.Output > 0 {
-			lines = append(lines, fmt.Sprintf("        maxTokens: %d", m.Output))
+			lines = append(lines, fmt.Sprintf("        maxTokens: %d", maxTokens(m)))
 		}
 		if m.Images {
 			lines = append(lines, "        inputModalities: [text, image]")
diff --git a/internal/agent/omp.go b/internal/agent/omp.go
index f1ba8a1f..a18032bc 100644
--- a/internal/agent/omp.go
+++ b/internal/agent/omp.go
@@ -184,7 +184,7 @@ type ompProviderEntry struct {
 func ompProvider() ompProviderEntry {
 	ms := []ompModel{}
 	for _, m := range magpieModels("omp") {
-		e := ompModel{ID: m.ID, Name: m.Name, Context: m.Context, MaxTokens: m.Output}
+		e := ompModel{ID: m.ID, Name: m.Name, Context: m.Context, MaxTokens: maxTokens(m)}
 		mode := "effort"
 		switch {
 		case slices.Contains(m.APIs, string(provider.Responses)):
diff --git a/internal/agent/route.go b/internal/agent/route.go
index b205e8d4..2f947328 100644
--- a/internal/agent/route.go
+++ b/internal/agent/route.go
@@ -86,6 +86,17 @@ func magpieModels(agent string) []catalog.Model {
 	return out
 }
 
+// maxTokens is the output limit an agent is handed for m, kept within the
+// context window it is handed with it: models.dev lists some models' output
+// above their window (deepseek-chat's 384000 against 128000). An unknown
+// window leaves the output as it is.
+func maxTokens(m catalog.Model) int {
+	if m.Context > 0 && m.Output > m.Context {
+		return m.Context
+	}
+	return m.Output
+}
+
 // group tags every option with a group name.
 func group(name string, opts []Option) []Option {
 	for i := range opts {
diff --git a/internal/agent/sync_test.go b/internal/agent/sync_test.go
index 1a50ef55..a90c1651 100644
--- a/internal/agent/sync_test.go
+++ b/internal/agent/sync_test.go
@@ -13,6 +13,7 @@ import (
 	"github.com/yetone/magpie/internal/edit"
 	"github.com/yetone/magpie/internal/provider"
 	"github.com/yetone/magpie/internal/settings"
+	"gopkg.in/yaml.v3"
 )
 
 // syncHome is a sandbox home with a models.dev catalog that knows glm-4.6's
@@ -103,6 +104,29 @@ func TestPiModelsCarryMaxTokens(t *testing.T) {
 	}
 }
 
+// An output limit above the model's window (models.dev lists deepseek-chat's
+// 384000 against 128000 of context) is cut to the window for every agent
+// magpie hands maxTokens; one whose window isn't known keeps its output.
+func TestMaxTokensWithinContextWindow(t *testing.T) {
+	syncHome(t)
+	check := func(limit, want string) {
+		t.Helper()
+		os.WriteFile(catalog.CachePath(), []byte(`{"zai":{"models":{"glm-4.6":{"id":"glm-4.6","name":"GLM-4.6","limit":{`+limit+`}}}}}`), 0o644)
+		catalog.Reset()
+		pi, _ := json.Marshal(magpieProviderJSON("pi"))
+		cline, _ := json.Marshal(clineModels(""))
+		omp, _ := yaml.Marshal(ompProvider())
+		dsh := strings.Join(dshProviderLines(true, ""), "\n")
+		if !strings.Contains(string(pi), `"maxTokens":`+want) || !strings.Contains(string(cline), `"maxTokens":`+want) ||
+			!strings.Contains(string(omp), "maxTokens: "+want) || !strings.Contains(dsh, "maxTokens: "+want) {
+			t.Fatalf("want maxTokens %s:\npi %s\ncline %s\nomp %s\ndsh %s", want, pi, cline, omp, dsh)
+		}
+	}
+	check(`"context":128000,"output":384000`, "128000")
+	check(`"output":384000`, "384000")
+	check(`"context":204800,"output":131072`, "131072")
+}
+
 // A provider added after a magpie model was picked reaches the lists agents
 // keep of magpie's models; a file magpie wrote nothing into stays as it is.
 func TestSyncCatalogRewritesAgentLists(t *testing.T) {