From ae5965241eec67188a2bba8cc12539709e796ef0 Mon Sep 17 00:00:00 2001
From: enieuwy <121954036+enieuwy@users.noreply.github.com>
Date: Sat, 8 Aug 2026 13:14:35 +0800
Subject: [PATCH] fix(session): count every real output block as a served turn
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Codex found the shared attribution predicate recognising only tool calls,
text and signed thinking. A native image response often arrives with no
text and no tool call at all, so an image-only turn was read as producing
nothing: attribution stayed on whichever model spoke before it, and the
empty-stop rule could classify a successful generation as empty.

Everything the assistant can emit now counts except two: unsigned thinking,
which is not provider-authenticated and was already excluded, and
Anthropic's `fallback` marker, which records that a request was routed
elsewhere rather than carrying output. Redacted thinking and server-tool
blocks are real work by the same argument as the image.

The `toolUse` arm keeps its stricter rule — an orphaned toolUse stop needs
a tool_use block to anchor a later tool_result, and an image cannot.

Also switches the new test to the namespace import AGENTS.md requires for
node builtins.
---
 packages/coding-agent/src/session/messages.ts | 30 ++++++++++++++-----
 .../persisted-agent-attribution.test.ts       | 20 ++++++++++++-
 2 files changed, 41 insertions(+), 9 deletions(-)

diff --git a/packages/coding-agent/src/session/messages.ts b/packages/coding-agent/src/session/messages.ts
index ce1297b26..ae5eba84b 100644
--- a/packages/coding-agent/src/session/messages.ts
+++ b/packages/coding-agent/src/session/messages.ts
@@ -504,15 +504,29 @@ function hasText(content: { text?: unknown }): boolean {
 	return typeof content.text === "string" && content.text.trim().length > 0;
 }
 
-/** A block a consumer can act on: a call, real text, or provider-signed thinking. */
+/**
+ * A block that is real output from the model.
+ *
+ * Everything the assistant can emit counts except two: unsigned thinking, which
+ * is not provider-authenticated and not actionable, and Anthropic's `fallback`
+ * marker, which records that the request was routed elsewhere rather than
+ * carrying any output. A native image response often arrives with no text and
+ * no tool call at all, so recognising only those would call it nothing.
+ */
 function isActionableContent(content: AssistantMessage["content"][number] | undefined): boolean {
-	if (!content) return false;
-	if (content.type === "toolCall") return true;
-	if (content.type === "text") return hasText(content);
-	// Unsigned thinking is not actionable; a signature is provider-authenticated.
-	return content.type === "thinking" && typeof content.thinkingSignature === "string"
-		? content.thinkingSignature.trim().length > 0
-		: false;
+	switch (content?.type) {
+		case "toolCall":
+		case "image":
+		case "redactedThinking":
+		case "anthropicServerTool":
+			return true;
+		case "text":
+			return hasText(content);
+		case "thinking":
+			return typeof content.thinkingSignature === "string" && content.thinkingSignature.trim().length > 0;
+		default:
+			return false;
+	}
 }
 
 /** A `stop`/`toolUse` turn that produced nothing actionable. Any other stop
diff --git a/packages/coding-agent/test/registry/persisted-agent-attribution.test.ts b/packages/coding-agent/test/registry/persisted-agent-attribution.test.ts
index da839c104..5a41ae004 100644
--- a/packages/coding-agent/test/registry/persisted-agent-attribution.test.ts
+++ b/packages/coding-agent/test/registry/persisted-agent-attribution.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from "bun:test";
-import path from "node:path";
+import * as path from "node:path";
 import { AgentRegistry } from "@oh-my-pi/pi-coding-agent/registry/agent-registry";
 import { registerPersistedSubagents } from "@oh-my-pi/pi-coding-agent/registry/persisted-agents";
 import { TempDir } from "@oh-my-pi/pi-utils";
@@ -126,6 +126,24 @@ describe("persisted agent model attribution", () => {
 		expect(history?.resolvedModelIsFallback).toBe(false);
 	});
 
+	it("credits a turn whose only output is an image", async () => {
+		using tempDir = TempDir.createSync("@omp-attribution-image-");
+		// A native image response can arrive with no text and no tool call at all.
+		// Recognising only those would call it nothing and leave the run credited
+		// to whichever model spoke before it.
+		const registry = await historyFor(tempDir.path(), "Painter", [
+			...transcriptHead(),
+			assistant("a1", "si", SONNET, "stop", [{ type: "text", text: "sonnet did the work" }]),
+			assistant("e1", "a1", SONNET, "error", []),
+			modelChange("m2", "e1", "openai-codex/gpt-5.6-sol", "fallback", true),
+			assistant("a2", "m2", SOL, "stop", [{ type: "image", data: "aGk=", mimeType: "image/png" }]),
+		]);
+
+		const history = registry.get("Painter")?.history;
+		expect(history?.resolvedModel).toBe("openai-codex/gpt-5.6-sol");
+		expect(history?.resolvedModelIsFallback).toBe(true);
+	});
+
 	it("treats a budget-exhausted length stop with nothing usable as unserved", async () => {
 		using tempDir = TempDir.createSync("@omp-attribution-length-");
 		// `length` is not an "empty stop", so the empty-stop rule never inspects it: