gsd-pi 2.82.0 → 3.0.0-dev.04f5ccf82

package/packages/pi-coding-agent/dist/tests/system-prompt-file-safety.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=system-prompt-file-safety.test.d.ts.map

package/packages/pi-coding-agent/dist/tests/system-prompt-file-safety.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"system-prompt-file-safety.test.d.ts","sourceRoot":"","sources":["../../src/tests/system-prompt-file-safety.test.ts"],"names":[],"mappings":""}

package/packages/pi-coding-agent/dist/tests/system-prompt-file-safety.test.js

@@ -0,0 +1,17 @@
+// GSD-2 pi-coding-agent system prompt file-safety regression tests.
+import test from "node:test";
+import assert from "node:assert/strict";
+import { buildSystemPrompt } from "../core/system-prompt.js";
+test("buildSystemPrompt: read and write tools require reading before overwrite", () => {
+    const prompt = buildSystemPrompt({ selectedTools: ["read", "write"] });
+    assert.match(prompt, /before editing or overwriting/i);
+    assert.match(prompt, /Before write creates or replaces a file, verify the target path/i);
+    assert.match(prompt, /if it exists, read it first/i);
+    assert.match(prompt, /Use write only for new files or complete rewrites after verifying the target path/i);
+});
+test("buildSystemPrompt: write-only tool guidance does not reference unavailable read tool", () => {
+    const prompt = buildSystemPrompt({ selectedTools: ["write"] });
+    assert.doesNotMatch(prompt, /read it first/i);
+    assert.match(prompt, /Use write only for new files or complete rewrites after verifying the target path/i);
+});
+//# sourceMappingURL=system-prompt-file-safety.test.js.map

package/packages/pi-coding-agent/dist/tests/system-prompt-file-safety.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"system-prompt-file-safety.test.js","sourceRoot":"","sources":["../../src/tests/system-prompt-file-safety.test.ts"],"names":[],"mappings":"AAAA,oEAAoE;AAEpE,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,MAAM,MAAM,oBAAoB,CAAC;AAExC,OAAO,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAE7D,IAAI,CAAC,0EAA0E,EAAE,GAAG,EAAE;IACrF,MAAM,MAAM,GAAG,iBAAiB,CAAC,EAAE,aAAa,EAAE,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC;IAEvE,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,gCAAgC,CAAC,CAAC;IACvD,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,kEAAkE,CAAC,CAAC;IACzF,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,8BAA8B,CAAC,CAAC;IACrD,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,oFAAoF,CAAC,CAAC;AAC5G,CAAC,CAAC,CAAC;AAEH,IAAI,CAAC,sFAAsF,EAAE,GAAG,EAAE;IACjG,MAAM,MAAM,GAAG,iBAAiB,CAAC,EAAE,aAAa,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;IAE/D,MAAM,CAAC,YAAY,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAC9C,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,oFAAoF,CAAC,CAAC;AAC5G,CAAC,CAAC,CAAC","sourcesContent":["// GSD-2 pi-coding-agent system prompt file-safety regression tests.\n\nimport test from \"node:test\";\nimport assert from \"node:assert/strict\";\n\nimport { buildSystemPrompt } from \"../core/system-prompt.js\";\n\ntest(\"buildSystemPrompt: read and write tools require reading before overwrite\", () => {\n\tconst prompt = buildSystemPrompt({ selectedTools: [\"read\", \"write\"] });\n\n\tassert.match(prompt, /before editing or overwriting/i);\n\tassert.match(prompt, /Before write creates or replaces a file, verify the target path/i);\n\tassert.match(prompt, /if it exists, read it first/i);\n\tassert.match(prompt, /Use write only for new files or complete rewrites after verifying the target path/i);\n});\n\ntest(\"buildSystemPrompt: write-only tool guidance does not reference unavailable read tool\", () => {\n\tconst prompt = buildSystemPrompt({ selectedTools: [\"write\"] });\n\n\tassert.doesNotMatch(prompt, /read it first/i);\n\tassert.match(prompt, /Use write only for new files or complete rewrites after verifying the target path/i);\n});\n"]}

package/packages/pi-coding-agent/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gsd/pi-coding-agent",
-  "version": "2.82.0",
+  "version": "3.0.0",
   "description": "Coding agent CLI (vendored from pi-mono)",
   "type": "module",
   "gsd": {

package/packages/pi-coding-agent/src/core/agent-session-thinking-level.test.ts

@@ -0,0 +1,79 @@
+// Regression test for #5102: disabling thinking on one reasoning-capable
+// model must not silently persist "off" as the global default.
+
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, it } from "node:test";
+
+import { Agent } from "@gsd/pi-agent-core";
+import { getModel } from "@gsd/pi-ai";
+import { AgentSession } from "./agent-session.js";
+import { AuthStorage } from "./auth-storage.js";
+import { ModelRegistry } from "./model-registry.js";
+import { DefaultResourceLoader } from "./resource-loader.js";
+import { SessionManager } from "./session-manager.js";
+import { SettingsManager } from "./settings-manager.js";
+
+let testDir: string;
+
+async function createSession(): Promise<{ session: AgentSession; settingsManager: SettingsManager }> {
+	const agentDir = join(testDir, "agent-home");
+	const authStorage = AuthStorage.inMemory({});
+	const modelRegistry = new ModelRegistry(authStorage, join(agentDir, "models.json"));
+	const settingsManager = SettingsManager.inMemory({ defaultThinkingLevel: "high" });
+	const resourceLoader = new DefaultResourceLoader({
+		cwd: testDir,
+		agentDir,
+		settingsManager,
+		noExtensions: true,
+		noPromptTemplates: true,
+		noThemes: true,
+	});
+	await resourceLoader.reload();
+
+	const session = new AgentSession({
+		agent: new Agent({
+			initialState: {
+				model: getModel("zai", "glm-5.1" as any),
+				thinkingLevel: "high",
+			},
+		}),
+		sessionManager: SessionManager.inMemory(testDir),
+		settingsManager,
+		cwd: testDir,
+		resourceLoader,
+		modelRegistry,
+	});
+
+	return { session, settingsManager };
+}
+
+describe("AgentSession thinking level persistence", () => {
+	beforeEach(() => {
+		testDir = mkdtempSync(join(tmpdir(), "agent-session-thinking-level-"));
+	});
+
+	afterEach(() => {
+		rmSync(testDir, { recursive: true, force: true });
+	});
+
+	it("does not persist off as the global default for reasoning-capable models", async () => {
+		const { session, settingsManager } = await createSession();
+
+		session.setThinkingLevel("off");
+
+		assert.equal(session.thinkingLevel, "off");
+		assert.equal(settingsManager.getDefaultThinkingLevel(), "high");
+	});
+
+	it("still persists non-off thinking levels as the global default", async () => {
+		const { session, settingsManager } = await createSession();
+
+		session.setThinkingLevel("low");
+
+		assert.equal(session.thinkingLevel, "low");
+		assert.equal(settingsManager.getDefaultThinkingLevel(), "low");
+	});
+});

package/packages/pi-coding-agent/src/core/agent-session.ts

@@ -1945,7 +1945,7 @@ export class AgentSession {
 
 		if (isChanging) {
 			this.sessionManager.appendThinkingLevelChange(effectiveLevel);
-			if (this.supportsThinking() || effectiveLevel !== "off") {
+			if (effectiveLevel !== "off") {
 				this.settingsManager.setDefaultThinkingLevel(effectiveLevel);
 			}
 			this._emitSessionStateChanged("set_thinking_level");

package/packages/pi-coding-agent/src/core/chat-controller-ordering.test.ts

@@ -1,3 +1,5 @@
+// Project/App: GSD-2
+// File Purpose: Regression tests for streamed interactive chat ordering.
 import assert from "node:assert/strict";
 import { test } from "node:test";
 
@@ -97,9 +99,7 @@ function createHost() {
 	return host;
 }
 
-test("chat-controller renders content blocks in content[] index order (tool-first stream)", async () => {
-	// ToolExecutionComponent uses the global theme singleton.
-	// Install a minimal no-op theme implementation for this unit test.
+function installTheme() {
 	(globalThis as any)[Symbol.for("@gsd/pi-coding-agent:theme")] = {
 		fg: (_key: string, text: string) => text,
 		bg: (_key: string, text: string) => text,
@@ -107,6 +107,12 @@ test("chat-controller renders content blocks in content[] index order (tool-firs
 		italic: (text: string) => text,
 		truncate: (text: string) => text,
 	};
+}
+
+test("chat-controller renders content blocks in content[] index order (tool-first stream)", async () => {
+	// ToolExecutionComponent uses the global theme singleton.
+	// Install a minimal no-op theme implementation for this unit test.
+	installTheme();
 
 	const host = createHost();
 	const toolId = "mcp-tool-1";
@@ -168,6 +174,50 @@ test("chat-controller renders content blocks in content[] index order (tool-firs
 	assert.equal(host.chatContainer.children[1]?.constructor?.name, "AssistantMessageComponent");
 });
 
+test("chat-controller skips empty GPT reasoning blocks before tool-only turns", async () => {
+	installTheme();
+
+	const host = createHost();
+	host.getMarkdownThemeWithSettings = () => ({});
+	const toolId = "gpt-tool-1";
+	const toolCall = {
+		type: "toolCall",
+		id: toolId,
+		name: "read",
+		arguments: { filePath: "todo.js" },
+	};
+	const content = [
+		{ type: "thinking", thinking: "", thinkingSignature: "encrypted" },
+		toolCall,
+	];
+
+	await handleAgentEvent(host, { type: "message_start", message: makeAssistant([]) } as any);
+
+	await handleAgentEvent(
+		host,
+		{
+			type: "message_update",
+			message: makeAssistant(content),
+			assistantMessageEvent: {
+				type: "toolcall_end",
+				contentIndex: 1,
+				toolCall: {
+					...toolCall,
+					externalResult: {
+						content: [{ type: "text", text: "todo contents" }],
+						details: {},
+						isError: false,
+					},
+				},
+				partial: makeAssistant(content),
+			},
+		} as any,
+	);
+
+	assert.equal(host.chatContainer.children.length, 1, "empty reasoning should not create a blank assistant block");
+	assert.equal(host.chatContainer.children[0]?.constructor?.name, "ToolExecutionComponent");
+});
+
 test("chat-controller renders serverToolUse before trailing text matching content[] index order", async () => {
 	(globalThis as any)[Symbol.for("@gsd/pi-coding-agent:theme")] = {
 		fg: (_key: string, text: string) => text,

package/packages/pi-coding-agent/src/core/compaction/compaction.test.ts

@@ -9,7 +9,14 @@ import { describe, it, mock } from "node:test";
 import type { AgentMessage } from "@gsd/pi-agent-core";
 import type { Model, AssistantMessage } from "@gsd/pi-ai";
 
-import { generateSummary, estimateTokens, chunkMessages, isDegenerateSummary, CompactionProducedNoSummaryError } from "./compaction.js";
+import {
+	generateSummary,
+	estimateTokens,
+	chunkMessages,
+	isDegenerateSummary,
+	CompactionProducedNoSummaryError,
+	calculateContextTokens,
+} from "./compaction.js";
 import { estimateSerializedTokens } from "./utils.js";
 
 // ---------------------------------------------------------------------------
@@ -187,6 +194,21 @@ describe("chunkMessages", () => {
 	});
 });
 
+describe("calculateContextTokens", () => {
+	it("uses prompt-relevant usage only (input + cacheRead + cacheWrite)", () => {
+		const usage = {
+			input: 65,
+			output: 39_846,
+			cacheRead: 2_945_563,
+			cacheWrite: 243_452,
+			totalTokens: 3_228_926,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+		};
+
+		assert.equal(calculateContextTokens(usage), 3_189_080);
+	});
+});
+
 // ---------------------------------------------------------------------------
 // generateSummary chunked fallback tests
 // ---------------------------------------------------------------------------

package/packages/pi-coding-agent/src/core/compaction/compaction.ts

@@ -107,10 +107,15 @@ export const DEFAULT_COMPACTION_SETTINGS: CompactionSettings = {
 
 /**
  * Calculate total context tokens from usage.
- * Uses the native totalTokens field when available, falls back to computing from components.
+ * Uses prompt-relevant components only:
+ * - input: current user/tool payload sent to model
+ * - cacheRead/cacheWrite: context replay + newly cached context
+ *
+ * Excludes output because output tokens are not part of the current prompt size.
+ * Excludes totalTokens because some providers report cumulative loop/session totals.
  */
 export function calculateContextTokens(usage: Usage): number {
-	return usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;
+	return usage.input + usage.cacheRead + usage.cacheWrite;
 }
 
 /**

package/packages/pi-coding-agent/src/core/sdk.ts

@@ -573,7 +573,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 					const removed = modelRegistry.authStorage.removeLegacyOAuthCredential(resolvedProvider);
 					if (removed) {
 						console.warn(
-							`[auth] Removed unsupported Anthropic OAuth credential from auth.json (#3952).`,
+							`[auth] Removed unsupported Anthropic OAuth credential from auth.json.`,
 						);
 					}
 					if (isClaudeCodeBinaryInPath()) {

package/packages/pi-coding-agent/src/core/system-prompt.ts

@@ -203,9 +203,9 @@ export function buildSystemPrompt(options: BuildSystemPromptOptions = {}): strin
 		addGuideline("Prefer grep/find/ls tools over bash for file exploration (faster, respects .gitignore)");
 	}
 
-	// Read before edit guideline
-	if (hasRead && hasEdit) {
-		addGuideline("Use read to examine files before editing. You must use this tool instead of cat or sed.");
+	// Read before file mutation guideline
+	if (hasRead && (hasEdit || hasWrite)) {
+		addGuideline("Use read to examine relevant existing files before editing or overwriting. Before write creates or replaces a file, verify the target path; if it exists, read it first. Use read instead of cat or sed for file inspection.");
 	}
 
 	// Edit guideline
@@ -215,7 +215,7 @@ export function buildSystemPrompt(options: BuildSystemPromptOptions = {}): strin
 
 	// Write guideline
 	if (hasWrite) {
-		addGuideline("Use write only for new files or complete rewrites");
+		addGuideline("Use write only for new files or complete rewrites after verifying the target path");
 	}
 
 	// LSP guideline

package/packages/pi-coding-agent/src/modes/interactive/components/__tests__/tool-execution.test.ts

@@ -39,6 +39,7 @@ function renderToolCollapsed(
 		details?: Record<string, unknown>;
 	},
 	toolDefinition?: { label?: string; renderCall?: (...args: any[]) => any; renderResult?: (...args: any[]) => any },
+	width = 120,
 ): string {
 	const component = new ToolExecutionComponent(
 		toolName,
@@ -48,7 +49,7 @@ function renderToolCollapsed(
 		{ requestRender() {} } as any,
 	);
 	if (result) component.updateResult(result);
-	return stripAnsi(component.render(120).join("\n"));
+	return stripAnsi(component.render(width).join("\n"));
 }
 
 describe("ToolExecutionComponent", () => {
@@ -294,6 +295,21 @@ describe("ToolExecutionComponent", () => {
 		assert.doesNotMatch(rendered, /\bok\b/);
 	});
 
+	test("uses available row width for compact bash command text", () => {
+		const rendered = renderToolCollapsed(
+			"bash",
+			{
+				command:
+					'grep -n "expanded\\|toolOutputExpanded\\|setExpanded\\|defaultExpanded" /tmp/project/src/tool-execution.ts',
+			},
+			{ content: [{ type: "text", text: "ok" }], isError: false, details: { cwd: "/tmp/project" } },
+			undefined,
+			200,
+		);
+
+		assert.match(rendered, /defaultExpanded/);
+	});
+
 	test("keeps failed tools expanded and error visible", () => {
 		const rendered = renderToolCollapsed(
 			"edit",

package/packages/pi-coding-agent/src/modes/interactive/components/dynamic-border.test.ts

@@ -65,6 +65,26 @@ describe("DynamicBorder spinner", () => {
 		}
 	});
 
+	it("unref's the spinner interval so it never blocks process exit", () => {
+		// The pinned-zone spinner is purely cosmetic. A ref'd interval keeps the
+		// Node event loop alive until stopSpinner() runs — which made test
+		// processes hang and CI time out.
+		const border = new DynamicBorder((s) => s);
+		const tui = makeTUI();
+		try {
+			border.startSpinner(tui as any, (s) => s);
+			const interval = (border as any).spinnerInterval as NodeJS.Timeout;
+			assert.ok(interval, "startSpinner should create the animation interval");
+			assert.equal(
+				interval.hasRef(),
+				false,
+				"spinner interval must be unref'd so it never blocks process exit",
+			);
+		} finally {
+			border.stopSpinner();
+		}
+	});
+
 	it("updates lastExternalRender on each render() call", () => {
 		const border = new DynamicBorder((s) => s);
 		const anyBorder = border as any;

package/packages/pi-coding-agent/src/modes/interactive/components/dynamic-border.ts

@@ -46,6 +46,9 @@ export class DynamicBorder implements Component {
 				ui.requestRender();
 			}
 		}, 200);
+		// The spinner is purely cosmetic — it must never keep the Node event loop
+		// alive on its own (otherwise a process can hang waiting for it to stop).
+		this.spinnerInterval.unref?.();
 		ui.requestRender();
 	}
 

package/packages/pi-coding-agent/src/modes/interactive/components/footer.ts

@@ -20,6 +20,17 @@ function sanitizeStatusText(text: string): string {
 		.trim();
 }
 
+function truncateFooterPath(text: string, width: number): string {
+	if (visibleWidth(text) <= width) return text;
+	const tailMatch = text.match(/( \([^)]+\)(?: • .*)?)$/);
+	if (!tailMatch) return truncateToWidth(text, width, "...");
+	const tail = tailMatch[1];
+	const tailWidth = visibleWidth(tail);
+	if (tailWidth >= width - 4) return truncateToWidth(text, width, "...");
+	const head = text.slice(0, -tail.length);
+	return `${truncateToWidth(head, width - tailWidth, "...")}${tail}`;
+}
+
 /**
  * Format token counts (similar to web-ui)
  */
@@ -222,10 +233,6 @@ export class FooterComponent implements Component {
 			}
 		}
 
-		// Apply dim to the stats group before handing it to the shared footer strip.
-		// statsLeft may contain color codes for context %, so keep coloring local to the group.
-		const dimStatsLeft = theme.fg("dim", statsLeft);
-
 		// Extension statuses right-aligned on the pwd line (sorted by key).
 		// Keeps the footer compact by avoiding a dedicated row when the content
 		// fits alongside pwd. Falls back to pwd-only if the combined line would
@@ -239,12 +246,21 @@ export class FooterComponent implements Component {
 						.join(" ")
 				: "";
 
+		const footerRight = [rightSide, extStatusText].filter(Boolean).join(" ");
+		const gsdSegment = theme.fg("accent", "● GSD");
+		const dimStatsLeft = theme.fg("dim", statsLeft);
+		const innerWidth = Math.max(1, width - 2);
+		const rightWidth = visibleWidth(footerRight);
+		const leftBudget = footerRight ? Math.max(1, innerWidth - rightWidth - 3) : innerWidth;
+		const sepWidth = visibleWidth("  │  ");
+		const pwdBudget = Math.max(1, leftBudget - visibleWidth(gsdSegment) - visibleWidth(dimStatsLeft) - sepWidth * 2);
+		const pwdSegment = theme.fg("dim", truncateFooterPath(pwd, pwdBudget));
+
 		const leftSegments = [
-			theme.fg("accent", "● GSD"),
-			theme.fg("dim", pwd),
+			gsdSegment,
+			pwdSegment,
 			dimStatsLeft,
 		];
-		const footerRight = [rightSide, extStatusText].filter(Boolean).join(" ");
 		return renderFooterStrip(leftSegments, footerRight, width);
 	}
 }

package/packages/pi-coding-agent/src/modes/interactive/components/tool-execution.ts

@@ -117,7 +117,7 @@ function formatElapsed(ms: number): string {
 }
 
 function formatCommandPreview(command: string): string {
-	return truncateToWidth(command.replace(/\s+/g, " ").trim(), 64, "");
+	return command.replace(/\s+/g, " ").trim();
 }
 
 function appendLineOrRange(displayPath: string | undefined, target: ToolTargetMetadata): string | undefined {