@selesai/code 0.8.3 → 0.8.5

package/dist/extensions/pi-subagents/test/unit/completion-guard.test.ts

@@ -26,7 +26,7 @@ function assistantText(text: string): Message {
 
 test("implementation task with no mutation triggers the completion guard", () => {
 	const result = evaluateCompletionMutationGuard({
-		agent: "builder",
+		agent: "worker",
 		task: "Implement the approved fix",
 		messages: [assistantText("Plan: update the files...")],
 	});
@@ -53,10 +53,28 @@ test("declared read-only builtin tools suppress implementation-word false positi
 	});
 });
 
+test("hyphenated fix adjectives in review tasks do not trigger the completion guard", () => {
+	const result = evaluateCompletionMutationGuard({
+		agent: "worker",
+		task: "Return a review with the top 2-3 must-fix items.",
+		messages: [assistantText("Review: findings with severity labels.")],
+	});
+
+	assert.deepEqual(result, {
+		expectedMutation: false,
+		attemptedMutation: false,
+		triggered: false,
+	});
+	assert.equal(
+		expectsImplementationMutation("worker", "Return a review with the top 2-3 must-fix items."),
+		false,
+	);
+});
+
 test("read-only issue drafting tasks do not trigger on suggested fix wording", () => {
 	const task = "Draft GitHub issue for pi-subagents bug from current conversation. Include title, environment/context, reproduction steps, actual/expected, logs excerpt, suspected cause, suggested fix. Terse but complete. No tools needed.";
 	const result = evaluateCompletionMutationGuard({
-		agent: "explorer",
+		agent: "delegate",
 		task,
 		messages: [assistantText("Title: completionGuard false positive\n\nSuggested fix: model read-only intent.")],
 		tools: ["read", "grep", "find", "ls", "bash", "edit", "write", "contact_supervisor"],
@@ -67,9 +85,9 @@ test("read-only issue drafting tasks do not trigger on suggested fix wording", (
 		attemptedMutation: false,
 		triggered: false,
 	});
-	assert.equal(expectsImplementationMutation("builder", task), false);
+	assert.equal(expectsImplementationMutation("worker", task), false);
 	assert.equal(
-		expectsImplementationMutation("builder", "Draft GitHub issue for a bug. Include suspected cause and suggested fix."),
+		expectsImplementationMutation("worker", "Draft GitHub issue for a bug. Include suspected cause and suggested fix."),
 		false,
 	);
 });
@@ -89,9 +107,9 @@ test("omitted, bash, unknown, write, and MCP tool capabilities stay conservative
 	assert.equal(evaluateCompletionMutationGuard({ ...base, tools: ["read", "grep"], mcpDirectTools: ["github/search"] }).triggered, true);
 });
 
-test("builder with mutating-capable tools still triggers when no mutation is observed", () => {
+test("worker with mutating-capable tools still triggers when no mutation is observed", () => {
 	const result = evaluateCompletionMutationGuard({
-		agent: "builder",
+		agent: "worker",
 		task: "Fix the test implementation",
 		messages: [assistantText("I will edit it next")],
 		tools: ["read", "edit"],
@@ -104,10 +122,10 @@ test("builder with mutating-capable tools still triggers when no mutation is obs
 	});
 });
 
-test("commentator review tasks with bash available do not require mutation", () => {
+test("oracle review tasks with bash available do not require mutation", () => {
 	const task = "Review prep findings and determine what to implement with playbooks instead of before.";
 	const result = evaluateCompletionMutationGuard({
-		agent: "commentator",
+		agent: "oracle",
 		task,
 		messages: [assistantText("Review complete with file-backed findings.")],
 		tools: ["read", "grep", "find", "ls", "bash", "intercom"],
@@ -121,57 +139,57 @@ test("commentator review tasks with bash available do not require mutation", ()
 });
 
 test("review-only, research, and framework output instructions do not expect mutation", () => {
-	assert.equal(expectsImplementationMutation("builder", "Review only: return findings, do not edit"), false);
-	assert.equal(expectsImplementationMutation("builder", "Do not edit files. Tell me how to fix the bug."), false);
-	assert.equal(expectsImplementationMutation("builder", "Review the diff and suggest fixes only. Do not edit files."), false);
-	assert.equal(expectsImplementationMutation("builder", "Implement this. Do not edit files outside this repo. Do not edit files."), false);
-	assert.equal(expectsImplementationMutation("builder", "Investigate why this failed"), false);
+	assert.equal(expectsImplementationMutation("worker", "Review only: return findings, do not edit"), false);
+	assert.equal(expectsImplementationMutation("worker", "Do not edit files. Tell me how to fix the bug."), false);
+	assert.equal(expectsImplementationMutation("worker", "Review the diff and suggest fixes only. Do not edit files."), false);
+	assert.equal(expectsImplementationMutation("worker", "Implement this. Do not edit files outside this repo. Do not edit files."), false);
+	assert.equal(expectsImplementationMutation("worker", "Investigate why this failed"), false);
 	assert.equal(expectsImplementationMutation("researcher", "Research the API behavior"), false);
 	assert.equal(expectsImplementationMutation("researcher", "Research this and patch the bug"), false);
-	assert.equal(expectsImplementationMutation("commentator", "Review this and fix any real issues"), false);
-	assert.equal(expectsImplementationMutation("commentator", "Review this and fix any real issues; regardless of findings, apply changes directly"), true);
-	assert.equal(expectsImplementationMutation("builder", "[Write to: /tmp/result.md]\n\nSummarize findings"), false);
-	assert.equal(expectsImplementationMutation("builder", "Write report"), false);
-	assert.equal(expectsImplementationMutation("builder", "Create a report"), false);
-	assert.equal(expectsImplementationMutation("builder", "Create a summary"), false);
-	assert.equal(expectsImplementationMutation("builder", "Add a report"), false);
-	assert.equal(expectsImplementationMutation("builder", "Update a summary"), false);
-	assert.equal(expectsImplementationMutation("builder", "Write to {chain_dir}"), false);
+	assert.equal(expectsImplementationMutation("reviewer", "Review this and fix any real issues"), false);
+	assert.equal(expectsImplementationMutation("reviewer", "Review this and fix any real issues; regardless of findings, apply changes directly"), true);
+	assert.equal(expectsImplementationMutation("worker", "[Write to: /tmp/result.md]\n\nSummarize findings"), false);
+	assert.equal(expectsImplementationMutation("worker", "Write report"), false);
+	assert.equal(expectsImplementationMutation("worker", "Create a report"), false);
+	assert.equal(expectsImplementationMutation("worker", "Create a summary"), false);
+	assert.equal(expectsImplementationMutation("worker", "Add a report"), false);
+	assert.equal(expectsImplementationMutation("worker", "Update a summary"), false);
+	assert.equal(expectsImplementationMutation("worker", "Write to {chain_dir}"), false);
 	assert.equal(
-		expectsImplementationMutation("builder", "Do async work\nUpdate progress at: /tmp/progress.md\n**Output:**\nWrite your findings to exactly this path: /tmp/out.md\nThis path is authoritative for this run.\nIgnore any other output filename or output path mentioned elsewhere."),
+		expectsImplementationMutation("worker", "Do async work\nUpdate progress at: /tmp/progress.md\n**Output:**\nWrite your findings to exactly this path: /tmp/out.md\nThis path is authoritative for this run.\nIgnore any other output filename or output path mentioned elsewhere."),
 		false,
 	);
 });
 
-test("builder implementation verbs win over investigative wording and scoped prohibitions", () => {
-	assert.equal(expectsImplementationMutation("builder", "Investigate why the builder did not edit files and fix it"), true);
-	assert.equal(expectsImplementationMutation("builder", "Do not modify tests; implement the fix"), true);
-	assert.equal(expectsImplementationMutation("builder", "Do not modify tests — implement the fix"), true);
-	assert.equal(expectsImplementationMutation("builder", "Research the current code path and patch the bug"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix the bug where no edits were made"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix lint"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix the build"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix TypeScript errors"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix CI"), true);
-	assert.equal(expectsImplementationMutation("builder", "Fix the failing test"), true);
-	assert.equal(expectsImplementationMutation("builder", "Patch the cold start test"), true);
-	assert.equal(expectsImplementationMutation("builder", "Implement the fix and return findings."), true);
+test("worker implementation verbs win over investigative wording and scoped prohibitions", () => {
+	assert.equal(expectsImplementationMutation("worker", "Investigate why the worker did not edit files and fix it"), true);
+	assert.equal(expectsImplementationMutation("worker", "Do not modify tests; implement the fix"), true);
+	assert.equal(expectsImplementationMutation("worker", "Do not modify tests — implement the fix"), true);
+	assert.equal(expectsImplementationMutation("worker", "Research the current code path and patch the bug"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix the bug where no edits were made"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix lint"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix the build"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix TypeScript errors"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix CI"), true);
+	assert.equal(expectsImplementationMutation("worker", "Fix the failing test"), true);
+	assert.equal(expectsImplementationMutation("worker", "Patch the cold start test"), true);
+	assert.equal(expectsImplementationMutation("worker", "Implement the fix and return findings."), true);
 });
 
-test("non-builder implementation tasks still expect mutation", () => {
-	assert.equal(expectsImplementationMutation("explorer", "Fix the bug where no edits were made"), true);
-	assert.equal(expectsImplementationMutation("explorer", "Apply the suggested fix to src/runs/shared/completion-guard.ts"), true);
-	assert.equal(expectsImplementationMutation("builder", "Draft a GitHub issue, then implement the fix"), true);
+test("non-worker implementation tasks still expect mutation", () => {
+	assert.equal(expectsImplementationMutation("delegate", "Fix the bug where no edits were made"), true);
+	assert.equal(expectsImplementationMutation("delegate", "Apply the suggested fix to src/runs/shared/completion-guard.ts"), true);
+	assert.equal(expectsImplementationMutation("worker", "Draft a GitHub issue, then implement the fix"), true);
 });
 
-test("builder edit intent covers common docs, config, and source tasks", () => {
-	assert.equal(expectsImplementationMutation("builder", "Update README to mention the native tool"), true);
-	assert.equal(expectsImplementationMutation("builder", "Remove share functionality and all Vercel references"), true);
-	assert.equal(expectsImplementationMutation("builder", "Replace the registered command with a render tool"), true);
-	assert.equal(expectsImplementationMutation("builder", "Create completion-guard.ts"), true);
-	assert.equal(expectsImplementationMutation("builder", "Add tests for the completion guard"), true);
-	assert.equal(expectsImplementationMutation("builder", "Implement the approved fixes. Do not edit files outside this repo."), true);
-	assert.equal(expectsImplementationMutation("builder", "Implement the fix. Do not edit unrelated files."), true);
+test("worker edit intent covers common docs, config, and source tasks", () => {
+	assert.equal(expectsImplementationMutation("worker", "Update README to mention the native tool"), true);
+	assert.equal(expectsImplementationMutation("worker", "Remove share functionality and all Vercel references"), true);
+	assert.equal(expectsImplementationMutation("worker", "Replace the registered command with a render tool"), true);
+	assert.equal(expectsImplementationMutation("worker", "Create completion-guard.ts"), true);
+	assert.equal(expectsImplementationMutation("worker", "Add tests for the completion guard"), true);
+	assert.equal(expectsImplementationMutation("worker", "Implement the approved fixes. Do not edit files outside this repo."), true);
+	assert.equal(expectsImplementationMutation("worker", "Implement the fix. Do not edit unrelated files."), true);
 });
 
 test("edit and write tool calls count as mutation attempts", () => {
@@ -225,7 +243,7 @@ test("read-only and quoted git commands do not count as mutation attempts", () =
 
 test("implementation task with mutation attempts does not trigger", () => {
 	const result = evaluateCompletionMutationGuard({
-		agent: "builder",
+		agent: "worker",
 		task: "Fix the failing test",
 		messages: [assistantToolCall("edit", { path: "test.ts" })],
 	});
@@ -309,7 +327,7 @@ test("provider checkpoint with a changed commit counts as mutation evidence", ()
 		assert.equal(hasMutationToolCall([message]), true);
 		assert.equal(
 			evaluateCompletionMutationGuard({
-				agent: "builder",
+				agent: "worker",
 				task: "Edit the target source file",
 				messages: [message],
 			}).triggered,
@@ -327,7 +345,7 @@ test("unchanged provider checkpoint does not bypass the completion guard", () =>
 		assert.equal(hasMutationToolCall([message]), false);
 		assert.equal(
 			evaluateCompletionMutationGuard({
-				agent: "builder",
+				agent: "worker",
 				task: "Edit the target source file",
 				messages: [message],
 			}).triggered,
@@ -338,7 +356,7 @@ test("unchanged provider checkpoint does not bypass the completion guard", () =>
 
 test("implementation task with Cursor edit thinking does not trigger", () => {
 	const result = evaluateCompletionMutationGuard({
-		agent: "builder",
+		agent: "worker",
 		task: "Edit docs/BACKEND_ARCHITECTURE.md",
 		messages: [
 			assistantThinking("Cursor edit: docs/BACKEND_ARCHITECTURE.md added 1 line\n"),

package/dist/extensions/pi-subagents/test/unit/completion-replay.test.ts

@@ -0,0 +1,187 @@
+import assert from "node:assert/strict";
+import * as fs from "node:fs";
+import { createRequire, syncBuiltinESMExports } from "node:module";
+import * as os from "node:os";
+import * as path from "node:path";
+import { describe, it } from "node:test";
+import { cleanupCompletionReplay, completionArchivePath, readCompletionArchive, readCompletionReplay, writeCompletionArchive } from "../../src/runs/background/completion-replay.ts";
+import { utf8Tail } from "../../src/shared/utf8.ts";
+import { collectWaitCompletions, recordWaitCompletion } from "../../src/runs/background/wait-completions.ts";
+import type { AsyncRunSummary } from "../../src/runs/background/async-status.ts";
+import type { SubagentState } from "../../src/shared/types.ts";
+
+const require = createRequire(import.meta.url);
+const fsCjs = require("node:fs") as typeof fs;
+
+function makeState(): SubagentState {
+	return {
+		baseCwd: "",
+		currentSessionId: "session-a",
+		asyncJobs: new Map(),
+		foregroundControls: new Map(),
+		lastForegroundControlId: null,
+		cleanupTimers: new Map(),
+		lastUiContext: null,
+		poller: null,
+		completionSeen: new Map(),
+		watcher: null,
+		watcherRestartTimer: null,
+		resultFileCoalescer: { schedule: () => false, clear: () => {} },
+	} as SubagentState;
+}
+
+describe("completion replay", () => {
+	it("surfaces a consumed completion after in-memory watcher state is lost", () => {
+		const resultsDir = fs.mkdtempSync(path.join(os.tmpdir(), "pi-completion-replay-"));
+		try {
+			const now = Date.now();
+			recordWaitCompletion(makeState(), "run-a", {
+				agent: "worker",
+				mode: "single",
+				state: "complete",
+				success: true,
+				results: [{ agent: "worker", success: true, outputState: "present", output: "finished output" }],
+			}, now, 60_000, { resultsDir, sessionId: "session-a" });
+
+			const replay = readCompletionReplay(resultsDir, "run-a", { sessionId: "session-a", now: now + 1 });
+			assert.equal(replay?.version, 1);
+			assert.equal(replay?.completion.archivePath, replay?.archivePath);
+			assert.equal(readCompletionArchive(replay!.archivePath)?.entries[0]?.text, "[worker]\nfinished output");
+
+			const terminal = [{ id: "run-a", sessionId: "session-a" }] as AsyncRunSummary[];
+			const completions = collectWaitCompletions(terminal, makeState(), resultsDir);
+			assert.equal(completions?.[0]?.runId, "run-a");
+			assert.equal(completions?.[0]?.results?.[0]?.agent, "worker");
+			assert.equal(completions?.[0]?.archivePath, replay?.archivePath);
+		} finally {
+			fs.rmSync(resultsDir, { recursive: true, force: true });
+		}
+	});
+
+	it("keeps UTF-8 tails valid at multibyte boundaries", () => {
+		const bounded = utf8Tail(`start-${"é".repeat(10)}-tail`, 9);
+		assert.equal(bounded.truncated, true);
+		assert.equal(bounded.text, "éé-tail");
+	});
+
+	it("does not delete untrusted or cross-run archive paths from replay records", () => {
+		const root = fs.mkdtempSync(path.join(os.tmpdir(), "pi-completion-replay-untrusted-"));
+		try {
+			const resultsDir = path.join(root, "results");
+			const replayDir = path.join(resultsDir, "completion-replay");
+			fs.mkdirSync(replayDir, { recursive: true });
+			const victim = path.join(root, "victim.txt");
+			fs.writeFileSync(victim, "keep", "utf-8");
+			fs.writeFileSync(path.join(replayDir, `${encodeURIComponent("run-c")}.json`), JSON.stringify({
+				version: 1,
+				runId: "run-c",
+				sessionId: "session-a",
+				completedAt: 1,
+				expiresAt: 2,
+				archivePath: victim,
+				completion: { runId: "run-c", archivePath: victim },
+			}), "utf-8");
+
+			const replayPath = path.join(replayDir, `${encodeURIComponent("run-c")}.json`);
+			assert.equal(readCompletionReplay(resultsDir, "run-c", { sessionId: "session-a", now: 1 }), undefined);
+			assert.equal(fs.readFileSync(victim, "utf-8"), "keep");
+			assert.equal(fs.existsSync(replayPath), false);
+
+			fs.writeFileSync(replayPath, JSON.stringify({
+				version: 1,
+				runId: "run-c",
+				sessionId: "session-a",
+				completedAt: 1,
+				expiresAt: 2,
+				archivePath: victim,
+				completion: { runId: "run-c", archivePath: victim },
+			}), "utf-8");
+			cleanupCompletionReplay(resultsDir, 3, 60_000);
+			assert.equal(fs.readFileSync(victim, "utf-8"), "keep");
+			assert.equal(fs.existsSync(replayPath), false);
+
+			const victimArchive = completionArchivePath(resultsDir, "run-a");
+			fs.mkdirSync(path.dirname(victimArchive), { recursive: true });
+			fs.writeFileSync(victimArchive, "keep", "utf-8");
+			fs.writeFileSync(replayPath, JSON.stringify({
+				version: 1,
+				runId: "run-a",
+				sessionId: "session-a",
+				completedAt: 1,
+				expiresAt: 2,
+				archivePath: victimArchive,
+				completion: { runId: "run-a", archivePath: victimArchive },
+			}), "utf-8");
+			cleanupCompletionReplay(resultsDir, 3, 60_000);
+			assert.equal(fs.readFileSync(victimArchive, "utf-8"), "keep");
+			assert.equal(fs.existsSync(replayPath), false);
+		} finally {
+			fs.rmSync(root, { recursive: true, force: true });
+		}
+	});
+
+	it("treats replay read deletion failures as best-effort cleanup", () => {
+		const root = fs.mkdtempSync(path.join(os.tmpdir(), "pi-completion-replay-rm-failure-"));
+		const originalRmSync = fsCjs.rmSync;
+		try {
+			const resultsDir = path.join(root, "results");
+			const replayDir = path.join(resultsDir, "completion-replay");
+			const replayPath = path.join(replayDir, `${encodeURIComponent("run-d")}.json`);
+			const archivePath = completionArchivePath(resultsDir, "run-d");
+			fs.mkdirSync(replayDir, { recursive: true });
+			fs.mkdirSync(path.dirname(archivePath), { recursive: true });
+			fs.writeFileSync(archivePath, "archive", "utf-8");
+			fs.writeFileSync(replayPath, JSON.stringify({
+				version: 1,
+				runId: "run-d",
+				sessionId: "session-a",
+				completedAt: 1,
+				expiresAt: 2,
+				archivePath,
+				completion: { runId: "run-d", archivePath },
+			}), "utf-8");
+
+			fsCjs.rmSync = ((target: Parameters<typeof fs.rmSync>[0], options?: Parameters<typeof fs.rmSync>[1]) => {
+				if (String(target) === replayPath || String(target) === archivePath) {
+					const error = new Error("busy") as NodeJS.ErrnoException;
+					error.code = "EPERM";
+					throw error;
+				}
+				return originalRmSync(target, options);
+			}) as typeof fs.rmSync;
+			syncBuiltinESMExports();
+
+			assert.equal(readCompletionReplay(resultsDir, "run-d", { sessionId: "session-a", now: 3 }), undefined);
+			assert.equal(fs.existsSync(replayPath), true);
+			assert.equal(fs.existsSync(archivePath), true);
+		} finally {
+			fsCjs.rmSync = originalRmSync;
+			syncBuiltinESMExports();
+			fs.rmSync(root, { recursive: true, force: true });
+		}
+	});
+
+	it("prefers saved outputs and bounds fallback output tails", () => {
+		const root = fs.mkdtempSync(path.join(os.tmpdir(), "pi-completion-archive-"));
+		try {
+			const savedOutput = path.join(root, "saved-output.md");
+			fs.writeFileSync(savedOutput, "saved", "utf-8");
+			const archivePath = writeCompletionArchive(root, "run-b", {
+				results: [
+					{ agent: "saved", output: "duplicate text", artifactPaths: { outputPath: savedOutput } },
+					{ agent: "fallback", output: `start-${"x".repeat(70 * 1024)}-tail` },
+				],
+			}, Date.now());
+			const archive = readCompletionArchive(archivePath);
+			assert.deepEqual(archive?.entries[0], { agent: "saved", source: "output-artifact", path: savedOutput });
+			const fallback = archive?.entries[1];
+			assert.equal(fallback?.source, "result-tail");
+			assert.equal(fallback?.truncated, true);
+			assert.ok(Buffer.byteLength(fallback?.text ?? "", "utf-8") <= 64 * 1024);
+			assert.match(fallback?.text ?? "", /-tail$/);
+			assert.equal((fallback?.text ?? "").includes("duplicate text"), false);
+		} finally {
+			fs.rmSync(root, { recursive: true, force: true });
+		}
+	});
+});