@selesai/code 0.6.3 → 0.8.0

package/dist/extensions/pi-subagents/test/e2e/real-session-subagent.test.ts

@@ -13,7 +13,6 @@
 
 import { afterEach, describe, it } from "node:test";
 import assert from "node:assert/strict";
-import * as fs from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
 import { tryImport } from "../support/helpers.ts";
@@ -24,17 +23,6 @@ const piAi = await tryImport<unknown>("@earendil-works/pi-ai");
 const available = Boolean(piCodingAgent && piAi);
 
 const CHILD_MARKER = "CHILD_REAL_SESSION_OK";
-
-/**
- * Reference-first tool results: completion content carries "Output saved to:
- * <path> (…)". Read the durable file to inspect the full child output.
- */
-function readSavedOutput(text: string): string {
-	const rest = text.split("Output saved to: ")[1];
-	assert.ok(rest, `expected a saved-output reference in: ${text.slice(0, 160)}`);
-	const outputPath = rest.split(" (")[0]!;
-	return fs.readFileSync(outputPath, "utf-8");
-}
 // Env vars the runner must clear so a parent that was itself spawned as a
 // subagent child can still launch fresh children. The values are deliberately
 // bogus sentinels (nonexistent paths) so a leaked value would break spawning.
@@ -49,7 +37,7 @@ const ISOLATED_ENV_KEYS = [
 	"SELESAI_SUBAGENT_EXTRA_AGENT_DIRS",
 	"SELESAI_SUBAGENT_PARENT_SESSION",
 	"SELESAI_SUBAGENT_PI_BINARY",
-	"SELESAI_SUBAGENTS_SELESAI_CODING_AGENT_PACKAGE_ROOT",
+	"SELESAI_SUBAGENTS_SELESAI_PACKAGE_ROOT",
 ] as const;
 
 describe("real Pi-session subagent E2E", { skip: !available ? "pi runtime packages not available" : undefined }, () => {
@@ -60,10 +48,10 @@ describe("real Pi-session subagent E2E", { skip: !available ? "pi runtime packag
 		run = undefined;
 	});
 
-	it("loads requested extension tools in direct and chain children and diagnoses missing providers", async () => {
+	it("loads requested extension tools in direct and workflow children and diagnoses missing providers", async () => {
 		const { runRealSubagentSession, subagentCall, subagentToolResults } = await import("../support/real-session-runner.ts");
 		const extensionAgent = `---
-name: extension-worker
+name: extension-builder
 description: Uses a child-only fixture tool
 tools: read, fixture_search
 subagentOnlyExtensions: ./fixture-extension.ts
@@ -71,7 +59,7 @@ completionGuard: false
 ---
 Use the available tools.`;
 		const missingAgent = `---
-name: missing-extension-worker
+name: missing-extension-builder
 description: Requests an extension tool without loading its provider
 tools: read, missing_search
 completionGuard: false
@@ -88,41 +76,42 @@ Use the available tools.`;
 }`;
 
 		run = await runRealSubagentSession({
-			prompt: "Run the direct, chain, and missing-provider child checks.",
+			prompt: "Run the direct, workflow, and missing-provider child checks.",
 			childText: CHILD_MARKER,
 			reportChildTools: true,
 			projectFiles: {
-				".selesai/agents/extension-worker.md": extensionAgent,
-				".selesai/agents/missing-extension-worker.md": missingAgent,
+				".selesai/agents/extension-builder.md": extensionAgent,
+				".selesai/agents/missing-extension-builder.md": missingAgent,
 				"fixture-extension.ts": fixtureExtension,
 			},
 			respond(context) {
 				const resultCount = (context.messages as Array<{ role?: string; toolName?: string }>).filter((message) => message.role === "toolResult" && message.toolName === "subagent").length;
 				if (resultCount === 0) {
-					return subagentCall({ agent: "extension-worker", task: "Report active tools.", context: "fresh", agentScope: "project" }, "call-direct-extension");
+					return subagentCall({ agent: "extension-builder", task: "Report active tools.", context: "fresh", async: false, clarify: false, agentScope: "project" }, "call-direct-extension");
 				}
 				if (resultCount === 1) {
-					return subagentCall({ chain: [{ agent: "extension-worker", task: "Report active tools." }], async: false, clarify: false, agentScope: "project" }, "call-chain-extension");
-				}
-				if (resultCount === 2) {
 					return subagentCall({
-						chain: [{
-							agent: "extension-worker",
-							task: "Submit the required structured marker.",
-							outputSchema: {
-								type: "object",
-								properties: { marker: { type: "string" } },
-								required: ["marker"],
-								additionalProperties: false,
+						workflowScript: `return await runs.all([
+							{ key: "extension", agent: "extension-builder", task: "Report active tools." },
+							{
+								key: "structured",
+								agent: "extension-builder",
+								task: "Submit the required structured marker.",
+								outputSchema: {
+									type: "object",
+									properties: { marker: { type: "string" } },
+									required: ["marker"],
+									additionalProperties: false,
+								},
 							},
-						}],
+						]);`,
 						async: false,
 						clarify: false,
 						agentScope: "project",
-					}, "call-structured-output");
+					}, "call-workflow-extension");
 				}
-				if (resultCount === 3) {
-					return subagentCall({ agent: "missing-extension-worker", task: "Report active tools.", context: "fresh", agentScope: "project" }, "call-missing-extension");
+				if (resultCount === 2) {
+					return subagentCall({ agent: "missing-extension-builder", task: "Report active tools.", context: "fresh", async: false, clarify: false, agentScope: "project" }, "call-missing-extension");
 				}
 				return "Child tool checks complete.";
 			},
@@ -131,26 +120,21 @@ Use the available tools.`;
 
 		const results = subagentToolResults(run.parentSession);
 		const toolMessages = run.parentSession.messages.filter((message) => message.role === "toolResult" && (message as { toolName?: string }).toolName === "subagent");
-		const chainDetails = JSON.stringify((toolMessages[1] as { details?: unknown } | undefined)?.details);
-		const structuredDetails = JSON.stringify((toolMessages[2] as { details?: unknown } | undefined)?.details);
-		assert.equal(results.length, 4);
-		const directOutput = readSavedOutput(results[0] ?? "");
-		assert.match(directOutput, /ACTIVE_TOOLS:[^\n]*fixture_search/);
-		assert.match(directOutput, /ACTIVE_TOOLS:[^\n]*read/);
-		const chainFirstChild = (toolMessages[1] as { details?: { results?: Array<{ savedOutputPath?: string }> } } | undefined)?.details?.results?.[0];
-		assert.ok(chainFirstChild?.savedOutputPath, "chain details should carry the saved output path");
-		const chainOutput = fs.readFileSync(chainFirstChild.savedOutputPath, "utf-8");
-		assert.match(chainOutput, /ACTIVE_TOOLS:[^\n]*fixture_search/);
-		assert.match(chainOutput, /ACTIVE_TOOLS:[^\n]*read/);
-		assert.match(structuredDetails, /STRUCTURED_OUTPUT_OK/);
-		assert.match(results[3] ?? "", /requested unavailable child tools: missing_search/);
-		assert.match(results[3] ?? "", /subagentOnlyExtensions/);
+		const workflowDetails = JSON.stringify((toolMessages[1] as { details?: unknown } | undefined)?.details);
+		assert.equal(results.length, 3);
+		assert.match(results[0] ?? "", /ACTIVE_TOOLS:[^\n]*fixture_search/);
+		assert.match(results[0] ?? "", /ACTIVE_TOOLS:[^\n]*read/);
+		assert.match(workflowDetails, /ACTIVE_TOOLS:[^\n]*fixture_search/);
+		assert.match(workflowDetails, /ACTIVE_TOOLS:[^\n]*read/);
+		assert.match(workflowDetails, /STRUCTURED_OUTPUT_OK/);
+		assert.match(results[2] ?? "", /requested unavailable child tools: missing_search/);
+		assert.match(results[2] ?? "", /subagentOnlyExtensions/);
 	});
 
 	it("accepts child tools registered by async before_agent_start hooks", async () => {
 		const { runRealSubagentSession, subagentCall, subagentToolResults } = await import("../support/real-session-runner.ts");
 		const asyncExtensionAgent = `---
-name: async-extension-worker
+name: async-extension-builder
 description: Uses a child-only tool registered from before_agent_start
 tools: read, fixture_async_search
 subagentOnlyExtensions: ./fixture-async-extension.ts
@@ -175,21 +159,20 @@ Report active tools.`;
 			childText: CHILD_MARKER,
 			reportChildTools: true,
 			projectFiles: {
-				".selesai/agents/async-extension-worker.md": asyncExtensionAgent,
+				".selesai/agents/async-extension-builder.md": asyncExtensionAgent,
 				"fixture-async-extension.ts": fixtureAsyncExtension,
 			},
 			respond(context) {
 				const resultCount = (context.messages as Array<{ role?: string; toolName?: string }>).filter((message) => message.role === "toolResult" && message.toolName === "subagent").length;
 				if (resultCount > 0) return "Async child tool check complete.";
-				return subagentCall({ agent: "async-extension-worker", task: "Report active tools.", context: "fresh", agentScope: "project" }, "call-async-extension");
+				return subagentCall({ agent: "async-extension-builder", task: "Report active tools.", context: "fresh", async: false, clarify: false, agentScope: "project" }, "call-async-extension");
 			},
 			timeoutMs: 60_000,
 		});
 
 		const results = subagentToolResults(run.parentSession);
 		assert.equal(results.length, 1);
-		const asyncOutput = readSavedOutput(results[0] ?? "");
-		assert.match(asyncOutput, /ACTIVE_TOOLS:[^\n]*fixture_async_search/);
+		assert.match(results[0] ?? "", /ACTIVE_TOOLS:[^\n]*fixture_async_search/);
 		assert.doesNotMatch(results[0] ?? "", /requested unavailable child tools/);
 	});
 
@@ -204,7 +187,7 @@ Report active tools.`;
 		process.env.SELESAI_SUBAGENT_EXTRA_AGENT_DIRS = BOGUS_EXTRA_DIRS;
 		process.env.SELESAI_SUBAGENT_PARENT_SESSION = "polluted-parent";
 		process.env.SELESAI_SUBAGENT_PI_BINARY = BOGUS_PI_BINARY;
-		process.env.SELESAI_SUBAGENTS_SELESAI_CODING_AGENT_PACKAGE_ROOT = BOGUS_PI_PACKAGE_ROOT;
+		process.env.SELESAI_SUBAGENTS_SELESAI_PACKAGE_ROOT = BOGUS_PI_PACKAGE_ROOT;
 
 		try {
 			run = await runRealSubagentSession({
@@ -216,6 +199,8 @@ Report active tools.`;
 						agent: "builder",
 						task: "Return the marker from the faux child provider.",
 						context: "fresh",
+						async: false,
+						clarify: false,
 						agentScope: "project",
 					},
 				}),
@@ -223,7 +208,7 @@ Report active tools.`;
 
 			const toolResults = subagentToolResults(run.parentSession);
 			assert.equal(toolResults.length, 1);
-			assert.match(readSavedOutput(toolResults[0]!), new RegExp(CHILD_MARKER));
+			assert.match(toolResults[0]!, new RegExp(CHILD_MARKER));
 			assert.match(run.responseText, new RegExp(CHILD_MARKER));
 			assert.doesNotMatch(run.responseText, /CHILD_MISSING/);
 			assert.ok(run.modelCalls >= 2, `expected parent tool-call and final turns, got ${run.modelCalls}`);
@@ -236,92 +221,4 @@ Report active tools.`;
 			}
 		}
 	});
-
-	function latestSubagentToolResultText(messages: Array<{ role?: string; toolName?: string; content?: unknown }>): string | undefined {
-		for (let i = messages.length - 1; i >= 0; i--) {
-			const message = messages[i]!;
-			if (message.role === "toolResult" && message.toolName === "subagent") {
-				return Array.isArray(message.content)
-					? message.content
-						.map((part) => part && typeof part === "object" && (part as { type?: unknown }).type === "text"
-							? String((part as { text?: unknown }).text ?? "")
-							: "")
-						.join("")
-					: "";
-			}
-		}
-		return undefined;
-	}
-
-	it("lists then delegates to a non-bundled discovered writer in a broad-mutation request", async () => {
-		const { runRealSubagentSession, subagentCall, subagentToolResults } = await import("../support/real-session-runner.ts");
-		const writerAgent = `---
-name: fixture-writer
-description: Scoped mutation-capable fixture writer
-aliases: fw
-tools: read, grep, find, ls, bash, edit, write
-acceptanceRole: writer
-defaultContext: fork
-completionGuard: false
----
-Implement the scoped fixture change and return the marker.`;
-
-		run = await runRealSubagentSession({
-			prompt: "Implement the fixture change across the codebase.",
-			childText: CHILD_MARKER,
-			projectFiles: {
-				".selesai/agents/fixture-writer.md": writerAgent,
-			},
-			respond(context) {
-				const messages = context.messages as Array<{ role?: string; toolName?: string; content?: unknown; details?: unknown }>;
-				const subagentResults = messages.filter((message) => message.role === "toolResult" && message.toolName === "subagent");
-				if (subagentResults.length === 0) {
-					return subagentCall({ action: "list", agentScope: "project" }, "call-list-writer");
-				}
-				if (subagentResults.length === 1) {
-					const listText = latestSubagentToolResultText(messages) ?? "";
-					assert.match(
-						listText,
-						/- fixture-writer \(project, context: fork, role: writer, aliases: fw, tools: read, grep, find, ls, bash, edit, write\)/,
-						"catalog must expose the custom writer with its runtime metadata",
-					);
-					const listedDetails = JSON.stringify(subagentResults.at(-1)?.details ?? {});
-					assert.match(listedDetails, /"catalog"/);
-					assert.match(listedDetails, /"fixture-writer"/);
-					assert.match(listedDetails, /"acceptanceRole":"writer"/);
-					return subagentCall(
-						{ agent: "fixture-writer", task: "Implement the change and return the marker.", context: "fresh", agentScope: "project" },
-						"call-fixture-writer",
-					);
-				}
-				return "Broad mutation work complete.";
-			},
-			timeoutMs: 60_000,
-		});
-
-		const results = subagentToolResults(run.parentSession);
-		assert.equal(results.length, 2);
-		assert.match(results[0] ?? "", /fixture-writer \(project, context: fork, role: writer/);
-		assert.match(readSavedOutput(results[1] ?? ""), new RegExp(CHILD_MARKER));
-	});
-
-	it("keeps tiny targeted reads local without a subagent call", async () => {
-		const { runRealSubagentSession, subagentToolResults } = await import("../support/real-session-runner.ts");
-		run = await runRealSubagentSession({
-			prompt: "What does the README say about subagents?",
-			childText: CHILD_MARKER,
-			projectFiles: {
-				"README.md": "Subagents are delegated workers.",
-			},
-			respond() {
-				return "The README says: Subagents are delegated workers.";
-			},
-			timeoutMs: 60_000,
-		});
-
-		const results = subagentToolResults(run.parentSession);
-		assert.equal(results.length, 0);
-		assert.match(run.responseText, /Subagents are delegated workers/);
-		assert.ok(run.modelCalls >= 1, `expected at least one parent turn, got ${run.modelCalls}`);
-	});
 });

package/dist/extensions/pi-subagents/test/integration/acceptance-file-report.test.ts

@@ -158,7 +158,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 			const outputPath = path.join(tempDir, "report.md");
 			conflictingReportsCall(outputPath, "satisfied", "not-satisfied");
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-file-only",
 				outputPath,
 				outputMode: "file-only",
@@ -179,7 +179,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 				writeFiles: [{ path: outputPath, content: fileReport }],
 			});
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-foreground-saved-receipt",
 				outputPath,
 				outputMode: "file-only",
@@ -203,7 +203,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 			const artifactsDir = path.join(tempDir, "report-only-artifacts");
 			mockPi.onCall({ output: acceptanceReport("satisfied", "report-only evidence") });
 
-			const result = await runSync!(tempDir, [makeAgent("worker", { completionGuard: false })], "worker", "Implement and report the fix.", {
+			const result = await runSync!(tempDir, [makeAgent("builder", { completionGuard: false })], "builder", "Implement and report the fix.", {
 				runId: "acceptance-report-only",
 				acceptance: { level: "checked", criteria: ["Report the findings"] },
 				artifactsDir,
@@ -225,7 +225,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 			const artifactsDir = path.join(tempDir, "rejected-artifacts");
 			conflictingReportsCall(outputPath, "satisfied", "not-satisfied");
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-inline-text-first",
 				outputPath,
 				acceptance: { level: "checked", criteria: ["Report the findings"] },
@@ -248,7 +248,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 			const outputPath = path.join(tempDir, "report.md");
 			conflictingReportsCall(outputPath, "not-satisfied", "satisfied");
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-inline-file-fallback-only",
 				outputPath,
 				acceptance: { level: "checked", criteria: ["Report the findings"] },
@@ -281,7 +281,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 				],
 			});
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-failed-write",
 				outputPath,
 				outputMode: "file-only",
@@ -304,7 +304,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 				writeFiles: [{ path: outputPath, content: malformedReport }],
 			});
 
-			const result = await runSync!(tempDir, makeAgentConfigs(["worker"]), "worker", "Write the findings report.", {
+			const result = await runSync!(tempDir, makeAgentConfigs(["builder"]), "builder", "Write the findings report.", {
 				runId: "acceptance-malformed-file-report",
 				outputPath,
 				outputMode: "file-only",
@@ -313,16 +313,16 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 
 			assert.equal(result.acceptance?.status, "rejected");
 			assert.equal(result.exitCode, 1);
-			assert.match(result.error ?? "", /Empty or unterminated acceptance-report fence.*configured output/);
+			assert.match(result.error ?? "", /Failed to parse acceptance-report: Expected property name.*configured output/);
 		});
 	});
 
 	describe("background runner", { skip: isAsyncAvailable && !isAsyncAvailable() ? "jiti not available" : undefined }, () => {
 		function runAsyncSingle(id: string, outputPath: string, outputMode: "inline" | "file-only", artifactConfig = DISABLED_ARTIFACTS) {
 			executeAsyncSingle!(id, {
-				agent: "worker",
+				agent: "builder",
 				task: "Write the findings report.",
-				agentConfig: makeAgent("worker", { completionGuard: false }),
+				agentConfig: makeAgent("builder", { completionGuard: false }),
 				ctx: { pi: { events: { emit() {} } }, cwd: tempDir, currentSessionId: "session-file-report" },
 				artifactConfig,
 				artifactsDir: path.join(tempDir, ".pi-subagents", "artifacts"),
@@ -350,7 +350,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 
 		it("file-only mode persists async acceptance metadata when final assistant text is only a receipt", async () => {
 			const outputPath = path.join(tempDir, "saved-review.md");
-			const fileReport = `# Review\n${acceptanceReport("satisfied", "saved reviewer verdict")}`;
+			const fileReport = `# Review\n${acceptanceReport("satisfied", "saved commentator verdict")}`;
 			mockPi.onCall({
 				jsonl: [...events.completedWrite(outputPath, fileReport), events.assistantMessage("Output saved to the configured file.")],
 				writeFiles: [{ path: outputPath, content: fileReport }],
@@ -361,12 +361,12 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 			const payload = await waitForAsyncResult(id);
 			assert.equal(payload.success, true);
 			assert.equal(payload.results[0]?.acceptance?.status, "checked");
-			assert.equal(payload.results[0]?.acceptance?.childReport?.criteriaSatisfied?.[0]?.evidence, "saved reviewer verdict");
+			assert.equal(payload.results[0]?.acceptance?.childReport?.criteriaSatisfied?.[0]?.evidence, "saved commentator verdict");
 			const metadataPath = payload.results[0]?.artifactPaths?.metadataPath;
 			assert.ok(metadataPath);
 			const metadata = JSON.parse(fs.readFileSync(metadataPath, "utf-8")) as { acceptance?: AcceptanceSummary };
 			assert.equal(metadata.acceptance?.status, "checked");
-			assert.equal(metadata.acceptance?.childReport?.criteriaSatisfied?.[0]?.evidence, "saved reviewer verdict");
+			assert.equal(metadata.acceptance?.childReport?.criteriaSatisfied?.[0]?.evidence, "saved commentator verdict");
 		});
 
 		it("inline mode accepts from the text report and strips fences from the resolved file output", async () => {
@@ -405,7 +405,7 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 				tempArtifactsDir: tempDir,
 				getSubagentSessionRoot: () => tempDir,
 				expandTilde: (p: string) => p,
-				discoverAgents: () => ({ agents: [makeAgent("worker", { completionGuard: false })] }),
+				discoverAgents: () => ({ agents: [makeAgent("builder", { completionGuard: false })] }),
 			});
 
 			const acceptance = { level: "checked", criteria: ["Report the findings"] };
@@ -413,8 +413,8 @@ describe("acceptance file reports", { skip: !runSync ? "pi packages not availabl
 				"acceptance-parallel-files",
 				{
 					tasks: [
-						{ agent: "worker", task: "Alpha task: write the report.", output: alphaPath, outputMode: "file-only", acceptance },
-						{ agent: "worker", task: "Beta task: write the report.", output: betaPath, outputMode: "file-only", acceptance },
+						{ agent: "builder", task: "Alpha task: write the report.", output: alphaPath, outputMode: "file-only", acceptance },
+						{ agent: "builder", task: "Beta task: write the report.", output: betaPath, outputMode: "file-only", acceptance },
 					],
 					async: true,
 					clarify: false,