@tea-agent/loop-agent 0.24.4 → 0.24.6

package/dist/worker/observe/static/views/dag-inspector.js

@@ -10,12 +10,13 @@ import {
 	UI_TEXT,
 } from "../constants.js";
 import { clearNode, el } from "../dom.js";
-import { fetchJson, fetchSpecEvidenceFile } from "../api.js";
+import { fetchJson, fetchSpecEvidenceFile, fetchDagNodeInput } from "../api.js";
 import { badge } from "../format.js";
 import { renderMarkdown } from "../markdown-render.js";
 import {
 	uiState,
 	makeSessionEventIdentity,
+	makeDagNodeInputIdentity,
 	openSpecEvidenceDetail,
 	closeSpecEvidenceDetail,
 	updateDagTimelineViewportState,
@@ -50,9 +51,13 @@ export function selectDagNode(dagRunId, nodeId) {
 		// into node B's spec-evidence panel (AC-007 regression guard).
 		uiState.specEvidenceDetail = null;
 		uiState.specEvidenceListEvidence = null;
+		uiState.dagNodeInputData = null;
+		uiState.dagNodeInputError = null;
+		uiState.dagNodeInputLoading = false;
 		// Bump the session-event identity token so any in-flight response for the
 		// previous node is rejected before it can mutate the current node cache.
 		uiState.sessionEventIdentity = makeSessionEventIdentity(dagRunId, nodeId);
+		uiState.dagNodeInputIdentity = makeDagNodeInputIdentity(dagRunId, nodeId);
 	}
 	void import("./dag.js").then((m) => m.renderDagDetail(dagRunId, false));
 }
@@ -530,16 +535,337 @@ function renderSpecEvidenceDetail(content, dagRunId, nodeId) {
 	content.appendChild(wrapper);
 }
 
+function sourceLabel(source) {
+	if (source === "defaults") return "继承默认值";
+	if (source === "schema-default") return "schema 默认";
+	if (source === "merged") return "合并";
+	return null;
+}
+
+function appendMetaRow(dl, label, value, source) {
+	if (value == null || value === "") return;
+	const dt = el("dt", null, label);
+	const dd = el("dd", null, String(value));
+	const inherited = sourceLabel(source);
+	if (inherited) {
+		dd.appendChild(document.createTextNode(" "));
+		dd.appendChild(el("span", "node-input-inherited", inherited));
+	}
+	dl.appendChild(dt);
+	dl.appendChild(dd);
+}
+
+function appendCodeList(parent, title, values) {
+	if (!values || values.length === 0) return;
+	const section = el("section", "node-input-section");
+	section.appendChild(el("h4", null, title));
+	const list = el("ul", "node-input-code-list");
+	for (const value of values) {
+		const item = document.createElement("li");
+		const code = document.createElement("code");
+		code.textContent = String(value);
+		item.appendChild(code);
+		list.appendChild(item);
+	}
+	section.appendChild(list);
+	parent.appendChild(section);
+}
+
+function renderDagNodeInputDetail(content, data) {
+	const wrap = el("div", "node-input-root");
+	wrap.appendChild(
+		el("p", "node-input-semantics", UI_TEXT.inputFrozenLabel),
+	);
+
+	if (data.summary) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "摘要"));
+		const dl = el("dl", "node-input-summary");
+		appendMetaRow(
+			dl,
+			"executor",
+			data.summary.executor,
+			data.summary.executorSource,
+		);
+		appendMetaRow(dl, "role", data.summary.role, null);
+		appendMetaRow(dl, "complexity", data.summary.complexity, null);
+		appendMetaRow(
+			dl,
+			"writePolicy",
+			data.summary.writePolicy,
+			data.summary.writePolicySource,
+		);
+		appendMetaRow(dl, "toolProfile", data.summary.toolProfile, null);
+		section.appendChild(dl);
+		wrap.appendChild(section);
+	}
+
+	if (data.dependency) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "依赖"));
+		const dl = el("dl", "node-input-summary");
+		appendMetaRow(
+			dl,
+			"depends_on",
+			(data.dependency.dependsOn ?? []).join(", ") || "(none)",
+			null,
+		);
+		if (data.dependency.dependsPolicy) {
+			appendMetaRow(dl, "dependsPolicy", data.dependency.dependsPolicy, null);
+		}
+		if ((data.dependency.failureAwareDependsOn ?? []).length > 0) {
+			appendMetaRow(
+				dl,
+				"failureAwareDependsOn",
+				data.dependency.failureAwareDependsOn.join(", "),
+				null,
+			);
+		}
+		if (data.dependency.runIf) {
+			appendMetaRow(dl, "runIf", data.dependency.runIf, null);
+		}
+		section.appendChild(dl);
+		wrap.appendChild(section);
+	}
+
+	if (data.prompt) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "任务正文"));
+		if (data.prompt.source?.path) {
+			const meta = el("p", "muted");
+			meta.textContent = `source: ${data.prompt.source.path}${
+				data.prompt.source.sha256Prefix
+					? ` (sha256 ${data.prompt.source.sha256Prefix}…)`
+					: ""
+			}`;
+			section.appendChild(meta);
+		}
+		section.appendChild(renderMarkdown(data.prompt.text ?? ""));
+		if (data.prompt.truncated) {
+			section.appendChild(
+				el(
+					"p",
+					"node-input-truncated",
+					`已截断，仅显示前 ${data.prompt.maxBytes ?? 0} 字节（UTF-8）。`,
+				),
+			);
+		}
+		wrap.appendChild(section);
+	}
+
+	if (data.executorInput) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "执行器输入摘要"));
+		const dl = el("dl", "node-input-summary");
+		appendMetaRow(dl, "kind", data.executorInput.kind, null);
+		if (data.executorInput.modelHint) {
+			appendMetaRow(dl, "modelHint", data.executorInput.modelHint, null);
+		}
+		section.appendChild(dl);
+		appendCodeList(section, "skills", data.executorInput.skills);
+		if (data.executorInput.shell) {
+			const shell = data.executorInput.shell;
+			const shellDl = el("dl", "node-input-summary");
+			appendMetaRow(shellDl, "preset", shell.preset, null);
+			appendMetaRow(shellDl, "cwd", shell.cwd, null);
+			appendMetaRow(shellDl, "timeoutMs", shell.timeoutMs, null);
+			appendMetaRow(
+				shellDl,
+				"nonZeroExitPolicy",
+				shell.nonZeroExitPolicy,
+				null,
+			);
+			appendMetaRow(
+				shellDl,
+				"gates",
+				(shell.gates ?? []).join(", ") || "(none)",
+				null,
+			);
+			section.appendChild(shellDl);
+			if ((shell.commands ?? []).length > 0) {
+				const list = el("ol", "node-input-command-list");
+				for (const command of shell.commands) {
+					const item = document.createElement("li");
+					const pre = document.createElement("pre");
+					const code = document.createElement("code");
+					code.textContent = String(command);
+					pre.appendChild(code);
+					item.appendChild(pre);
+					list.appendChild(item);
+				}
+				section.appendChild(list);
+			}
+		}
+		if (data.executorInput.static) {
+			const staticBlock = data.executorInput.static;
+			const staticDl = el("dl", "node-input-summary");
+			appendMetaRow(staticDl, "status", staticBlock.status, null);
+			section.appendChild(staticDl);
+			section.appendChild(renderMarkdown(staticBlock.resultPreview ?? ""));
+			if (staticBlock.truncated) {
+				section.appendChild(
+					el(
+						"p",
+						"node-input-truncated",
+						`已截断，仅显示前 ${staticBlock.maxBytes ?? 0} 字节（UTF-8）。`,
+					),
+				);
+			}
+		}
+		wrap.appendChild(section);
+	}
+
+	if (data.boundaries) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "边界与输出契约"));
+		const dl = el("dl", "node-input-summary");
+		appendMetaRow(dl, "outputMode", data.boundaries.outputMode, null);
+		appendMetaRow(
+			dl,
+			"firstProtocolLine",
+			data.boundaries.firstProtocolLine,
+			null,
+		);
+		section.appendChild(dl);
+		if (data.boundaries.outputContract) {
+			section.appendChild(el("h5", null, "outputContract"));
+			section.appendChild(renderMarkdown(data.boundaries.outputContract));
+		}
+		appendCodeList(section, "allowedPaths", data.boundaries.allowedPaths);
+		appendCodeList(section, "forbiddenPaths", data.boundaries.forbiddenPaths);
+		appendCodeList(section, "writeSet", data.boundaries.writeSet);
+		wrap.appendChild(section);
+	}
+
+	if (data.promptFingerprint?.exists) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "Assembled prompt 指纹"));
+		const dl = el("dl", "node-input-summary");
+		appendMetaRow(
+			dl,
+			"sha256",
+			data.promptFingerprint.sha256Prefix ?? "(parse failed)",
+			null,
+		);
+		appendMetaRow(
+			dl,
+			"lengthChars",
+			data.promptFingerprint.lengthChars,
+			null,
+		);
+		appendMetaRow(dl, "artifact", data.promptFingerprint.artifact, null);
+		section.appendChild(dl);
+		section.appendChild(
+			el("p", "muted", UI_TEXT.inputFingerprintNote),
+		);
+		wrap.appendChild(section);
+	}
+
+	if ((data.warnings ?? []).length > 0) {
+		const section = el("section", "node-input-section");
+		section.appendChild(el("h4", null, "警告"));
+		for (const warning of data.warnings) {
+			section.appendChild(el("p", "node-input-warning", String(warning)));
+		}
+		wrap.appendChild(section);
+	}
+
+	content.appendChild(wrap);
+}
+
+function unavailableMessage(reason) {
+	if (reason === "unsupported-dynamic-node") {
+		return UI_TEXT.inputDynamicUnsupported;
+	}
+	if (reason === "run-spec-unavailable") {
+		return UI_TEXT.inputRunSpecUnavailable;
+	}
+	if (reason === "task-not-found") {
+		return UI_TEXT.inputUnavailable;
+	}
+	return UI_TEXT.noInput;
+}
+
+async function renderDagNodeInput(content, dagRunId, node) {
+	const nodeId = node.nodeId;
+	const identity = makeDagNodeInputIdentity(dagRunId, nodeId);
+	uiState.dagNodeInputIdentity = identity;
+	uiState.dagNodeInputLoading = true;
+	uiState.dagNodeInputError = null;
+
+	clearNode(content);
+	const loading = el("p", "muted", UI_TEXT.inputLoading);
+	loading.setAttribute("role", "status");
+	loading.setAttribute("aria-live", "polite");
+	content.appendChild(loading);
+
+	const result = await fetchDagNodeInput(dagRunId, nodeId);
+	if (
+		uiState.dagNodeInputIdentity !== identity ||
+		uiState.selectedDagNodeId !== nodeId ||
+		uiState.currentDagRunId !== dagRunId ||
+		uiState.dagInspectorTab !== "input" ||
+		!content.isConnected ||
+		content.dataset.dagNodeId !== String(nodeId) ||
+		content.dataset.dagRunId !== String(dagRunId)
+	) {
+		return;
+	}
+
+	uiState.dagNodeInputLoading = false;
+	clearNode(content);
+
+	if (!result.ok) {
+		const message =
+			result.body?.error ||
+			(result.status === 0
+				? UI_TEXT.inputError
+				: `${UI_TEXT.inputError}（HTTP ${result.status}）`);
+		uiState.dagNodeInputError = message;
+		uiState.dagNodeInputData = null;
+		const error = el("p", "node-input-error", message);
+		error.setAttribute("role", "alert");
+		content.appendChild(error);
+		return;
+	}
+
+	const data = result.body;
+	uiState.dagNodeInputData = data;
+	uiState.dagNodeInputError = null;
+	if (!data?.available) {
+		const empty = el(
+			"p",
+			"empty",
+			unavailableMessage(data?.availabilityReason),
+		);
+		empty.setAttribute("role", "status");
+		content.appendChild(empty);
+		if ((data?.warnings ?? []).length > 0) {
+			for (const warning of data.warnings) {
+				content.appendChild(el("p", "node-input-warning", String(warning)));
+			}
+		}
+		return;
+	}
+
+	renderDagNodeInputDetail(content, data);
+}
+
 	function fillDagInspectorContent(content, dagRunId, node) {
 		const previousRenderedNode = content.dataset.dagNodeId ?? "";
-		const nodeChanged = previousRenderedNode !== String(node.nodeId ?? "");
+		const previousRenderedRun = content.dataset.dagRunId ?? "";
+		const nodeChanged =
+			previousRenderedNode !== String(node.nodeId ?? "") ||
+			previousRenderedRun !== String(dagRunId ?? "");
 		if (nodeChanged) {
 			// Drop any open spec-evidence preview + cached list so a stale A detail
 			// cannot survive into node B (AC-007 regression guard).
 			uiState.specEvidenceDetail = null;
 			uiState.specEvidenceListEvidence = null;
+			uiState.dagNodeInputData = null;
+			uiState.dagNodeInputError = null;
 		}
-		const activeTab = ["timeline", "spec-evidence"].includes(
+		const activeTab = ["input", "timeline", "spec-evidence"].includes(
 			uiState.dagInspectorTab,
 		)
 			? uiState.dagInspectorTab
@@ -548,12 +874,15 @@ function renderSpecEvidenceDetail(content, dagRunId, nodeId) {
 		content.setAttribute("role", "tabpanel");
 		content.setAttribute("aria-labelledby", `dag-inspector-tab-${activeTab}`);
 		content.dataset.dagNodeId = String(node.nodeId ?? "");
+		content.dataset.dagRunId = String(dagRunId ?? "");
 		// Drop timeline-only classes when switching tabs so output layout stays clean.
 		content.className = "dag-inspector-content";
 		if (activeTab === "spec-evidence") {
 			void renderSpecEvidence(content, dagRunId, node.nodeId);
 		} else if (activeTab === "timeline") {
 			renderSessionTimeline(content, node.nodeId);
+		} else if (activeTab === "input") {
+			void renderDagNodeInput(content, dagRunId, node);
 		} else if (!node.outputPreview && !node.errorPreview) {
 			content.appendChild(el("p", "empty", UI_TEXT.noOutput));
 		} else {
@@ -592,6 +921,7 @@ function buildDagInspectorHeader(dagRunId, node) {
 	header.appendChild(controls);
 
 	const tabDefs = [
+		["input", "节点输入", "ri-login-box-line"],
 		["output", "节点输出", "ri-file-text-line"],
 		["timeline", "执行过程", "ri-route-line"],
 		["spec-evidence", "规范证据", "ri-book-open-line"],
@@ -660,6 +990,17 @@ function syncDagInspectorHeader(panel, dagRunId, node) {
 }
 
 function dagInspectorSignature(dagRunId, node) {
+	const inputRevision =
+		uiState.dagInspectorTab === "input"
+			? [
+					uiState.dagNodeInputIdentity,
+					uiState.dagNodeInputLoading,
+					uiState.dagNodeInputError ?? "",
+					uiState.dagNodeInputData?.availabilityReason ?? "",
+					uiState.dagNodeInputData?.promptFingerprint?.sha256Prefix ?? "",
+					node.status ?? "",
+				]
+			: null;
 	return JSON.stringify([
 		dagRunId,
 		node.nodeId,
@@ -667,6 +1008,7 @@ function dagInspectorSignature(dagRunId, node) {
 		node.outputPreview ?? "",
 		node.errorPreview ?? "",
 		uiState.dagInspectorTab === "timeline" ? uiState.sessionEventOffset : null,
+		inputRevision,
 	]);
 }
 
@@ -702,17 +1044,30 @@ export function renderDagInspector(dagRunId, node, initial = true) {
 		syncDagInspectorHeader(panel, dagRunId, node);
 		if (panel.dataset.contentSignature === nextContentSignature) return;
 		const previousNode = existingContent.dataset.dagNodeId ?? "";
-		const nodeChanged = previousNode !== String(node.nodeId ?? "");
-		const followLatest =
-			nodeChanged ||
-			computeFollowLatest(
-				existingContent.scrollTop,
-				existingContent.scrollHeight,
-				existingContent.clientHeight,
-			);
-		const regionKey = `dag:${dagRunId}:${node.nodeId}:output`;
-		uiState.dagNodeOutputViewportState = updateDagOutputViewportState(
-			uiState.dagNodeOutputViewportState,
+		const previousRun = existingContent.dataset.dagRunId ?? "";
+		const nodeChanged =
+			previousNode !== String(node.nodeId ?? "") ||
+			previousRun !== String(dagRunId ?? "");
+		const activeTab = ["input", "timeline", "spec-evidence"].includes(
+			uiState.dagInspectorTab,
+		)
+			? uiState.dagInspectorTab
+			: "output";
+		const isInputTab = activeTab === "input";
+		const followLatest = isInputTab
+			? false
+			: nodeChanged ||
+				computeFollowLatest(
+					existingContent.scrollTop,
+					existingContent.scrollHeight,
+					existingContent.clientHeight,
+				);
+		const regionKey = `dag:${dagRunId}:${node.nodeId}:${isInputTab ? "input" : "output"}`;
+		const viewportUpdater = isInputTab
+			? "dagNodeInputViewportState"
+			: "dagNodeOutputViewportState";
+		uiState[viewportUpdater] = updateDagOutputViewportState(
+			uiState[viewportUpdater],
 			regionKey,
 			existingContent.scrollLeft,
 			existingContent.scrollTop,
@@ -729,7 +1084,8 @@ export function renderDagInspector(dagRunId, node, initial = true) {
 		if (followLatest) {
 			existingContent.scrollTop = existingContent.scrollHeight;
 		} else {
-			existingContent.scrollTop = uiState.dagNodeOutputViewportState.scrollTop;
+			const saved = uiState[viewportUpdater];
+			existingContent.scrollTop = nodeChanged ? 0 : (saved?.scrollTop ?? 0);
 		}
 		for (const [index, pre] of [
 			...existingContent.querySelectorAll("pre"),

package/dist/workflows/dag/backend-test-markdown-workflow.js

@@ -413,6 +413,18 @@ export async function validateBackendMarkdownCases(input) {
             if (!hasAssertableExpectedResult(expected)) {
                 findings.push(`${testCase.id} has no structured expected result`);
             }
+            const expectedScript = expectedBackendTestPytestScriptForMarkdownModule(relativeFile);
+            const mappedScripts = extractMappedPytestScripts(testCase.body);
+            if (mappedScripts.length === 0) {
+                findings.push(`${testCase.id} has no mapped pytest script in Automation Notes/自动化映射 (expected ${expectedScript} from module ${path.basename(relativeFile)})`);
+            }
+            else {
+                for (const script of mappedScripts) {
+                    if (script !== expectedScript) {
+                        findings.push(`${testCase.id} automation mapping ${script} must equal module one-to-one path ${expectedScript} (from ${path.basename(relativeFile)})`);
+                    }
+                }
+            }
         }
     }
     const missing = input.requiredRequirementIds.filter((id) => !coveredAc.has(id));
@@ -454,6 +466,27 @@ export async function validateBackendMarkdownCases(input) {
 function collectMarkdownCaseIds(markdown) {
     return splitCases(markdown).map((testCase) => testCase.id);
 }
+/**
+ * Derive the stable module stem used for Markdown ↔ pytest one-to-one mapping.
+ * `testcase/md/<module>.md` → stem → `testcase/test_<stem>.py`.
+ * Example: `BE-HEALTH.md` / `be-health.md` / `order-api.md` → `be_health` / `order_api`.
+ */
+export function normalizeBackendTestModuleStem(moduleFileName) {
+    const base = path.basename(moduleFileName).replace(/\.md$/i, "");
+    const stem = base
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, "_")
+        .replace(/^_+|_+$/g, "")
+        .replace(/_+/g, "_");
+    if (!stem) {
+        throw new Error(`cannot derive backend-test module stem from: ${moduleFileName}`);
+    }
+    return stem;
+}
+/** Expected pytest script path for a Markdown module file (basename or relative path). */
+export function expectedBackendTestPytestScriptForMarkdownModule(moduleFileName) {
+    return `testcase/test_${normalizeBackendTestModuleStem(moduleFileName)}.py`;
+}
 function extractMappedPytestScripts(testCaseBody) {
     const automation = sectionBody(testCaseBody, CASE_SECTION_ALIASES.automationNotes);
     if (!automation.trim())
@@ -462,6 +495,15 @@ function extractMappedPytestScripts(testCaseBody) {
         .map((match) => match[1].replaceAll("\\", "/"))
         .filter((value) => !value.includes("..")));
 }
+function isSafeBackendPytestScript(script, workspaceRoot) {
+    const absolute = path.resolve(workspaceRoot, script);
+    const relative = path.relative(workspaceRoot, absolute).replaceAll(path.sep, "/");
+    return (!relative.startsWith("..") &&
+        !path.isAbsolute(relative) &&
+        script.startsWith("testcase/") &&
+        !script.includes("..") &&
+        /^test_.*\.py$/i.test(path.basename(script)));
+}
 function testFunctionRegion(input) {
     const functionLineStart = input.source.lastIndexOf("\n", input.functionIndex - 1) + 1;
     const functionLine = input.source.slice(functionLineStart, input.functionHeaderEnd);
@@ -610,26 +652,50 @@ export async function collectBackendTestMappedPytestScripts(workspaceRoot) {
     for (const file of files) {
         if (path.basename(file).toLowerCase() === "readme.md")
             continue;
+        const relativeFile = path.relative(workspaceRoot, file).replaceAll(path.sep, "/");
+        const expectedScript = expectedBackendTestPytestScriptForMarkdownModule(relativeFile);
         const markdown = await readFile(file, "utf8");
+        const mappedFromModule = new Set();
         for (const testCase of splitCases(markdown)) {
-            for (const script of extractMappedPytestScripts(testCase.body))
+            for (const script of extractMappedPytestScripts(testCase.body)) {
+                mappedFromModule.add(script);
+            }
+        }
+        // Prefer explicit mappings, but recover to the deterministic module stem path
+        // when the model wrote the correct one-to-one file while Markdown still names
+        // a drifted script (common: health.md maps test_health.py but writer emitted
+        // test_be_health.py for BE-HEALTH.md, or the reverse).
+        if (mappedFromModule.size === 0) {
+            if (await exists(path.resolve(workspaceRoot, expectedScript))) {
+                scripts.add(expectedScript);
+            }
+            continue;
+        }
+        for (const script of mappedFromModule) {
+            if (!isSafeBackendPytestScript(script, workspaceRoot)) {
+                throw new Error(`unsafe mapped pytest script: ${script}`);
+            }
+            if (await exists(path.resolve(workspaceRoot, script))) {
                 scripts.add(script);
+                continue;
+            }
+            if (isSafeBackendPytestScript(expectedScript, workspaceRoot) &&
+                (await exists(path.resolve(workspaceRoot, expectedScript)))) {
+                scripts.add(expectedScript);
+                continue;
+            }
+            throw new Error(`mapped pytest script is missing: ${script} (module one-to-one path ${expectedScript} also missing)`);
         }
     }
     if (scripts.size === 0) {
         throw new Error("no pytest scripts are mapped by final Markdown cases");
     }
     const safeScripts = [];
-    for (const script of unique(scripts)) {
-        const absolute = path.resolve(workspaceRoot, script);
-        const relative = path.relative(workspaceRoot, absolute);
-        if (relative.startsWith("..") ||
-            path.isAbsolute(relative) ||
-            !script.startsWith("testcase/") ||
-            !/^test_.*\.py$/i.test(path.basename(script))) {
+    for (const script of unique([...scripts]).sort()) {
+        if (!isSafeBackendPytestScript(script, workspaceRoot)) {
             throw new Error(`unsafe mapped pytest script: ${script}`);
         }
-        if (!(await exists(absolute))) {
+        if (!(await exists(path.resolve(workspaceRoot, script)))) {
             throw new Error(`mapped pytest script is missing: ${script}`);
         }
         safeScripts.push(script);