@tea-agent/loop-agent 0.44.0-next.12 → 0.44.0-next.14

package/dist/workflows/dag/frontend-test-case-checklist.js

@@ -142,12 +142,18 @@ function executableListStep(value) {
             return null;
         return normalizeCommand(inlineCode[1]);
     }
-    const prompted = /^[$>]\s*\S/.test(trimmed);
     const normalized = normalizeCommand(trimmed);
     if (isNonExecutableBlockedProse(normalized))
         return null;
-    if (prompted || isCommandLikeExecutable(normalized))
+    // A `$ `/`> ` prompt marker (stripped by normalizeCommand) must NOT turn
+    // Markdown documentation — e.g. a `> 协议说明` blockquote inside 测试步骤 —
+    // into an executable instruction: every line, prompted or not, follows the
+    // same command-like rule. Real commands after the marker (e.g.
+    // `> npx playwright test`) still normalize to command-like executables and
+    // stay gated by the allowlist.
+    if (isCommandLikeExecutable(normalized)) {
         return normalized;
+    }
     return null;
 }
 function hasShellControl(command) {
@@ -330,7 +336,10 @@ export async function validateFrontendCaseChecklist(input) {
         }
         if (expectedPath && casePath.replaceAll("\\", "/") !== expectedPath)
             issues.push({ ruleId: "case-path-mismatch", caseId: id, casePath, detail: `${casePath} must equal ${expectedPath}` });
-        const body = await readFile(absolute, "utf8");
+        // `playwright_cli` (underscore) is the structured tool name; normalize it to
+        // the canonical `playwright-cli` executable spelling before gating so the
+        // same command is accepted regardless of which spelling the case used.
+        const body = (await readFile(absolute, "utf8")).replace(/\bplaywright_cli\b/g, "playwright-cli");
         if (!openRe.test(body))
             issues.push({ ruleId: "open-prefix", caseId: id, casePath, detail: "missing playwright-cli open --browser=chrome <absolute-url>" });
         const match = body.match(/playwright-cli\s+open\s+--browser=chrome\s+(https?:\/\/\S+)/i);

package/dist/workflows/dag/frontend-test-environment-probe.js

@@ -2,8 +2,6 @@ import { spawnSync } from "node:child_process";
 import { mkdir, readFile, writeFile } from "node:fs/promises";
 import path from "node:path";
 import { defaultFrontendTestLayout, frontendTestContextRel, frontendTestProbeRel, } from "./frontend-test-layout.js";
-export const FRONTEND_TEST_CONTEXT_REL = "testcase/frontend/rag/context.md";
-export const FRONTEND_TEST_PROBE_REL = "testcase/frontend/rag/environment-probe.json";
 const CONNECTION_REFUSED_HINT = "被测服务未在 baseUrl 监听（connection refused）。请先启动本地前端（例如 scripts/serve.sh 或项目约定的 npm start），确认可访问后再从 materialize-frontend-test-execution-shell 续跑。";
 export function classifyFrontendBaseUrlProbeFailure(input) {
     if (input.spawnError)

package/dist/workflows/dag/frontend-test-layout.js

@@ -1,4 +1,5 @@
 import path from "node:path";
+import { canonicalTestArtifactRoot, } from "./test-artifact-namespace.js";
 /**
  * Frontend-test artifact layout (configurable testRoot).
  *
@@ -6,14 +7,16 @@ import path from "node:path";
  * frozen on the spec for the whole run. Writers, gates, Playwright fences
  * and TaskSpec allowedPaths all derive from this single object.
  *
- * Default layout keeps the historical frozen contract:
- *   testcase/frontend/rag/**
- *   testcase/frontend/cases/**
- *   testcase/frontend/evidence/**
- *   testcase/frontend/reports/**
- *   testcase/frontend/fixtures/**
+ * Default layout (since 2026-09-17) lives under the shared runtime namespace:
+ *   .harness/testcase/frontend/rag/**
+ *   .harness/testcase/frontend/cases/**
+ *   .harness/testcase/frontend/evidence/**
+ *   .harness/testcase/frontend/reports/**
+ *   .harness/testcase/frontend/fixtures/**
  */
-export const FRONTEND_TEST_DEFAULT_TEST_ROOT = "testcase/frontend";
+export const FRONTEND_TEST_DEFAULT_TEST_ROOT = canonicalTestArtifactRoot("frontend-test");
+/** Historical pre-2026-09-17 default root, kept for text rewrites only. */
+export const FRONTEND_TEST_LEGACY_DEFAULT_TEST_ROOT = "testcase/frontend";
 const FORBIDDEN_TEST_ROOT_PREFIXES = [
     "src",
     "app",
@@ -59,6 +62,10 @@ function normalizeRelPosix(value, label) {
     return segments.join("/");
 }
 function assertTestRootAllowed(testRoot) {
+    // D3: the canonical harness test-artifact root is the ONLY hidden-root
+    // exception. Every other hidden or forbidden top segment stays rejected.
+    if (testRoot === FRONTEND_TEST_DEFAULT_TEST_ROOT)
+        return;
     const topSegment = testRoot.split("/")[0];
     if (FORBIDDEN_TEST_ROOT_PREFIXES.some((forbidden) => topSegment === forbidden || topSegment.startsWith(`${forbidden}.`))) {
         throw new Error(`frontendTest.testRoot must not target production/governance trees: ${testRoot}`);
@@ -67,14 +74,25 @@ function assertTestRootAllowed(testRoot) {
         throw new Error(`frontendTest.testRoot must not target hidden directories: ${testRoot}`);
     }
 }
+function resolveTestRootInput(raw, canonical) {
+    // D3: only the EXACT canonical harness test-artifact root may be a hidden
+    // path. Compare after trimming + stripping trailing slashes so
+    // `.harness/testcase/frontend/` matches, but every other hidden spelling
+    // falls through to the fail-closed normalizer below.
+    const trimmed = raw.trim().replace(/\/+$/, "");
+    if (trimmed === canonical)
+        return canonical;
+    return normalizeRelPosix(raw, "testRoot");
+}
 /**
  * Resolve the frozen frontend-test artifact layout from task config.
- * Throws on unsafe or ambiguous configuration (fail-closed).
+ * Roots of the form `.harness/testcase/frontend` are accepted only as the
+ * exact canonical default; all other hidden roots fail closed (D3).
  */
 export function resolveFrontendTestLayout(config) {
     const testRoot = config?.testRoot === undefined || config?.testRoot === null
         ? FRONTEND_TEST_DEFAULT_TEST_ROOT
-        : normalizeRelPosix(config.testRoot, "testRoot");
+        : resolveTestRootInput(config.testRoot, FRONTEND_TEST_DEFAULT_TEST_ROOT);
     assertTestRootAllowed(testRoot);
     const isDefault = testRoot === FRONTEND_TEST_DEFAULT_TEST_ROOT;
     return {
@@ -98,10 +116,16 @@ export function frontendTestLayoutFromSpec(spec) {
     return resolveFrontendTestLayout({ testRoot: frozen.testRoot });
 }
 const DEFAULT_PREFIX = `${FRONTEND_TEST_DEFAULT_TEST_ROOT}/`;
+const LEGACY_DEFAULT_PREFIX = `${FRONTEND_TEST_LEGACY_DEFAULT_TEST_ROOT}/`;
 /**
- * Longest-prefix rewrite of historical `testcase/frontend/` paths.
- * No-op when layout.isDefault so default DAGs stay byte-identical.
- * Never replaces the short token `testcase/` (would collide with backend-test).
+ * Longest-prefix rewrite of frontend test-root paths in template text.
+ *
+ * No-op when layout.isDefault (default DAGs stay byte-identical to their
+ * templates, which already use the canonical `.harness/testcase/frontend`
+ * root). Custom roots rewrite BOTH the canonical default prefix and the
+ * historical `testcase/frontend/` prefix, so legacy templates and frozen
+ * specs converge onto the configured root. Never replaces the short token
+ * `testcase/` alone (would collide with backend-test).
  */
 export function applyFrontendTestLayoutToText(text, layout) {
     if (layout.isDefault)
@@ -119,6 +143,11 @@ export function applyFrontendTestLayoutToText(text, layout) {
             i += DEFAULT_PREFIX.length;
             continue;
         }
+        if (text.startsWith(LEGACY_DEFAULT_PREFIX, i)) {
+            output += replacement;
+            i += LEGACY_DEFAULT_PREFIX.length;
+            continue;
+        }
         output += text[i];
         i += 1;
     }

package/dist/workflows/dag/hybrid/sources.js

@@ -7,6 +7,7 @@ import { DAG_AGENT_RUNTIME_PI_ONLY, DAG_REPAIR_WRITER_PROTOCOL_EXPLICIT_NODE_V1,
 import { planMavenVerification } from "../../../verification/maven/index.js";
 import { pathMatchesPattern } from "../../../shared/git-progress.js";
 import { BASELINE_FORBIDDEN_PATHS } from "../governance-constants.js";
+import { isAnyTestArtifactPath } from "../test-artifact-namespace.js";
 import { resolveAdapter } from "../../../adapters/index.js";
 import { loadHarnessManifest } from "../../../governance/harness.js";
 import { mergeDocumentIndexCompanions } from "../../../governance/document-index-closure.js";
@@ -31,16 +32,16 @@ export const HYBRID_DEFAULTS = {
     executor: "pi",
     piBackend: "sdk-first",
     contextProfile: "slim",
-    skills: ["ai-engineering-context"],
+    skills: ["builtin:ai-engineering-context"],
     writePolicy: "read-only",
 };
 export const HYBRID_SKILLS_BY_ROLE = {
-    planner: ["loop-agent"],
-    scout: ["ai-engineering-context", "codebase-scout"],
-    implementer: ["test-driven-development", "verification-before-completion"],
-    reviewer: ["requesting-code-review", "code-review-core"],
-    verifier: ["verification-before-completion", "systematic-debugging"],
-    closeout: ["loop-agent", "verification-before-completion"],
+    planner: ["builtin:loop-agent"],
+    scout: ["builtin:ai-engineering-context", "builtin:codebase-scout"],
+    implementer: ["builtin:test-driven-development", "builtin:verification-before-completion"],
+    reviewer: ["builtin:requesting-code-review", "builtin:code-review-core"],
+    verifier: ["builtin:verification-before-completion", "builtin:systematic-debugging"],
+    closeout: ["builtin:loop-agent", "builtin:verification-before-completion"],
 };
 export const FRONTEND_DEFAULTS = {
     ...HYBRID_DEFAULTS,
@@ -55,13 +56,13 @@ export const FRONTEND_SKILLS_BY_ROLE = {
     verifier: [],
     closeout: [],
 };
-export const FRONTEND_CONTRACT_SKILLS = ["frontend-contract"];
-export const FRONTEND_SCOUT_SKILLS = ["frontend-scout"];
-export const FRONTEND_PLAN_SKILLS = ["frontend-plan"];
-export const FRONTEND_BOUNDED_IMPLEMENT_SKILLS = ["frontend-bounded-implement"];
-export const FRONTEND_DESIGN_REVIEW_SKILLS = ["frontend-design-review"];
-export const FRONTEND_REVIEW_SKILLS = ["frontend-review"];
-export const FRONTEND_VERIFICATION_SKILLS = ["frontend-verification"];
+export const FRONTEND_CONTRACT_SKILLS = ["builtin:frontend-contract"];
+export const FRONTEND_SCOUT_SKILLS = ["builtin:frontend-scout"];
+export const FRONTEND_PLAN_SKILLS = ["builtin:frontend-plan"];
+export const FRONTEND_BOUNDED_IMPLEMENT_SKILLS = ["builtin:frontend-bounded-implement"];
+export const FRONTEND_DESIGN_REVIEW_SKILLS = ["builtin:frontend-design-review"];
+export const FRONTEND_REVIEW_SKILLS = ["builtin:frontend-review"];
+export const FRONTEND_VERIFICATION_SKILLS = ["builtin:frontend-verification"];
 export const IMPLEMENT_WRITESET_PLACEHOLDER = "REPLACE/WITH/NARROW/IMPLEMENT/PATHS/**";
 export const GENERATED_DAG_RUNTIME_CONTRACT = {
     schemaVersion: DAG_RUNTIME_CONTRACT_SCHEMA_VERSION,
@@ -1534,20 +1535,38 @@ export function patternStaticPrefix(pattern) {
     const wildcard = normalized.search(/[?*]/);
     return (wildcard >= 0 ? normalized.slice(0, wildcard) : normalized).replace(/\/+$/, "");
 }
-export function allowedPathHitsBaselineForbidden(allowedPath) {
+export function allowedPathHitsBaselineForbidden(allowedPath, options) {
     const probes = probePathsForWriteSetEntry(allowedPath);
     for (const forbidden of BASELINE_FORBIDDEN_PATHS) {
         if (probes.some((probe) => pathMatchesPattern(probe, forbidden)) ||
             pathMatchesPattern(allowedPath, forbidden) ||
             pathMatchesPattern(forbidden, allowedPath)) {
+            // D3 test-artifact exception: dedicated test tasks may own exactly the
+            // canonical `.harness/testcase` roots in task allowedPaths; every other
+            // `.harness` overlap stays a baseline violation.
+            if (options?.testArtifactNamespace &&
+                isHarnessNamespacePattern(forbidden) &&
+                probes.every((probe) => isAnyTestArtifactPath(probe))) {
+                continue;
+            }
             return forbidden;
         }
     }
     return undefined;
 }
+function isHarnessNamespacePattern(pattern) {
+    const normalized = pattern.replace(/\\/g, "/").replace(/^\.\//, "");
+    return (normalized === ".harness" ||
+        normalized === ".harness/**" ||
+        normalized.startsWith(".harness/"));
+}
 export function assertTaskAllowedPathsPreflight(taskConfig) {
+    const testArtifactNamespace = taskConfig.taskKind === "backend-test" ||
+        taskConfig.taskKind === "frontend-test";
     for (const allowedPath of taskConfig.allowedPaths) {
-        const forbidden = allowedPathHitsBaselineForbidden(allowedPath);
+        const forbidden = allowedPathHitsBaselineForbidden(allowedPath, {
+            testArtifactNamespace,
+        });
         if (forbidden) {
             throw new Error(`task allowedPaths entry "${allowedPath}" overlaps baseline forbidden path "${forbidden}"; task source paths are main-session owned and must not be executor writeSet`);
         }