@guanlin4924/devflow-pi 0.1.19 → 0.1.21

package/plugins/devflow-quality/skills/workflow-generate-mainflow-cases/scripts/compile_dsl.py

@@ -531,14 +531,25 @@ def compile_case(case_desc: dict[str, Any]) -> dict[str, Any]:
     }
 
 
-def compile_all(payload: dict[str, Any]) -> dict[str, Any]:
+def compile_all(payload: dict[str, Any], frozen_mainflow_tc_ids: set[str] | None = None) -> dict[str, Any]:
     """Compile all cases in the payload."""
     metadata = dict(payload.get("metadata") or {})
-    cases = [compile_case(c) for c in payload.get("cases", [])]
+    descriptions = list(payload.get("cases", []))
+    tc_ids: list[str] = []
+    if frozen_mainflow_tc_ids is not None:
+        for index, description in enumerate(descriptions):
+            tc_id = description.get("tc_id") if isinstance(description, dict) else None
+            if not isinstance(tc_id, str) or tc_id not in frozen_mainflow_tc_ids:
+                raise ValueError(f"cases[{index}].tc_id must reference a frozen api/page TC that is positive")
+            tc_ids.append(tc_id)
+        if len(tc_ids) != len(set(tc_ids)):
+            raise ValueError("each frozen positive api/page TC may generate at most one mainflow case")
+    cases = [compile_case(c) for c in descriptions]
 
     return {
         "metadata": {
             **metadata,
+            **({"frozen_tc_ids": tc_ids} if frozen_mainflow_tc_ids is not None else {}),
             "compiled_at": datetime.now(timezone.utc).isoformat(),
             "compiled_by": "compile_dsl.py (hard-constraint layer)",
         },
@@ -546,6 +557,41 @@ def compile_all(payload: dict[str, Any]) -> dict[str, Any]:
     }
 
 
+def frozen_mainflow_tc_ids_from_cwd(cwd: Path) -> set[str] | None:
+    """Return the frozen API/page suite for new workflows, else preserve legacy compilation."""
+    state_path = cwd / ".planning" / "STATE.md"
+    if not state_path.is_file():
+        return None
+    match = re.search(r"^current_feature:\s*['\"]?([^'\"\n]+)", state_path.read_text(encoding="utf-8"), re.MULTILINE)
+    if not match:
+        return None
+    suite_path = cwd / ".planning" / "requirements" / match.group(1).strip() / "tests.json"
+    if not suite_path.is_file():
+        return None
+    suite = json.loads(suite_path.read_text(encoding="utf-8"))
+    if suite.get("frozen") is not True:
+        raise ValueError("current TC suite exists but is not frozen and fresh")
+    eligible = {
+        node.get("id")
+        for node in suite.get("nodes", [])
+        if isinstance(node, dict)
+        and node.get("kind") == "test_case"
+        and node.get("status") == "confirmed"
+        and isinstance(node.get("data"), dict)
+        and node["data"].get("test_type") in {"api", "page"}
+        # A generated API TC with an error-code oracle is a negative/exception
+        # test. It belongs to run-acceptance-tests, never to the positive
+        # mainflow importer even though its broad test_type is "api".
+        and not (
+            isinstance(node["data"].get("generated_case_key"), str)
+            and isinstance(node["data"].get("execution"), dict)
+            and isinstance(node["data"]["execution"].get("expect"), dict)
+            and "error_code" in node["data"]["execution"]["expect"]
+        )
+    }
+    return {value for value in eligible if isinstance(value, str)}
+
+
 # ---------------------------------------------------------------------------
 # CLI
 # ---------------------------------------------------------------------------
@@ -568,7 +614,7 @@ def main() -> int:
         raw = {"metadata": {}, "cases": raw}
 
     try:
-        result = compile_all(raw)
+        result = compile_all(raw, frozen_mainflow_tc_ids_from_cwd(Path.cwd()))
     except ValueError as exc:
         print(f"COMPILE ERROR: {exc}", file=__import__("sys").stderr)
         return 1

package/plugins/devflow-quality/skills/workflow-run-acceptance-tests/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: workflow-run-acceptance-tests
+description: "Execute the frozen TC acceptance suite after code review. Use the requirement baseline and frozen TC nodes for test intent; use current code only to bind execution adapters."
+---
+
+# 冻结验收测试执行
+
+此步骤在代码审查后、UAT 前运行。它的目标是执行冻结的 `TC-*`，而不是根据交付总结重新猜测要测什么。
+
+## 输入边界
+
+测试意图只能读取：
+
+- `.planning/requirements/{feature}/tests.json` 中已冻结且未 stale 的 `TC-*`；
+- 图谱中被 TC 引用的已确认 REQ / BR / BC / AC、契约与架构风险；
+- 初始化发现的测试工具链能力。
+
+可以读取当前源码和代码索引，但**仅**用来绑定入口、选择器、请求地址、测试数据与命令适配器。
+
+禁止读取或使用以下内容来扩大、缩小或补造验收范围：
+
+- `.planning/phases/{phase}/SUMMARY.md`；
+- agent 的交付自述、代码 diff 或“已完成”结论；
+- 未冻结的临时需求列表。
+
+## 过程
+
+1. 用 `devflow_acceptance_tests(feature)` 检查测试集已由 CP-3 冻结。若有 stale 或候选 TC，停止并回到需求/测试设计。
+2. 读取冻结 TC 与其 oracle、前置数据、身份和自动化模式。
+   对 `api_refs`、`workload_refs` 和 `risk_refs` 指向的专项用例，保留其契约、性能/迁移或韧性义务；没有适配器或环境时必须写为 blocked，不能用普通主流程替代。
+3. 读取代码审查报告、代码索引和实际源码，仅为每个 TC 绑定运行适配器；不得改变 TC 的断言、范围或预期。
+4. 写入 `.planning/features/{feature}/acceptance-test-bindings.json`：
+   - 固定 `test_suite.graph_revision` 与 `fingerprint`；
+   - 每个 TC 恰好一条绑定；
+   - `mode` 必须与冻结 TC 的 `automation_mode` 一致；
+   - 通用自动化用例使用 `{ "kind": "command", "argv": [ ... ] }`，不使用 shell 字符串；
+   - Core 生成的 API TC 可直接使用 `{ "kind": "http", "base_url_env": "DEVFLOW_TEST_BASE_URL", "headers_env": { "Authorization": "DEVFLOW_TEST_AUTHORIZATION" } }`。请求、HTTP 状态和异常码断言来自冻结 TC 的 `execution`，绑定只能提供环境地址和鉴权，不得重写测试意图；若场景需要专用夹具，可额外配置 `{ "setup": { "argv": [ ... ] } }`，Core 会先记录并执行该受控命令，再发 HTTP 请求；
+   - 人工或环境阻塞用例必须有 owner 与原因/阻塞源。
+5. 调用 `devflow_run_acceptance_tests(action="validate", feature=...)`。不通过时修正绑定，不得绕过。
+6. 调用 `devflow_run_acceptance_tests(action="run", feature=...)`。该 Core 工具记录每个 TC 的命令哈希、退出码和输出摘要。
+7. 失败、未执行或 blocked 的 TC 只能报告为 `blocked`/`failed`，不得在 CP-4 或 UAT 中宣称验收通过。
+
+## 输出
+
+- `.planning/features/{feature}/acceptance-test-bindings.json`
+- `.planning/test-runs/{feature}/acceptance-results.json`
+
+`acceptance-results.json` 是 Core-owned 的执行事实；Markdown 结论不能替代它。

package/plugins/devflow-quality/skills/workflow-run-acceptance-tests/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Run Frozen Acceptance Tests"
+  short_description: "执行冻结 TC 验收测试"
+  default_prompt: "基于冻结的 TC-* 绑定并执行验收测试；源码只能用于运行适配，不能改变测试意图。"

package/plugins/devflow-quality/skills/workflow-verify-work/SKILL.md

@@ -72,18 +72,24 @@ Verification must include these gates when project artifacts exist:
    - Run or require the commands declared by `.planning/toolchain.md`, `.planning/execution-contract.md`, `.context/workflow/verification.md`, or the change list.
    - Include target tests, full tests, build, codegen/mock/DI checks, migration checks, and integration tests when applicable.
 
-5. **Compatibility gate**
+5. **冻结验收测试门禁**
+   - 读取 `.planning/requirements/{feature}/tests.json`、`.planning/features/{feature}/acceptance-test-bindings.json` 和 `.planning/test-runs/{feature}/acceptance-results.json`。
+   - 调用 `devflow_run_acceptance_tests(action="validate_results", feature=...)`；它验证结果仍引用当前冻结 suite，且每个 `TC-*` 都已通过。
+   - 任一 TC 未执行、失败、人工待执行或因环境阻塞时，验证状态只能是 `blocked` 或 `failed`，CP-4 不得宣称验收通过。
+   - Markdown、UAT 口述或 agent 自述不能替代该 Core-owned 结果。
+
+6. **Compatibility gate**
    - Check interface, database, config, data migration, permission, and consumer compatibility risks.
    - For frontend-visible API/VO/schema changes, record backward compatibility or migration handling.
    - For database/model/config changes, record migration, rollback, and deployment ordering risk.
    - Record accepted residual risks explicitly.
 
-6. **Baseline decision gate**
+7. **Baseline decision gate**
    - Compare implementation against `.planning/requirements/baseline.md`, especially `反向审视与用户拍板`, `待确认事项`, and `下游输入契约`.
    - Confirm implemented behavior maps to `FP-*`, `BR-*`, `BC-*`, and confirmed `CR-*` decisions.
    - Any implemented scope based on a pending decision is a verification failure unless the user explicitly approved it.
 
-7. **Lessons extraction gate**
+8. **Lessons extraction gate**
    - After verification is complete, review the findings for lessons worth preserving.
    - Candidate triggers for a lesson:
      - A regression was caught early — what pattern prevented it?

package/src/constants.ts

@@ -1,5 +1,5 @@
 export const PRODUCT_NAME = "Devflow Pi";
-export const PRODUCT_VERSION = "0.1.0";
+export const PRODUCT_VERSION = "0.1.21";
 export const PROJECT_SCHEMA_VERSION = 1;
 export const OMP_RUNTIME_VERSION = "17.2.7";
 

package/src/doctor.ts

@@ -48,8 +48,12 @@ export async function runHostDoctor(project: ProjectContext): Promise<HostDoctor
   const sdkManifest = ompSdkManifestPath();
   let installedVersion = "missing";
   if (await exists(sdkManifest)) {
-    const parsed = JSON.parse(await readFile(sdkManifest, "utf8")) as { version?: string };
-    installedVersion = parsed.version || "unknown";
+    try {
+      const parsed = JSON.parse(await readFile(sdkManifest, "utf8")) as { version?: unknown };
+      installedVersion = typeof parsed.version === "string" && parsed.version ? parsed.version : "unknown";
+    } catch {
+      installedVersion = "invalid manifest";
+    }
   }
   add("omp-sdk", installedVersion === OMP_RUNTIME_VERSION, installedVersion);
   add("agent-dir", true, devflowAgentDir());

package/src/project-context.ts

@@ -1,5 +1,5 @@
 import { randomUUID } from "node:crypto";
-import { access, lstat, mkdir, readFile, realpath, rename, writeFile } from "node:fs/promises";
+import { access, lstat, mkdir, readFile, realpath, readdir, rename, writeFile } from "node:fs/promises";
 import path from "node:path";
 
 import { PROJECT_SCHEMA_VERSION } from "./constants.ts";
@@ -26,6 +26,15 @@ const CONTROL_BOUNDARY_PATHS = [
   ".planning/artifacts",
 ];
 
+async function assertNoSymlinkDescendants(target: string): Promise<void> {
+  const entries = await readdir(target, { withFileTypes: true }).catch(() => []);
+  for (const entry of entries) {
+    const child = path.join(target, entry.name);
+    if (entry.isSymbolicLink()) throw new Error(`Refusing symlinked Devflow control path: ${child}`);
+    if (entry.isDirectory()) await assertNoSymlinkDescendants(child);
+  }
+}
+
 async function exists(target: string): Promise<boolean> {
   try {
     await access(target);
@@ -112,6 +121,7 @@ export async function assertPlanningBoundary(root: string): Promise<string> {
       if (!canonicalTarget.startsWith(`${canonicalRoot}${path.sep}`)) {
         throw new Error(`Devflow control path escapes project boundary: ${canonicalTarget}`);
       }
+      if (targetStat.isDirectory()) await assertNoSymlinkDescendants(target);
     }
   }
   return planningRoot;

package/src/runtime.ts

@@ -16,6 +16,12 @@ export interface RuntimeOptions {
   approvalMode?: "write" | "yolo";
 }
 
+const TOOL_CALL_SAFETY_PROMPT = `Devflow tool-call rules:
+- Read always requires a non-empty string path; never call Read with {} or omit path.
+- skill:// URLs identify one exact packaged resource. Never use Glob or Grep with a skill:// URL or wildcard; choose an exact resource and use Read.
+- A generated project path such as .context/spec/frontend/index.md is not automatically a packaged skill template path.
+- The project-context validator is a final gate. Do not run project-context-validate while files are still being planned or written.`;
+
 const MANAGED_SETTINGS: Record<string, unknown> = {
   modelRoleStorage: "user",
   disabledProviders: ["claude", "codex", "cursor", "gemini", "github", "opencode"],
@@ -135,6 +141,7 @@ export async function createEmbeddedRuntime(project: ProjectContext, options: Ru
     hasUI: options.hasUI ?? true,
     enableMCP: options.enableMCP ?? false,
     enableLsp: options.enableLsp ?? true,
+    appendSystemPrompt: TOOL_CALL_SAFETY_PROMPT,
   });
   return { omp, runtime, sessionDir };
 }

package/src/usage-stats.ts

@@ -153,13 +153,11 @@ export async function aggregateUsageReport(
   const perSession: Record<string, UsageTotals> = {};
   const perSkill: Record<string, UsageTotals> = {};
   const files = await jsonlFiles(sessionRoot);
-  totals.sessionFiles = files.length;
 
   for (const file of files) {
     const session = sessionLabel(sessionRoot, file);
     if (range.session && range.session !== "all" && range.session !== session) continue;
-    const sessionTotals = (perSession[session] ??= emptyUsage());
-    sessionTotals.sessionFiles = 1;
+    let sessionTotals: UsageTotals | undefined;
     let currentModel = "unknown";
     const content = await readFile(file, "utf8").catch(() => "");
     for (const line of content.split("\n")) {
@@ -181,6 +179,8 @@ export async function aggregateUsageReport(
       if (!usage) continue;
       const model = currentModel;
       if (range.models?.length && !range.models.includes(model)) continue;
+      sessionTotals ??= (perSession[session] ??= { ...emptyUsage(), sessionFiles: 1 });
+      totals.sessionFiles = Math.max(totals.sessionFiles, Object.keys(perSession).length);
       const modelTotals = (perModel[model] ??= emptyUsage());
       addUsage(totals, usage);
       addUsage(sessionTotals, usage);