homegraph 1.1.2 → 1.2.0

package/scripts/exp_boundary_eval/data/agents.json

@@ -0,0 +1,109 @@
+{
+  "default": "claude",
+  "agents": {
+    "claude": {
+      "name": "Claude Code",
+      "binary": "claude",
+      "flags": {
+        "print": "-p",
+        "verbose": "--verbose",
+        "output_format": ["--output-format", "stream-json"],
+        "session": ["--session-id"],
+        "resume": ["--resume"],
+        "max_turns": ["--max-turns"],
+        "skip_permissions": ["--dangerously-skip-permissions"]
+      },
+      "parser": "claude_stream_json",
+      "note": "Claude Code CLI. stream-json mode requires --verbose."
+    },
+    "cursor": {
+      "name": "Cursor Agent",
+      "binary": "cursor-agent",
+      "flags": {
+        "print": "-p",
+        "output_format": ["--output-format", "stream-json"],
+        "resume": ["--resume"],
+        "continue": ["--continue"]
+      },
+      "parser": "cursor_stream_json",
+      "missing": ["session", "max_turns", "skip_permissions", "verbose"],
+      "note": "Verified 2026-06-01 with cursor-agent v2.x. No --session-id, --max-turns, or --skip-permissions flags. Uses --resume <chatId> / --continue for session persistence."
+    },
+    "codex": {
+      "name": "OpenAI Codex CLI",
+      "binary": "codex",
+      "flags": {
+        "print": "exec",
+        "verbose": "--verbose",
+        "output_format": ["--json"],
+        "session": ["--session"],
+        "resume": ["--resume"],
+        "max_turns": ["--max-turns"],
+        "skip_permissions": ["--yes"]
+      },
+      "parser": "codex_json",
+      "note": "OpenAI Codex CLI. Uses 'codex exec' for non-interactive mode. Output format is JSON with '--json' flag."
+    },
+    "opencode": {
+      "name": "OpenCode",
+      "binary": "opencode",
+      "flags": {
+        "print": "run",
+        "verbose": "--verbose",
+        "output_format": ["--format", "json"],
+        "session": ["--session-id"],
+        "resume": ["--resume"],
+        "max_turns": ["--max-turns"],
+        "skip_permissions": ["--auto-approve"]
+      },
+      "parser": "opencode_json",
+      "note": "OpenCode CLI. Uses 'opencode run' for non-interactive mode. Output format may vary — check your version."
+    },
+    "deveco": {
+      "name": "DevEco Studio AI",
+      "binary": "deveco",
+      "binary_env": "DEVECO_BIN",
+      "flags": {
+        "print": "run",
+        "output_format": ["--format", "json"],
+        "session": ["--session"],
+        "resume": ["--session"],
+        "continue": ["--continue"],
+        "skip_permissions": ["--dangerously-skip-permissions"]
+      },
+      "parser": "deveco_json",
+      "missing": ["session", "max_turns", "verbose"],
+      "note": "Verified 2026-06-01 with deveco CLI. Uses 'deveco run --format json <prompt>' for non-interactive. No --max-turns flag. --session only works for resume (existing sessions), not for creating new ones. Has --dangerously-skip-permissions."
+    },
+    "generic": {
+      "name": "Generic Agent (JSON)",
+      "binary": "agent",
+      "flags": {
+        "print": "--print",
+        "verbose": "--verbose",
+        "output_format": ["--json"],
+        "session": ["--session"],
+        "resume": ["--resume"],
+        "max_turns": ["--max-turns"],
+        "skip_permissions": ["--yes"]
+      },
+      "parser": "generic_json",
+      "note": "Generic fallback. Replace 'binary' and flags with your agent's actual CLI. The generic_json parser auto-detects event structure."
+    },
+    "text": {
+      "name": "Text-Only Agent",
+      "binary": "echo",
+      "flags": {
+        "print": "",
+        "verbose": "",
+        "output_format": [],
+        "session": [],
+        "resume": [],
+        "max_turns": [],
+        "skip_permissions": []
+      },
+      "parser": "text",
+      "note": "Fallback for agents that only produce plain text. Parses file paths from text lines heuristically."
+    }
+  }
+}

package/scripts/exp_boundary_eval/data/experiments.json

@@ -0,0 +1,140 @@
+{
+  "meta": {
+    "version": "2.0",
+    "project": "OpenHarmony Photos",
+    "repo_url": "https://gitcode.com/openharmony/applications_photos.git",
+    "default_branch": "weekly_20260601",
+    "suite": "boundary-gear-switching"
+  },
+  "experiments": [
+    {
+      "id": "1-1",
+      "title": "零缺口任务 — 快速检索档位",
+      "category": "档位切换",
+      "type": "oneshot",
+      "max_turns": 10,
+      "prompt": "在 common/src/main/ets/default/model/common/Constants.ts 文件中，找到 GRID_GUTTER 常量的定义行，把它的值从 2 改成 4。只修改这一个文件。",
+      "target": "Constants.ts:739",
+      "trap": ".ts 扩展名而非 .ets"
+    },
+    {
+      "id": "1-2",
+      "title": "弱缺口任务 — 快速→中深度升级",
+      "category": "档位切换",
+      "type": "oneshot",
+      "max_turns": 30,
+      "prompt": "在 common 模块的 ThirdSelectAlbumGridBase 组件中，当前 grid 间距是硬编码引用 Constants.GRID_GUTTER。请把这个间距改成可通过构造函数参数配置的值（保留默认值 2），只修改 ThirdSelectAlbumGridBase.ets 一个文件。",
+      "target": "feature/thirdselect/.../ThirdSelectAlbumGridBase.ets",
+      "trap": "文件在 feature/thirdselect 而非 common"
+    },
+    {
+      "id": "1-3",
+      "title": "强缺口任务 — 中深度→深度升级",
+      "category": "档位切换",
+      "type": "oneshot",
+      "max_turns": 30,
+      "prompt": "当图片加载失败时，我想显示一个统一的占位图组件。请帮我梳理：目前图片加载失败的处理逻辑在哪些文件中？涉及哪些组件和错误状态？如果要实现统一的占位图，需要修改哪些文件？只需分析和报告，不要修改代码。",
+      "target": "ImageGridItemComponent.ets, NoPhotoComponent.ets, PhotoItem.ets, PreviewItem.ets"
+    },
+    {
+      "id": "2",
+      "title": "跨层调用链探索 — 图片编辑链路",
+      "category": "跨层探索",
+      "type": "oneshot",
+      "max_turns": 50,
+      "prompt": "请帮我梳理从用户点击图片编辑按钮到最终调用 C++ 滤镜引擎的完整调用链路。需要覆盖从 ArkTS UI 层到 native.ts 再到 libImageEditor.so 的所有关键函数和文件。只需要分析和报告，不要修改代码。",
+      "trap": "native.ts 仅一行 import，C++ 在 imageEditor/common/src/main/cpp/"
+    },
+    {
+      "id": "3-1",
+      "title": "幻觉测试 — 不存在的模块",
+      "category": "失败模式",
+      "type": "oneshot",
+      "max_turns": 20,
+      "prompt": "请帮我找到 PhotoWatermarkEngine 模块，并解释它是如何实现图片水印添加的。",
+      "trap": "PhotoWatermarkEngine 完全不存在"
+    },
+    {
+      "id": "3-2",
+      "title": "遗漏测试 — 命名不直观的关键文件",
+      "category": "失败模式",
+      "type": "oneshot",
+      "max_turns": 40,
+      "prompt": "请找出所有与图片分享功能相关的文件，并解释分享流程。只需要分析和报告，不要修改代码。",
+      "trap": "KnockShareUtil.ets — 命名以 Knock 开头而非 Share"
+    },
+    {
+      "id": "3-3",
+      "title": "过度探索测试 — 72 个相似 Utils 文件",
+      "category": "失败模式",
+      "type": "oneshot",
+      "max_turns": 25,
+      "prompt": "我需要一个工具函数，用于判断图片是否为竖屏（高度大于宽度）。请在项目中找一个最合适的位置（Utils 文件）添加这个函数，并实现它。",
+      "target": "common/.../utils/ImageUtil.ets",
+      "utils_count": 72
+    },
+    {
+      "id": "4-1",
+      "title": "资源配置修改 — 应用名称",
+      "category": "非代码资产",
+      "type": "oneshot",
+      "max_turns": 15,
+      "prompt": "把应用的名字从 'Gallery' 改成 '我的相册'。",
+      "target": "AppScope/resources/base/element/string.json",
+      "trap": "15+ 语言版本"
+    },
+    {
+      "id": "4-2",
+      "title": "权限配置添加",
+      "category": "非代码资产",
+      "type": "oneshot",
+      "max_turns": 15,
+      "prompt": "给应用添加读取存储的权限（ohos.permission.READ_IMAGEVIDEO）。",
+      "target": "product/phone/src/main/module.json5",
+      "trap": "权限已存在"
+    },
+    {
+      "id": "4-3",
+      "title": "构建依赖添加",
+      "category": "非代码资产",
+      "type": "oneshot",
+      "max_turns": 25,
+      "prompt": "我需要使用一个第三方图片压缩库，请告诉我如何在这个项目中添加依赖。需要查看哪些配置文件？具体步骤是什么？只需要分析和报告，不要修改代码。",
+      "trap": "ohpm + oh-package.json5 而非 npm"
+    },
+    {
+      "id": "5",
+      "title": "会话持久性与上下文衰减",
+      "category": "会话持久性",
+      "type": "session",
+      "max_turns_per_round": 25,
+      "rounds": [
+        {
+          "round": 1,
+          "title": "建立上下文",
+          "prompt": "帮我理解这个图库项目的整体架构，包括主要模块和它们的职责。只分析和报告，不修改代码。"
+        },
+        {
+          "round": 2,
+          "title": "添加刷新按钮",
+          "prompt": "在 PhotoGridPage.ets 组件的标题栏右侧添加一个刷新按钮（使用 SymbolGlyph 的 arrow.clockwise 图标）。只修改 PhotoGridPage.ets 一个文件。"
+        },
+        {
+          "round": 3,
+          "title": "刷新联动媒体库",
+          "prompt": "点击上一轮添加的刷新按钮后，重新扫描媒体库并更新照片列表。需要查看 MediaDataSource 和相关数据刷新逻辑。"
+        },
+        {
+          "round": 4,
+          "title": "修改详情页动画",
+          "prompt": "修改图片详情页的滑动动画时长为 300ms。在 PhotoSwiper.ets 中查找动画相关常量并修改。"
+        },
+        {
+          "round": 5,
+          "title": "回溯修改按钮位置",
+          "prompt": "把之前在第 2 轮添加到 PhotoGridPage 的刷新按钮移到导航栏左侧。"
+        }
+      ]
+    }
+  ]
+}