tsingyuai · JunkaiWang-TheoPhy · Apr 3, 2026 · Apr 7, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -32,6 +32,32 @@ jobs:
       - name: Type check
         run: pnpm exec tsc --noEmit
 
+      - name: Verify packed plugin stays dependency-free
+        run: |
+          PACK_FILE="$(npm pack --silent)"
+          TMP_DIR="$(mktemp -d)"
+          tar -xzf "$PACK_FILE" -C "$TMP_DIR"
+
+          node - "$TMP_DIR/package/package.json" <<'EOF'
+          const fs = require("node:fs");
+          const manifestPath = process.argv[2];
+          const manifest = JSON.parse(fs.readFileSync(manifestPath, "utf8"));
+
+          const deps = manifest.dependencies ?? {};
+          if (Object.keys(deps).length > 0) {
+            console.error("Packed plugin must not publish runtime dependencies.");
+            console.error(JSON.stringify(deps, null, 2));
+            process.exit(1);
+          }
+
+          const peerDeps = manifest.peerDependencies ?? {};
+          if (Object.keys(peerDeps).length > 0) {
+            console.error("Packed plugin must not publish peer dependencies.");
+            console.error(JSON.stringify(peerDeps, null, 2));
+            process.exit(1);
+          }
+          EOF
+
   release:
     needs: build
     if: github.event_name == 'push' && github.ref == 'refs/heads/main'

diff --git a/README.en.md b/README.en.md
@@ -90,9 +90,8 @@ Driven by multi-agent iteration: the orchestrator holds hypotheses and all accum
 │                          │──→│  edit them too               │
 │  arxiv_search            │   └──────────────────────────────┘
 │  openalex_search         │
-│  github_search           │
-│  paper_browser           │
-│  code_executor           │
+│  platform / local exec   │
+│  / skill steps           │
 └──────────────────────────┘
 ```
 
@@ -116,14 +115,23 @@ Pipeline internally uses multi-agent iteration: the orchestrator holds hypothese
 
 ### Tool Layer
 
-The agents' hands and eyes:
+Scientify organizes research capabilities as a unified semantic layer spanning the runtime registration surface and the workflow execution surface:
 
-| Tool | Capability |
-|------|-----------|
-| `arxiv_search` / `openalex_search` | Search academic papers (arXiv + cross-disciplinary) |
-| `github_search` | Search open-source code implementations |
-| `paper_browser` | Paginated paper reading, avoids context overflow |
-| `code_executor` | Execute experiment code in `uv`-isolated environment |
+| Capability semantic | Primary carrier | Typical target | Execution path | Typical outputs |
+|---------------------|-----------------|----------------|----------------|-----------------|
+| `arxiv_search`-like capability | Runtime-registered tool | arXiv metadata and candidate paper sets | Registered in the plugin entrypoint and invoked through standard tool calls | search results, candidate paper lists, survey inputs |
+| `openalex_search`-like capability | Runtime-registered tool | cross-disciplinary metadata, DOI, citation, and OA signals | Registered in the plugin entrypoint and invoked through standard tool calls | search results, complementary literature sources, survey inputs |
+| `github_search`-like capability | Workflow execution step | open-source implementations, baseline repos, reproduction leads | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | `survey_res.md`, implementation references, repo links |
+| `paper_browser`-like capability | Workflow execution step | `.md`, `.tex`, long papers, and drafts | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | chunked reading results, local excerpts, structured notes |
+| `code_executor`-like capability | Workflow execution step | workspace training, evaluation, scripts, and experiment runs | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | run logs, validation outputs, experiment artifacts |
+| `arxiv_download`-like capability | Workflow execution step | arXiv source or PDF acquisition | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | paper files under `papers/` |
+| `openreview_lookup`-like capability | Workflow execution step | reviews, decisions, and forum context | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | review evidence, comparison notes, review summaries |
+| `unpaywall_download`-like capability | Workflow execution step | OA PDF acquisition by DOI | carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills | downloaded PDFs and provenance records |
+
+At the implementation boundary:
+- the `tool registry` exposes stable, named, directly invokable runtime interfaces
+- `skills` define stage ordering, input/output contracts, and artifact materialization
+- file reads, command execution, downloads, and external-site interactions are carried out through OpenClaw platform capabilities, local environment commands, or steps defined inside skills
 
 > Scientify runs on [OpenClaw](https://github.com/openclaw/openclaw), natively leveraging the platform's MCP servers (Slack / Feishu push), browser automation (paywalled paper downloads), multi-session concurrency (parallel multi-direction research), and more.
 
@@ -275,25 +283,19 @@ Check status anytime:
 
 | Skill | Description |
 |-------|-------------|
+| **paper-download** | Acquire paper files into the project workspace for downstream reading and analysis. |
+| **metabolism** | Run day-0 bootstrapping or day-N daily knowledge metabolism for a research topic. |
 | **write-review-paper** | Draft a review/survey paper from project research outputs. |
-| **research-subscription** | Create/list/remove scheduled Scientify jobs via `scientify_cron_job` (research digests or plain reminders). |
 
 </details>
 
 <details>
-<summary><b>Tools (available to LLM)</b></summary>
+<summary><b>Runtime-registered Tools</b></summary>
 
 | Tool | Description |
 |------|-------------|
 | `arxiv_search` | Search arXiv papers. Returns metadata (title, authors, abstract, ID). Supports sorting by relevance/date and date filtering. |
-| `arxiv_download` | Batch download papers by arXiv ID. Prefers .tex source files (PDF fallback). |
 | `openalex_search` | Search cross-disciplinary academic papers via OpenAlex API. Returns DOI, authors, citation count, OA status. |
-| `openreview_lookup` | Lookup OpenReview evidence by title/ID/forum. Returns decision, review rating/confidence aggregates, and review summaries. |
-| `unpaywall_download` | Download open access PDFs by DOI via Unpaywall API. Non-OA papers are silently skipped. |
-| `github_search` | Search GitHub repositories. Returns repo name, description, stars, URL. Supports language filtering and sorting. |
-| `paper_browser` | Paginated browsing of large paper files (.tex/.md) to avoid context overflow. |
-| `scientify_cron_job` | Manage scheduled Scientify jobs (`upsert`/`list`/`remove`). |
-| `scientify_literature_state` | Persistent incremental state for subscriptions: dedupe, record, feedback, and status inspection. |
 
 </details>
 
@@ -308,9 +310,7 @@ Check status anytime:
 | `/projects` | List all projects |
 | `/project-switch <id>` | Switch active project |
 | `/project-delete <id>` | Delete a project |
-| `/research-subscribe ...` | Create/update scheduled Scientify jobs |
-| `/research-subscriptions` | Show your scheduled Scientify jobs |
-| `/research-unsubscribe [job-id]` | Remove your scheduled Scientify jobs |
+| `/metabolism-status` | Show knowledge metabolism status for the active project |
 
 </details>
 

diff --git a/README.md b/README.md
@@ -90,9 +90,8 @@ Scientify 采用**新陈代谢模式**——持续地摄入、消化、沉淀、
 │                          │──→│                              │
 │  arxiv_search            │   └──────────────────────────────┘
 │  openalex_search         │
-│  github_search           │
-│  paper_browser           │
-│  code_executor           │
+│  平台能力 / 本地命令      │
+│  / skill 执行步骤        │
 └──────────────────────────┘
 ```
 
@@ -116,14 +115,23 @@ Pipeline 内部是多 Agent 迭代：编排器持有假设，spawn 子 agent 执
 
 ### 工具层
 
-Agent 的手和眼：
-
-| 工具 | 能力 |
-|------|------|
-| `arxiv_search` / `openalex_search` | 搜索学术论文（arXiv + 跨学科） |
-| `github_search` | 搜索开源代码实现 |
-| `paper_browser` | 分页精读论文，避免上下文溢出 |
-| `code_executor` | 在 `uv` 隔离环境中执行实验代码 |
+Scientify 将研究能力组织为统一的能力语义层，并分别落在运行时注册面与工作流执行面中：
+
+| 能力语义 | 主要载体 | 典型对象 | 执行路径 | 典型产物 |
+|---------|----------|----------|----------|----------|
+| `arxiv_search` 类能力 | 运行时注册 tool | arXiv 论文元数据、候选论文集合 | 由插件入口注册，并通过标准 tool call 调用 | 检索结果、候选论文列表、survey 输入 |
+| `openalex_search` 类能力 | 运行时注册 tool | 跨学科论文元数据、DOI、引用与 OA 信息 | 由插件入口注册，并通过标准 tool call 调用 | 检索结果、补充文献来源、survey 输入 |
+| `github_search` 类能力 | 工作流执行步骤 | 开源实现、baseline 仓库、复现线索 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | `survey_res.md`、实现参考、仓库链接 |
+| `paper_browser` 类能力 | 工作流执行步骤 | `.md`、`.tex`、长论文文本、草稿 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | 分段阅读结果、局部摘录、结构化笔记 |
+| `code_executor` 类能力 | 工作流执行步骤 | 工作区中的训练、评测、脚本与实验运行 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | 运行日志、验证结果、实验产物 |
+| `arxiv_download` 类能力 | 工作流执行步骤 | arXiv source / PDF 获取 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | `papers/` 下的论文文件 |
+| `openreview_lookup` 类能力 | 工作流执行步骤 | review、decision、forum 上下文 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | 评审证据、对比说明、review notes |
+| `unpaywall_download` 类能力 | 工作流执行步骤 | DOI 对应的 OA PDF 获取 | 由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成 | 下载后的 PDF、来源记录 |
+
+从实现边界看：
+- `tool registry` 提供稳定、命名、可直接调用的运行时接口
+- `skill` 负责编排研究阶段中的执行顺序、输入输出契约与产物落盘
+- 文件读取、命令执行、下载与外部站点交互由 OpenClaw 平台能力、本地环境命令或 skill 内步骤来完成
 
 > Scientify 运行在 [OpenClaw](https://github.com/openclaw/openclaw) 之上，天然可调用平台的 MCP 服务器（Slack / 飞书推送）、浏览器自动化（付费文献下载）、多会话并发（多方向并行研究）等能力。
 

diff --git a/package.json b/package.json
@@ -39,9 +39,6 @@
     "prepublishOnly": "npm run clean && npm run build",
     "release": "npm version patch && git push && git push --tags"
   },
-  "dependencies": {
-    "@sinclair/typebox": "^0.32.0"
-  },
   "devDependencies": {
     "@semantic-release/changelog": "^6.0.3",
     "@semantic-release/commit-analyzer": "^13.0.1",
@@ -55,9 +52,6 @@
     "semantic-release": "^25.0.3",
     "typescript": "^5.3.0"
   },
-  "peerDependencies": {
-    "openclaw": ">=2024.0.0"
-  },
   "openclaw": {
     "extensions": [
       "./dist/index.js"