← 返回 Skills 市场

literature-recommendation

Name: literature-recommendation
Author: zhcolin0313

作者 Colin · GitHub ↗ · v1.0.3 · MIT-0

cross-platform ✓ 安全检测通过

总下载

当前安装

版本数

在 OpenClaw 中安装

/install literature-recommendation

功能描述

用于论文推荐、文献推荐、论文匹配、按偏好找论文、根据研究方向推荐论文；从数据库读取成员偏好，召回候选论文并输出给OpenClaw。

使用说明 (SKILL.md)

Literature Recommendation

What it does

This skill implements a minimal literature recommendation workflow for a team:

store member research profiles
fetch candidate papers from arXiv
score papers with lightweight rules
persist run history and recommendation output in PostgreSQL
output personal card payload for OpenClaw delivery

触发词

优先使用中文触发，建议包括：

论文推荐
文献推荐
论文匹配
按偏好找论文
根据研究方向推荐论文
给我推荐相关论文
帮我推荐几篇论文

Usage

Run the CLI from the project root:

python entrypoints/run_paper_matching.py --dry-run

Workflow

Load member profiles from PostgreSQL.
Fetch candidate papers from arXiv.
Score papers with direction and keyword matching.
Store papers and recommendations in PostgreSQL.
Build delivery payloads: personal card JSON only.
Return payload to OpenClaw for actual Feishu sending.

职责边界（规范）

本 Skill 负责：
- 从数据库读取成员偏好
- 从 arXiv 召回候选论文
- 用硬规则打分并给出规则理由（rule_reason）
- 输出 openclaw_rerank_payload 与 personal_cards
OpenClaw 负责：
- 读取 openclaw_rerank_payload 并调用 LLM 做语义复排
- 判定是否推荐（keep/drop）与最终排序
- 生成最终推荐理由（可覆盖 rule_reason）
- 推荐理由需解释论文摘要核心内容，并指出与成员偏好的匹配点
- 执行私聊发送和定时调度

输出契约（推荐结果）

候选输入给 OpenClaw：openclaw_rerank_payload
- strategy: hard_rule_recall
- llm_reason_contract:
  - language: zh-CN
  - reason_style: explanatory
  - required_source:
    - paper.title
    - paper.abstract
    - profile.primary_direction
    - profile.keywords
  - forbidden:
    - title_only_reason
  - reason_generation_steps:
    - 先阅读 paper.abstract 提取论文核心方法/任务
    - 再对照成员偏好字段判断匹配点
    - 最后生成一句概括 + 一句匹配说明
  - must_include:
    - 论文核心内容一句话概括
    - 与该成员偏好匹配的具体点
  - max_length: 120
- by_member[record_id].candidates[*]:
  - paper_id
  - rule_score
  - rule_reason
  - paper (title/abstract/tags/url 等)
- llm_output_schema:
  - by_member[record_id].items[*]:
    - paper_id
    - keep
    - rank
    - score
    - reason
最终发送：delivery_payload.personal_cards
- record_id
- paper_id
- rank
- score
- reason
- paper_title
- paper_abstract
- card

Notes

The MVP uses rule-based matching only.
The weekly push is part of the MVP, not a later iteration.
The payload now targets personal direct messages only.
Feishu sending, Bitable sync, and feedback storage are handled by the data-hub skill.
This skill is invoked manually from chat or by OpenClaw orchestration; it does not schedule itself.
If you need timed execution, let OpenClaw or an external scheduler trigger this skill on a schedule.

安全使用建议

This skill appears to do what it says: it needs a PostgreSQL DSN and will create and modify tables in that database, fetch papers from arXiv, apply local rule-based scoring, and write reports/payloads to the configured output directory. Before installing or running: 1) Resolve the manifest inconsistency (registry metadata vs skill.yaml). 2) Run it against a test or dedicated database (not a production DB) because it will create tables and persist data. 3) Provide only a DB user with the minimal privileges needed (avoid giving a DB superuser). 4) Ensure required Python dependencies are installed in a controlled environment. 5) Review ~/.config/literature-recommendation/.env (or OPENCLAW_ENV_FILE) to ensure you don't place unrelated secrets there. 6) If you need stricter review, inspect the storage._init_schema and SQL statements and run the code in a sandboxed environment to verify behavior.

功能分析

Type: OpenClaw Skill Name: literature-recommendation Version: 1.0.3 The literature-recommendation skill bundle is a legitimate tool designed to fetch research papers from arXiv and match them against user profiles stored in a PostgreSQL database. The code uses standard libraries (requests, psycopg) with secure practices like parameterized SQL queries to prevent injection, and the SKILL.md instructions are strictly aligned with its stated purpose of generating literature recommendations and formatting Feishu cards.

能力评估

✓ Purpose & Capability

Name/description (paper/literature recommendation) align with the implementation: it fetches from arXiv, loads member profiles from PostgreSQL, does rule-based scoring, stores papers and recommendations, and builds personal-card payloads for OpenClaw.

✓ Instruction Scope

SKILL.md instructs running the CLI to load profiles from the DB, fetch arXiv, score, persist, and output payloads. The runtime instructions match the code (pipeline, storage, fetcher). There are no instructions to read unrelated local files or contact unexpected external endpoints beyond arXiv and the configured database.

ℹ Install Mechanism

There is no explicit install spec in the registry (no packaged install step), but the repo includes Python source and requirements.txt. That means the code runs directly but expects dependencies (requests, python-dotenv, psycopg[binary]) to be installed by the operator. This is reasonable for a Python CLI but the package omission (no install instructions beyond the CLI usage) is worth noting.

ℹ Credentials

The skill requires a PostgreSQL DSN (OPENCLAW_DB_DSN) and an output directory (OPENCLAW_OUTPUT_DIR) — both declared in skill.yaml and env-example and are appropriate for its function. Note: the top-level registry summary in the prompt incorrectly listed 'Required env vars: none' which is inconsistent with the included skill.yaml/env-example; this mismatch should be resolved before deployment.

✓ Persistence & Privilege

The skill is not always-enabled and uses normal autonomous invocation defaults. It will initialize database schema (CREATE TABLE IF NOT EXISTS) on first connect and persist papers/recommendations to the configured PostgreSQL database — this is expected for its purpose but means it will modify the specified DB. It does not modify other skills or system-wide configs.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install literature-recommendation
安装完成后，直接呼叫该 Skill 的名称或使用 /literature-recommendation 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.3

- Updated the llm_reason_contract in the output contract to specify required input sources, forbidden reasoning styles, and explicit reasoning steps. - Added delivery payload fields: paper_title and paper_abstract to personal_cards schema. - No code changes; documentation and output contract clarifications only.

v1.0.2

- Added new trigger phrase: “帮我推荐几篇论文”. - Clarified responsibilities between this skill and OpenClaw, detailing input/output and role boundaries. - Specified output contracts, including required structure for candidate recommendations and reasoning fields. - No code or workflow changes; documentation and usage clarified.

v1.0.1

- Removed setup.sh for streamlined deployment and maintenance. - Delivery payload is now personal card JSON only; group summary text is no longer included. - Documentation updated to clarify that the payload targets personal direct messages only.

v1.0.0

- Initial release of the literature recommendation skill. - Supports storing member research profiles and preferences. - Fetches candidate papers from arXiv and scores them based on direction and keyword matching. - Persists recommendation history and paper data in PostgreSQL. - Outputs both group summary text and individual card recommendations for delivery via OpenClaw. - Manual trigger via chat or orchestration; scheduling is managed externally.

元数据

Slug literature-recommendation

版本 1.0.3

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 4

常见问题