功能描述

Pluggable AI report engine generating multi-preset, multi-source briefings with smart scoring, deduplication, LLM editing, quality control, rendering, and em...

使用说明 (SKILL.md)

🦞 LunaClaw Brief — OpenClaw Skill

Name: Ai Cv Weekly
Author: llx9826

Pluggable AI-Powered Report Engine | 插件化 AI 智能简报引擎

Overview / 概述

LunaClaw Brief is an OpenClaw Skill that generates intelligent reports across multiple domains. It features a plugin architecture with 8-stage pipeline, 8 data sources, 7 editor strategies, hybrid intent routing, streaming output, and multi-channel delivery.

LunaClaw Brief 是一个 OpenClaw Skill，支持多领域智能报告生成。采用插件化架构，8 阶段管线，8 个数据源，7 种编辑器策略，混合意图路由，流式输出，多渠道推送。

Presets / 预设

Preset	Type	Description
`ai_cv_weekly`	Tech	AI/CV/多模态技术深度周报
`ai_daily`	Tech	AI 技术日报
`finance_weekly`	Finance	金融投资周报
`finance_daily`	Finance	金融快报日刊
`stock_a_daily`	Stock	A 股日报（大盘/板块/北向资金/IPO/异动）
`stock_hk_daily`	Stock	港股日报（恒生/南向资金/中概股/IPO）
`stock_us_daily`	Stock	美股日报（S&P/NASDAQ/科技巨头/IPO）
Custom	Any	via `--create-preset` natural language

Data Sources / 数据源

Source	Coverage
`github`	Open-source projects + competitor discovery
`arxiv`	Academic papers
`hackernews`	Tech community discussions
`paperswithcode`	ML papers with code
`finnews`	Financial news (HN + BBC/NYT RSS)
`yahoo_finance`	Yahoo Finance RSS + market indices
`eastmoney`	东方财富 (A-share news, IPO calendar)
`xueqiu`	雪球 (hot stocks, sentiment)

Usage / 使用方式

Via CLI

python run.py                                    # AI/CV Weekly (default)
python run.py --preset stock_a_daily             # A-share daily
python run.py --hint "今天A股怎么样"               # Auto-route → stock_a_daily
python run.py --hint "帮我看看腾讯和阿里"          # LLM auto-route
python run.py --preset ai_cv_weekly --email      # Generate + email
python run.py --create-preset "我是新能源基金经理"  # Create custom preset
python -m brief.scheduler                        # Run scheduled jobs

Via OpenClaw Skill API

from run import generate_report

result = generate_report({
    "preset": "stock_hk_daily",
    "hint": "重点关注腾讯和美团",
    "send_email": True,
})

Trigger Phrases / 触发短语

"生成周报" / "生成日报" / "generate weekly/daily"
"生成金融周报/日报" / "generate finance report"
"生成A股/港股/美股日报" / "generate A-share/HK/US brief"

Key Capabilities / 核心能力

Hybrid Intent Router: Regex (instant) + LLM classification (fallback) for preset routing
Unified Markdown Schema: ## Section → ### N. Item → **Label**: → **🦞 Claw**: structure
Streaming Output: pipeline.run_stream() yields progress events + content chunks
Multi-Channel Delivery: Email (SMTP), Webhook (Slack/DingTalk/Feishu/custom)
Cron Scheduler: Auto-generate reports on schedule with multi-channel push
Custom Presets: Create domain-specific presets via natural language
Quality Control: Auto-check structure/word count, retry if below threshold
Historical Dedup: Content fingerprinting with configurable time window

Architecture / 架构

Pattern	Where
Adapter	`BaseSource` → 8 source adapters
Strategy	`BaseEditor` → 7 editor strategies
Pipeline	8-stage `ReportPipeline` (sync + stream)
Registry	`@register_source` / `@register_editor` decorators
Observer	`MiddlewareChain` for timing, metrics, custom hooks
Factory	`create_sources()` / `create_editor()`
Cache	`FileCache` with TTL for API responses

Configuration / 配置

Global config in config.yaml. Secrets in config.local.yaml (gitignored).

# config.local.yaml (not committed)
llm:
  api_key: "your-key"

Auto-detects OpenClaw environment (OPENCLAW_API_KEY / OPENCLAW_BASE_URL) — no manual LLM config needed when running as an OpenClaw skill.

Output / 输出

HTML: Self-contained dark-theme report with embedded Luna logo
Markdown: Raw LLM output
PDF: Via WeasyPrint (optional)
Email: HTML body + PDF attachment
Webhook: Slack/DingTalk/Feishu/custom HTTP POST

Built by llx & Luna 🐱 — where the claw meets the code. 🦞

安全使用建议

This package appears to implement the described AI report engine, but there are a few things to check before running it with real credentials or on production hosts: - LLM credentials: The code requires an LLM API key (OPENCLAW_API_KEY / OPENCLAW_TOKEN or BAILIAN_API_KEY) and base URL, but the skill metadata lists no required env vars — ensure you only supply keys with least privilege and avoid using production/privileged credentials until you're confident. Consider running with a test key. - Delivery channels: The skill can send email (SMTP) and post to arbitrary webhooks (Slack/DingTalk/Feishu/custom). Review brief/sender.py and any config examples — do not provide webhook URLs or SMTP credentials you care about until you audit the code paths that perform delivery. - Network access: The package fetches from multiple public sources (GitHub, arXiv, HackerNews, Yahoo Finance, Eastmoney, Xueqiu, etc.). If you need to sandbox network access, run in an isolated environment or container. - Persistence: It writes caches, dedup stores, issue counters, and saved presets under data/ — inspect these directories for any sensitive persisted content and set appropriate filesystem permissions. - Inspect static assets: The static/luna_logo_b64.txt is a large base64 blob. Open it and verify it is actually an image/logo and not hidden data or scripts. - Review delivery and source adapters: If you plan to use Eastmoney/Xueqiu adapters, confirm whether those adapters scrape pages or require credentials; scraping can break and sometimes involve more network surface. Also review rate limiting and error handling. - Run in a safe test environment first: Execute the package locally with dummy config.local.yaml (no real secrets) to observe behavior, then enable credentials stepwise. If you want, I can: (a) highlight lines in specific files that reference environment variables and external endpoints, (b) list functions that perform network calls or write to disk, or (c) extract and summarize the sender.py and llm.py behaviors in more detail.

功能分析

Type: OpenClaw Skill Name: lunaclaw-brief Version: 4.0.0 LunaClaw Brief is a comprehensive AI-powered report generation engine that aggregates data from multiple technical and financial sources (including GitHub, arXiv, Eastmoney, and Yahoo Finance). The codebase implements a well-structured 8-stage pipeline (Fetch, Score, Select, Dedup, Edit, Quality, Render, Output) using professional design patterns like Adapters and Registries. It supports automated delivery via SMTP and webhooks, and includes a built-in cron scheduler for recurring reports. While the skill possesses broad network and file system access, these capabilities are strictly aligned with its stated purpose of content aggregation and report distribution. No evidence of malicious intent, unauthorized data exfiltration, or prompt injection attacks was found; sensitive credentials like API keys and SMTP passwords are appropriately handled via environment variables or local configuration files.

能力评估

ℹ Purpose & Capability

The name/description (AI report engine) aligns with the included code: sources, scoring, editors, pipeline, renderer and delivery. However the skill metadata declared no required env vars while the code clearly requires LLM credentials and optionally delivery credentials (SMTP / webhook). That mismatch between declared requirements and actual code is an inconsistency to be aware of.

ℹ Instruction Scope

SKILL.md instructions are consistent with the code (run run.py, create config.local.yaml for secrets, or rely on OPENCLAW env). The instructions do not request unrelated system files, but they do instruct creation of local secret file and running networked fetch/delivery operations. The SKILL.md contained a large embedded-logo/base64 mention (pre-scan flagged 'base64-block') which appears intended for embedding a logo but can also be used as a covert data blob — expected for embedding but worth reviewing.

✓ Install Mechanism

No install spec was provided (instruction-only in registry), and the repository files are included directly. That lowers installer risk, but means the code will run as-is when invoked — inspect the package before execution. There are no suspicious external download URLs in an install step.

⚠ Credentials

The skill metadata declares no required environment variables, but brief/llm.py requires an LLM API key via OPENCLAW_API_KEY/OPENCLAW_TOKEN (or BAILIAN_API_KEY) and base URL variables. The project also supports SMTP and arbitrary webhook delivery (sender.py), which will need delivery credentials/URLs not declared in metadata. Requesting LLM keys and delivery credentials is reasonable for this functionality, but the omission from requires.env is a discrepancy and increases risk of accidental credential exposure. The package also persists data (custom_presets, issue counters, dedup store) to local disk — expected but note persisted content may include source text or generated output.

✓ Persistence & Privilege

The skill writes to its own data directories (data/custom_presets, cache, issue counter, used-item JSON) for caching, deduplication and presets. It does not request always:true or attempt to modify other skills. Local persistence is reasonable for this application but consider filesystem permissions and where data is stored.

版本历史

v4.0.0

V2 architecture: unified schema, 8 data sources, hybrid LLM router, streaming, scheduler, webhook delivery, custom presets

v3.1.0

v3.1: Plugin registry, 8-stage pipeline, finance presets, Luna logo, bilingual docs, quality checker, middleware chain

v3.0.0

Version 3.0.0 — Major feature release with architecture and extensibility upgrades. - Introduces a fully plugin-based, modular report engine supporting multiple presets (AI/CV/Finance weekly/daily, extendable). - Implements 8-stage pipeline with middleware hooks for fetch, score, deduplication, LLM-based editing, rendering, and email delivery. - Adds design patterns: Adapter, Strategy, Pipeline, Observer/Middleware, Registry, Factory, and Dataclass, for robust extensibility. - Provides clear guides to extend sources, editors, presets, and templates. - Includes quality control (auto validation/retry), elegant report rendering (HTML/PDF), and SMTP/email integration. - Now with bilingual documentation (English & 中文) for broader accessibility.

元数据

Slug lunaclaw-brief

版本 4.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 3

常见问题

Ai Cv Weekly 是什么？

Pluggable AI report engine generating multi-preset, multi-source briefings with smart scoring, deduplication, LLM editing, quality control, rendering, and em... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 255 次。

如何安装 Ai Cv Weekly？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install lunaclaw-brief」即可一键安装，无需额外配置。

Ai Cv Weekly 是免费的吗？

是的，Ai Cv Weekly 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Ai Cv Weekly 支持哪些平台？

Ai Cv Weekly 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Ai Cv Weekly？

由 llx9826（@llx9826）开发并维护，当前版本 v4.0.0。

Ai Cv Weekly