中文 | English

---

The Problem

You're tracking the AI coding tools space. Every day you check Hacker News, GitHub Trending, Reddit, X, Product Hunt… a dozen sources, an hour of scanning, mostly noise, and you still almost miss that one critical signal.

Signex does this for you. Describe what you care about in one sentence, and it automatically collects from 15+ data sources, deduplicates, analyzes, and delivers a report with actionable insights. You read the conclusions, give feedback, and it learns what matters to you.

Who Is This For?

• Indie developers — tracking product opportunities, competitors, tech trends
• Startup founders — discovering unmet needs, validating product direction
• Tech leads — following industry direction, evaluating new tools and frameworks
• Product managers — monitoring user feedback, feature requests, market signals
• Investors / analysts — tracking market dynamics, spotting early signals
• Content creators — catching trending topics, sourcing writing material
• Researchers — continuously following developments in a specific domain

What is Signex?

Signex is your AI intelligence analyst. You define what you care about (a "Watch"), and it autonomously collects data from multiple sources, analyzes it through different lenses, and delivers actionable reports. It remembers your feedback and adjusts future analysis accordingly.

Architecturally, Claude Code IS the runtime. There is no standalone app, server, or CLI wrapper. The agent's behavior is defined entirely in CLAUDE.md, and its capabilities are modular skills in .claude/skills/. You interact with it by talking to Claude Code.

Core Concepts

Concept	What it does
Watch	A continuous monitoring intent. Defines what direction to watch and what signals matter.
Sensor	Data collection probes. Each sensor fetches from a specific source — Hacker News, GitHub, Reddit, search APIs, RSS, etc.
Lens	Analysis perspectives. Choose how to look at the data — deep insight, quick brief, pro/con evaluation, or timeline trace.
Vault	Cross-watch insight storage. Valuable findings that transcend individual watches get deposited here.

Architecture

%%{init: {
  "theme": "base",
  "flowchart": { "curve": "basis", "nodeSpacing": 42, "rankSpacing": 60 },
  "themeVariables": {
    "background": "transparent",
    "mainBkg": "transparent",
    "fontFamily": "ui-sans-serif, system-ui, -apple-system, Segoe UI, Roboto, Helvetica, Arial",
    "fontSize": "14px",
    "lineColor": "#94A3B8",
    "textColor": "#0F172A"
  }
}}%%

flowchart TB
  UA["You · shapes & calibrates"]:::actor
  UA -. "watch-shape · feedback · calibrate" .-> W
  UA -. "identity-shape · shape profile" .-> ID

  ID["Identity<br/>identity.md"]:::identity -. "context" .-> W

  W["Watches<br/>intent.md + memory.md"]:::core --> S["Sensor Selection"]:::core

  subgraph Sensors["Sensors (extensible)"]
    direction TB
    S --> S1["Hacker News"]:::sensor
    S --> S2["GitHub Trending"]:::sensor
    S --> S3["Tavily / Brave / Exa"]:::sensor
    S --> S4["Reddit / X / V2EX"]:::sensor
    S --> S5["Product Hunt / RSS / …"]:::sensor
    S --> S6["arXiv / OpenAlex"]:::sensor
    S --> S7["+ Your Own Sensor"]:::sensorAlt
  end

  S1 & S2 & S3 & S4 & S5 & S6 & S7 --> DB[("SQLite")]:::store
  DB --> L{"Lens"}:::decision

  subgraph Lenses["Lenses (extensible)"]
    direction TB
    L --> L1["Deep Insight"]:::lens
    L --> L2["Flash Brief"]:::lens
    L --> L3["Dual Take"]:::lens
    L --> L4["Timeline Trace"]:::lens
    L --> L5["+ Your Own Lens"]:::lensAlt
  end

  L1 & L2 & L3 & L4 & L5 --> R["Reports & Alerts"]:::out
  R --> UB["You · reads & iterates"]:::actor
  UB -. "cross-watch insights" .-> V["Vault"]:::vault

  classDef core      fill:#EEF2FF,stroke:#4F46E5,stroke-width:2px,color:#0F172A,rx:14,ry:14
  classDef identity  fill:#F5F3FF,stroke:#7C3AED,stroke-width:1.6px,color:#0F172A,rx:12,ry:12
  classDef sensor    fill:#ECFEFF,stroke:#06B6D4,stroke-width:1.6px,color:#0F172A,rx:12,ry:12
  classDef sensorAlt fill:#F0FDFA,stroke:#14B8A6,stroke-width:1.6px,color:#0F172A,stroke-dasharray:4 3,rx:12,ry:12
  classDef store     fill:#F1F5F9,stroke:#64748B,stroke-width:1.7px,color:#0F172A,rx:14,ry:14
  classDef decision  fill:#FFFFFF,stroke:#0EA5E9,stroke-width:2px,color:#0F172A,rx:18,ry:18
  classDef lens      fill:#FFF7ED,stroke:#F59E0B,stroke-width:1.6px,color:#0F172A,rx:12,ry:12
  classDef lensAlt   fill:#FFFBEB,stroke:#F59E0B,stroke-width:1.6px,color:#0F172A,stroke-dasharray:4 3,rx:12,ry:12
  classDef out       fill:#ECFDF5,stroke:#10B981,stroke-width:1.8px,color:#0F172A,rx:14,ry:14
  classDef vault     fill:#FDF4FF,stroke:#A855F7,stroke-width:1.6px,color:#0F172A,rx:14,ry:14
  classDef actor     fill:#FFF1F2,stroke:#FB7185,stroke-width:1.6px,color:#0F172A,rx:14,ry:14

  style Sensors fill:transparent,stroke:#CBD5E1,stroke-width:1.2px,rx:16,ry:16
  style Lenses  fill:transparent,stroke:#CBD5E1,stroke-width:1.2px,rx:16,ry:16

  linkStyle default stroke:#94A3B8,stroke-width:1.5px

Quick Start

Prerequisites

• Python 3.11+
• uv (Python package manager)
• Claude Code (the CLI)

Setup

# Clone the repo
git clone https://github.com/zhiyuzi/Signex.git
cd signex

# Install dependencies
uv sync

# Configure API keys (at minimum, set one search API key)
cp .env.example .env
# Edit .env with your API keys

# Start Claude Code in the project directory
claude

First Run

You:  Hi

Signex initializes automatically on first greeting — creates your profile, watch templates, and vault. Then it gives you a situational briefing.

You:  Help me watch AI coding tools — new IDEs, agent features, community reactions.

Signex creates a new Watch with your intent, picks relevant sensors, and is ready to run.

You:  Run it.

Sensors fire, data flows into SQLite, the lens analyzes, and you get a report.

Skills

Sensors (data collection)

Skill	Source	API Key	Get Key
fetch-hacker-news	Hacker News front page & search	—
fetch-github-trending	GitHub Trending repos	—
fetch-v2ex	V2EX (Chinese tech community)	—
fetch-reddit	Reddit posts & search	—
fetch-rss	Any RSS/Atom feed	—
fetch-tavily	Tavily web search	Yes	tavily.com
fetch-brave-search	Brave Search	Yes	brave.com
fetch-exa	Exa AI semantic search	Yes	exa.ai
fetch-product-hunt	Product Hunt launches	Yes	producthunt.com
fetch-request-hunt	RequestHunt feature requests	Yes	requesthunt.com
fetch-news-api	NewsAPI.org	Yes	newsapi.org
fetch-gnews	GNews	Yes	gnews.io
fetch-x	X / Twitter search	Yes	developer.x.com
fetch-arxiv	arXiv preprints	—
fetch-openalex	OpenAlex academic papers	Yes	openalex.org
extract-content	Extract full article text from URLs (supplements sensor data when needed)	—

Lenses (analysis)

Skill	Purpose
lens-deep-insight	Comprehensive analysis — key findings, trends, action items (default)
lens-flash-brief	3–5 bullet quick summary
lens-dual-take	Pro/con evaluation of a topic
lens-timeline-trace	Event timeline reconstruction

Database

Skill	Purpose
db-save-items	Store sensor data (auto-dedup)
db-query-items	Query items by watch, source, time
db-save-analysis	Record analysis runs
db-stats	Run history & statistics
db-source-health	Data source health monitoring

Shaping

Skill	Purpose
identity-shape	Conversational user profile shaping & iteration
watch-shape	Conversational watch cognitive structure design & iteration

Pipeline

Skill	Purpose
run-watch	Execute full watch cycle (collect → analyze → report)
save-report	Write reports and alerts to disk
update-memory	Integrate user feedback into watch memory
webhook-notify	Push report summary to IM tools (Feishu, Discord, WeCom, etc.)

Setup

Skill	Purpose
setup	Project initialization (directories, templates, database)
webhook-setup	Interactive webhook configuration wizard
skill-creator	Guide for creating new skills

Project Structure

signex/
├── CLAUDE.md                  # Agent behavior definition (the brain)
├── .claude/skills/            # All skills (sensor, lens, db, action)
├── profile/identity.md        # User identity & preferences
├── watches/                   # Watch definitions
│   ├── index.md               # Watch registry
│   └── {watch-name}/
│       ├── intent.md          # What to monitor
│       ├── memory.md          # Feedback memory & cognitive evolution
│       └── state.json         # Run state
├── vault/                     # Cross-watch insights
│   ├── index.md               # Vault index
│   └── *.md                   # Individual insight notes
├── reports/{date}/{watch}/    # Analysis output
├── alerts/{date}/             # High-signal alerts
├── knowledge/                 # Skill knowledge base (reference docs)
├── data/signex.db             # SQLite database
├── src/                       # Python scripts (HTTP + SQLite only)
└── .env                       # API keys (not committed)

License

Copyright (c) 2026 Li Ze

This project is licensed under the GNU Affero General Public License v3.0.

You are free to use, modify, and distribute this software under the terms of the AGPL-3.0. If you run a modified version as a network service, you must make the source code available to its users.