MAG Community Intelligence Agent

Autonomous OpenClaw agent that transforms high-volume community discussions into anonymized, human-readable intelligence reports.

What it solves

Community conversations contain useful context, disagreement, and audience signals, but they are difficult to review at scale. This project turns permitted events into an aggregate report while preserving uncertainty and privacy. It is an automation and analysis portfolio project — not an investment advisory product.

Key features

• Deterministic, credential-free Demo Mode with 45 synthetic messages.
• Reply-context reconstruction and canonical entity recognition.
• Topic, sentiment, argument-quality, disagreement, and emotional-response analysis.
• Aggregate-only Markdown reporting with no raw message or participant output.
• OpenClaw agent and custom skill contracts.
• Schedule example, checkpoint behavior, security gate, tests, and CI.

Architecture

flowchart TD
    A[Permitted Community Source] --> B[Collector]
    B --> C[Reply Context Reconstruction]
    C --> D[Entity and Topic Recognition]
    D --> E[Sentiment and Argument Analysis]
    E --> F[Anonymized Intelligence Report]
    F --> G[Optional Delivery]
    G --> H[Checkpoint]
    I[OpenClaw Schedule] --> B

Loading

Demo

python scripts/demo.py

The command uses only synthetic demo data and writes the current report. See an example report.

OpenClaw integration

The public agent contract calls the custom community intelligence skill after data ingestion. The repository deliberately ships no Telegram authorization, source identifier, session, or delivery configuration. See OpenClaw setup.

Trigger / cron automation

config/schedule.example.yaml shows a disabled example schedule in Europe/Moscow. A deployment should enable it only after configuring an authorized source, private runtime storage, and a delivery policy. See cron example.

Privacy and security

The project contains no real conversations, participant identities, credentials, sessions, or runtime state. Demo Mode is offline and deterministic. Read Privacy and Security before connecting any real community.

Quick start

Requires Python 3.11+; no third-party packages are required for Demo Mode.

python scripts/validate_project.py
python -m unittest discover -s tests -v

Tests

Tests cover fixture loading, reply chains, aliases, positive/negative/mixed sentiment, emotional versus reasoned signals, disagreement, off-topic handling, privacy-safe reporting, and the secret scan. GitHub Actions runs these checks without repository secrets.

Project structure

src/        deterministic public pipeline
skills/     custom OpenClaw skill
data/       synthetic demo fixture
config/     safe examples only
docs/       architecture, privacy, portfolio, and submission material
scripts/    demo, validation, and secret scanner
tests/      standard-library test suite

Real-world use cases

• Community operations and moderation insight.
• Product-feedback and release-discussion analysis.
• Support-community theme and disagreement tracking.
• Research workflows that need privacy-aware aggregate reporting.

Roadmap

See ROADMAP.md for multi-community, RSS, storage, cross-source, and deployment extensions.

Disclaimer

This project describes community discussion signals. It does not verify facts independently and does not provide financial, legal, medical, or investment recommendations.

Русская краткая версия

Это автономный агент OpenClaw для анонимизированного анализа обсуждений: темы, сущности, настроения, аргументы и расхождения во мнениях. Репозиторий использует только синтетические данные; Demo Mode запускается локально без Telegram и ключей.