liufeng1976/bossai-intelligence-brief-skill

BossAI Intelligence Brief

Use this Skill when the user asks to:

• collect the latest items from configured public sources;
• generate a morning or daily intelligence brief;
• reduce Hacker News, ArXiv, Reddit, RSS, or Atom feeds into a short reading list;
• classify information into MUST_READ, QUICK_READ, and SKIP;
• extract evidence-backed content angles from the highest-ranked items;
• validate or troubleshoot the intelligence source configuration;
• optionally push a generated brief through a user-configured webhook.

Resolve the Skill home

Resolve the project home in this order:

1. read skillHome from config.json beside this file;
2. use environment variable BOSSAI_INTELLIGENCE_SKILL_HOME;
3. use the repository root when this Skill is running directly from the cloned repository;
4. otherwise report that scripts/install-skill.mjs must be run.

Never guess a project path.

Required first-run verification

Before a live scan:

1. run npm test in the Skill home;
2. run the fixture demo;
3. inspect outputs/demo/manifest.json and confirm ok=true;
4. validate the user's live config;
5. only then run a live collection.

Commands:

npm test
npm run demo
node bin/bossai-intelligence-brief.mjs validate --config config/sources.example.json

Generate a brief

node bin/bossai-intelligence-brief.mjs run \
  --config config/sources.local.json \
  --output outputs/latest

Add --push only when:

• notification.enabled=true in the selected config;
• the user explicitly configured the webhook environment variable;
• the user asked for delivery.

Never print the webhook URL or any API key.

Read the result

Prefer brief.json for Agent reasoning and brief.md for user delivery.

The Agent should report:

• number of configured, successful, and failed sources;
• MUST_READ and QUICK_READ counts;
• the top items with source URLs;
• whether AI review was actually used;
• any source failures;
• where the output files were written.

Do not claim AI review was used when meta.llm.used=false.

Truthfulness rules

• Public posts and articles are source evidence, not verified customer orders or guaranteed market demand.
• Preserve the original source URL for every recommendation.
• Clearly separate source facts from the Agent's inference.
• Never invent engagement, revenue, budgets, market size, quotes, or publication dates.
• Never hide failed feeds to make the run look healthier.
• Demo fixture data must remain labeled as demo data.

Safety boundaries

This Skill is read-only toward source platforms. It must not:

• log in to social platforms;
• bypass access controls;
• publish content;
• send customer messages;
• purchase, pay, refund, delete, or modify external data;
• scrape private pages or feeds without explicit authorization.

Licensing

This source-available Skill is free for personal non-commercial use. Commercial operation, enterprise internal use for profit, paid services, SaaS integration, white-label delivery, or paid redistribution requires written authorization from 刘风 / BossAI.