geeknews-bot

Things to consider before installing or running this skill: - Required credentials are not declared in the registry metadata. The runtime needs TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID (mandatory) and the run script requires ANTHROPIC_API_KEY (it will exit if missing). It also recommends GITHUB_TOKEN to avoid rate limits. Don't run the cron script unless you intend the pipeline to call Anthropic. - Implementation mismatch: SKILL.md describes scraping/RSS fallback, but the provided automation delegates to Anthropic's API (claude) with a web_search tool. If you prefer local scraping (so content never leaves your host), you'll need to implement that yourself or modify run_geeknews_brief.sh. Running as-is will transmit prompts and rely on Anthropic; evaluate whether that data flow (site URLs and prompt content) is acceptable. - Network endpoints used: api.anthropic.com (if ANTHROPIC_API_KEY set) and api.telegram.org. The Telegram bot token must be protected; any compromise of that token would allow message sends as your bot. Keep tokens in a secure place and avoid committing them to repo. - Review message contents: the script sends the full generated markdown via send_telegram.py. Ensure no sensitive information is included in the generated brief or prompt context (the prompt instructs the model to "check the main page" — the model/tool may retrieve and process site content). - Actionable steps: ask the publisher to update registry metadata to list required env vars; if you don't trust remote LLM calls, modify the pipeline to perform local RSS parsing/scraping and GitHub queries yourself; store tokens in a secure secret store; test send_telegram.py manually with a test bot/chat before enabling cron. Overall, the skill is not evidently malicious but has clear inconsistencies and a surprising dependency on an external LLM for the automated pipeline — treat it cautiously and correct the manifest and/or implementation before enabling automated runs.