Tencent Fact Check

What it solves

Fact checking is often treated as a one-line search, but the practical problem is harder: users may provide a claim, article, chat log, or screenshot and want a sourced verification result, not a model rewrite based on memory. This skill connects Tencent’s fact-check capability to an agent workflow. Its core task is not to generate an explanatory article, but to call the fact-checking ability in tencent-news-cli and present the CLI’s structured Markdown output directly to engineers.

How it works

• Environment readiness: It first checks CLI state with cli-state, inspecting fields such as cliExists, update.needUpdate, and apiKey.status to decide whether to install, update, or configure the API key. Setup follows the provided guides and does not automatically open a browser.
• Fact-check execution: Except for cli-state, commands are run through run-cli. The agent must execute help jiaozhen first and then choose the matching command and arguments from the current help output. It does not hard-code business commands or guess where --jiaozhen belongs.
• Input handling: Short claims can be checked directly. Long text prefers whole-content checking when available; otherwise it extracts 1–3 core verifiable claims. Images and screenshots are first parsed with multimodal vision to extract text and key facts, then checked through the CLI.
• Result presentation: The final output must be based on the CLI’s Markdown. It preserves the conclusion, verification process, confidence assessment, source numbers, titles, and links, without summarizing, rewriting, or inventing links.

Boundaries and notes

This skill depends on the live Tencent fact-check CLI, and commands may change across versions. If the CLI returns empty output, a non-zero exit code, a timeout, or an error, it should be treated as a same-day quota limit. The agent should point users to the official fact-checking website instead of retrying, using WebSearch, or substituting other sources.