Building AI agent skills & automation tools — I turn repetitive daily workflows into skills that agents can run on their own: read-only by default, evidence-first, with hard guardrails instead of vibes.
Omni Scroll — 主流平台万能刷流 An agent skill that scrolls the mainstream platforms for you once a day: searches the last 7 days by keyword, reads the full text / subtitles of every item before judging it, and writes one cross-platform daily brief. Covers Bilibili, Xiaohongshu, Douyin, Kuaishou, Weibo, Tieba, Zhihu, YouTube and X — read-only, injection-resistant, distilled from dsh-bili-scroll.
tcping-sweep — 全国 31 省三网 TCP-Ping 批量测速 Zero-dependency Python tool that TCP-pings test nodes in all 31 Chinese provinces across the three major carriers (93 endpoints) and reports median connect latency per node.
- Agent skills — reusable, host-agnostic playbooks (Muse / Muse / Hermes style runtimes)
- Content intelligence — crawlers + LLM triage pipelines that read originals, not summaries
- Self-hosted tooling — small utilities for networks, proxies and automation
- Read the source, then judge — never conclude from a headline
- Whitelist gates at the capability layer, not willpower in the prompt
- Ship the artifact, report the evidence
Open to feedback and PRs on any public repo here.