competitor-recon
以投诉为驱动的竞品调研：列出竞争对手，提取功能、定价、定位以及来自评论、论坛和问题跟踪器的用户投诉，整理投诉类别，并交付包含可借鉴项和应跳过项列表的对比表。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:ChenneyZhuang/competitor-recon#26ff46f91298d8746dce84c610b9b2dd48bef270Competitor Recon 以 competitor-recon 发布，当前版本为 0.2.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Run competitor reconnaissance before building any feature: the review sections of existing products are the cheapest requirements research that exists.
动手做功能之前先跑竞品侦察：现有产品的用户评论区，是世界上最便宜的需求调研。
Marketing pages state intentions; user reviews state outcomes. One grounded complaint is worth more than any amount of feature-page copy — and a complaint repeated by three independent users is an unmet need you can build against.
营销页写的是意图，用户评论写的是结果。一条有出处的真实抱怨，胜过整页功能文案；三个互不相识的用户抱怨同一件事，就是一个可以动手满足的未满足需求。
Sources actually fetched during the run: V2EX threads (via curl), the target's GitHub issue tracker (via API — reaction counts double as frequency statistics), community forums.
| Complaint cluster | Evidence | Verdict |
|---|---|---|
| Ads / subscription inflation driving users out | multiple independent forum posts naming the incumbents | differentiate: stay clean, price predictably |
| "AI auto-capture might be wrong, I re-check manually" | feature request with 👍×3 in the target's own tracker | build: recognition → confirmation loop |
| Reimbursement / refund flows too complex | complaints across two products, praise for a third's | build: expense-return workflow done simply |
| Multi-currency locked behind a paywall | a long-time user defected over exactly this | skip: no demand evidence beyond the wall itself |
The run produced the full comparison table plus a skip list with reasons — 3 build ideas, each cited; 3 skips, each justified.
git clone https://github.com/ChenneyZhuang/competitor-recon ~/.claude/skills/competitor-recon
One SKILL.md. Needs only web search and any fetch client. MIT. v0.2.0 — live-tested end-to-end.
单个 SKILL.md，只需搜索与抓取能力。MIT 许可，v0.2.0，完整实测通过。