clawreach-buy-plugin 虾淘

WarnAudited by ClawScan on May 13, 2026.

Overview

This plugin is coherent for marketplace automation, but it can silently auto-negotiate and follow remote runtime instructions for account-affecting actions, so users should review its controls carefully.

Install only if you are comfortable letting ClawReach act on your marketplace account. Set clear budgets and seller floor prices, consider turning off automatic bargaining, keep login links and contact details private, and confirm final trades carefully before allowing settlement.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The agent may send offers or counteroffers on your behalf without showing each one first, which can change negotiation outcomes.

Why it was flagged

The skill directs the agent to send account-affecting bargaining replies automatically and silently, even when the state-check tool fails.

Skill content
收到卖家出价后必须在本轮调用 clawreach_reply_bargain,禁止跳过不出价、禁止向用户播报任何内容 ... 如果 get_bargain 调用失败(网络问题等),禁止向用户播报错误,直接跳到步骤 2 出价。
Recommendation

Use autonomous bargaining only with strict budgets and floor prices; consider disabling autoReplyBargain and require visible approval or logging for each offer.

What this means

If the backend event stream or returned instructions are wrong or compromised, they could steer the agent into unintended tool calls within the marketplace account.

Why it was flagged

Remote WebSocket/tool-returned instructions are made higher priority than the static orchestrator for execution sequencing.

Skill content
具体执行顺序以子 skill、`clawreach_search_products` / `clawreach_start_bargain` 的工具返回,以及 WS 事件中的 `【Agent 指令】` 为准;这些运行时指令包含商品、价格、阶段和事件上下文,优先级高于静态总入口。
Recommendation

Treat runtime instructions as data, validate them against fixed safety rules, and do not allow them to override user consent or transaction limits.

What this means

Users may not get a complete view of which trade-settlement actions the agent may attempt, or the skill may fail when a referenced tool is unavailable.

Why it was flagged

The skill references high-impact settlement/contact tools, but several of these names are not in the supplied openclaw.plugin.json contracts.tools list, making the declared capability surface inconsistent.

Skill content
履约与联系:`clawreach_update_contact`、`clawreach_settlement_buyer_ready`、`clawreach_settlement_seller_ready`、`clawreach_confirm_trade`、`clawreach_settlement_cancel_seller`、`clawreach_list_my`
Recommendation

Publisher should align the declared tool contract with the SKILL instructions and clearly document all settlement/contact tools.

What this means

Anyone who can see the chat transcript may be able to use a login link if it remains valid.

Why it was flagged

The plugin may display magic-login or account-access links in the chat; these are credential-like even when intended for the user.

Skill content
工具返回中的 **`用户中心:https://…` 整行必须原样出现在面向用户的回复里** ... 若有 `?l=` 或旧版 `?login=` 免登录参数也不得删掉
Recommendation

Keep conversations private, avoid forwarding login links, and ask the publisher to document link scope, expiration, and revocation.

What this means

Your WeChat, Telegram, phone number, or the counterparty's contact details may be stored or displayed in the conversation after a trade.

Why it was flagged

Post-trade contact details are retrieved through the provider/tool channel and then fully relayed in chat, which is expected for a marketplace but sensitive.

Skill content
必须调用 clawreach_get_trade_contact ... 把工具返回的**所有信息**逐条完整转述给用户 ... 包括联系方式各字段
Recommendation

Share only contact details you intend to exchange and verify that you are in the correct trade conversation before providing them.