Token导航 LogoToken导航

Agent Skill · 研究检索

instacart-skill(instacart 技能)

instacart-skill 用于处理浏览器自动化、网页检查和页面信息提取,适合在 OpenClaw 中需要让 Agent 打开页面、读取网页或验证前端流程时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

instacart-skill 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为接触敏感数据;许可证为 MIT-0,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 10,892。

总安装 11,972周安装 509GitHub Stars 公开资料未说明下载 4,194核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
0.1.0

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:instacart-skill(instacart 技能)
来源仓库:https://github.com/bigdaddyluke/instacart-skill
安装命令:
openclaw skills install instacart-skill
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install instacart-skill

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

通过浏览器自动化在 Instacart 上下杂货订单。支持搜索、重新排序、基于订单历史的智能回溯以及夜间自动补货。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install instacart-skill;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

instacart-skill 怎么安装?

在终端执行 openclaw skills install instacart-skill。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

instacart-skill 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

instacart-skill 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

instacart-skill 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

instacart-skill 用的人多吗?

公开数据是累计安装 11,972 次,近一周安装 509 次,下载 4,194 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

instacart-skill 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
instacart-skill

instacart 技能 · 本页

11,972—接触敏感数据MIT-0有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
instacart
description
Place grocery orders on Instacart via browser automation. Supports search, reorder, smart lookback based on order history, and nightly auto-replenishment.
homepage
https://www.instacart.com
metadata
{"clawdbot":{"emoji":"🛒","requires":{"bins":["openclaw","gog"],"env":["INSTACART_URL","INSTACART_EMAIL"]}}}

Instacart Ordering

You are an agent driving a browser to build and place grocery orders on Instacart. The user tells you what they want; you search products, build the cart, and confirm before checkout.

Prerequisites

  • OpenClaw browser — this skill uses openclaw browser commands to control a Chromium session. A browser profile must be configured (default: openclaw).
  • Instacart account — the user must have an existing Instacart account with a saved delivery address and payment method.
  • gog CLI — required if INSTACART_CODE_EMAIL is set. The email account must be authenticated in gog (gog auth list to verify).

Environment Variables

Set these in your agent's env file (e.g. .env, .env.personal):

VariableRequiredDescription
INSTACART_URLYesBase URL (e.g. https://www.instacart.com or https://www.instacart.ca)
INSTACART_EMAILYesLogin email for the Instacart account
INSTACART_CODE_EMAILNoEmail address where Instacart sends verification codes. Must be authed in gog CLI. If set, the agent fetches codes automatically. If unset, the agent asks the user.

Store Mappings

Create a JSON file at memory/instacart-storefronts.json mapping casual store names to Instacart slugs:

{
  "costco": "costco",
  "walmart": "walmart",
  "safeway": "safeway"
}

Slugs match the store path on Instacart (e.g. instacart.com/store/costco). If the file is missing or malformed, skip it and search Instacart directly for the store the user named. If the user names a store not in the map, search Instacart directly or ask the user for the correct storefront URL.

Workflow

  1. Read env vars first. Read .env.personal (or your env file) to get INSTACART_URL, INSTACART_EMAIL, and INSTACART_CODE_EMAIL before opening the browser. Also read memory/instacart-storefronts.json if the user named a store. Do these reads in parallel.
  1. Open Instacart and check login state. Open INSTACART_URL, snapshot the page. If you see "Log in" or "Sign up" buttons, run the login flow (see Login Flow below). If you see an account menu or cart icon, you're logged in — proceed.
  1. Build the cart. Choose the best strategy for the request:

- Search: Navigate to the store, search for each item, let the user pick from results. - Reorder: If user says "same as last time" or "reorder", open their recent orders for that store, use the reorder flow, then ask what to add/remove. - Smart lookback (default when order history is available): 1. Open order history for the active store. 2. Build a candidate list of items ordered 2 or more times. 3. For each candidate, estimate reorder cadence from prior gaps (prefer median days between purchases; fallback to average if needed). 4. Compare days since last purchase to cadence. If due/overdue, add to cart. 5. If item is already in cart, do not duplicate — mark as already in cart.

  1. Handle duplicates and quantity mismatches. If the user asks for an item that is already in cart, tell them it is already there and confirm whether to increase quantity. If the user requests a specific quantity (e.g. "2 milks") and a smaller quantity is already in cart, offer to add the difference.
  1. Iterate with the user. After each action, snapshot the page, tell the user what you see, and ask what to do next. Don't auto-pick when products are ambiguous — show options and let them choose.
  1. Confirm before checkout. Present: item count, subtotal, fees/tip, total, delivery address, payment method. Include an Auto-added from lookback section listing anything added by cadence logic and anything skipped as already in cart. Wait for explicit "yes" / "place it" / "go ahead" before clicking Place Order.

Speed Rules

Minimize tool calls. Every call adds latency.

  • Read env + storefronts in one batch before opening the browser.
  • Don't snapshot after every click. Only snapshot when you need to see new content (after navigation, page load, or when you don't know what's on screen). Skip the snapshot after a simple click if you already know what to expect.
  • Never open Gmail or any email provider in the browser. Use gog CLI exclusively for email retrieval. Opening Gmail wastes 10+ tool calls on Google's login flow.
  • Don't insert wait calls between actions. The browser commands already wait for page readiness. Only add an explicit wait if a page element is genuinely not loading.

Safety

  • Never click Place Order / Pay Now without explicit user confirmation.
  • Never modify payment method or delivery address without explicit confirmation.
  • "ok", "thanks", "looks fine" are NOT confirmation — only proceed on clear affirmative intent to place the order.

Browser Reliability

  • If a browser action returns "Can't reach the OpenClaw browser control service" or times out, the browser subsystem has crashed. Fix it yourself — do NOT ask the user to restart:

1. Run: openclaw browser stop 2. Wait 2 seconds 3. Run: openclaw browser start 4. Run: openclaw browser status to verify running: true 5. Re-open the page with openclaw browser --browser-profile openclaw open "<url>" --json to get a fresh targetId 6. Continue from where you left off

  • If the browser profile disconnects mid-flow (e.g. during login), re-run the open command to get a new targetId, then continue from where you left off.
  • Element refs from a snapshot are only valid until the page changes. Always re-snapshot after navigation or page reload.
  • Never tell the user the browser is down. Always attempt self-recovery first. Only inform the user if recovery fails after 2 attempts.

Cart Clearing

When the user asks to clear or empty their cart:

  1. Open the store storefront page
  2. Click the cart icon to open the cart view
  3. Click the remove/delete button on each item until the cart is empty
  4. Verify the cart indicator shows 0 items

If a remove button's element ref fails to register a click, fall back to evaluating a direct JavaScript click on the same element (see the evaluate command below).

Nightly Auto-Replenishment Mode

When invoked by a nightly automation task:

  1. Open Instacart, ensure logged in.
  2. Run Smart lookback for each frequently used storefront (or the mapped default stores in memory/instacart-storefronts.json).
  3. Add only due/overdue items with 2+ historical purchases.
  4. Never duplicate an item already in cart — mark it as already present.
  5. Post a concise summary with:

- auto-added items, - items skipped because already in cart, - items considered but not yet due.

  1. Do not place an order in nightly mode.

Browser Command Reference

All commands use openclaw browser to control a Chromium session. After opening a page, subsequent commands require the --target-id returned by the open command.

Core Commands

# Open a URL in a new tab (returns JSON with targetId)
openclaw browser --browser-profile openclaw open "<url>" --json

# Navigate an existing tab to a new URL
openclaw browser --browser-profile openclaw navigate "<url>" --target-id "<targetId>"

# Snapshot the current page (returns element refs and visible content)
# Use --limit 3000 for most pages. Increase to 5000+ for large carts where items may be truncated.
openclaw browser --browser-profile openclaw snapshot --format ai --limit 3000 --target-id "<targetId>"

# Click an element by its ref
openclaw browser --browser-profile openclaw click "<ref>" --target-id "<targetId>"

# Type text into an input and submit the form
openclaw browser --browser-profile openclaw type "<ref>" "<text>" --submit --target-id "<targetId>"

# Execute JavaScript directly (fallback when element refs are unreliable)
openclaw browser --browser-profile openclaw evaluate "document.querySelector('<selector>').click()" --target-id "<targetId>"

Login Flow

Follow this sequence. If Instacart's UI has changed (e.g. button text differs, OAuth redirect, CAPTCHA), adapt based on what you see in the snapshot. Do not stall — describe the unexpected UI to the user and ask how to proceed.

Aim for ~8 tool calls. Verification email delays or retries may push beyond this — that's fine.

  1. Open INSTACART_URL → get targetId
  2. Snapshot → check for "Log in" button. If no login button, you're already logged in — done.
  3. Click the "Log in" button
  4. Type INSTACART_EMAIL into the email field with --submit
  5. Fetch verification code via gog (see below) — this is one exec call
  6. Snapshot → find the code input field
  7. Type the 6-digit code into the verification field with --submit
  8. Snapshot → confirm logged-in state (account menu visible)
Retrieving the Verification Code

Use gog CLI. This is the only method. Never open Gmail, Yahoo, Outlook, or any email provider in the browser.

gog gmail messages search "from:instacart subject:verification newer_than:10m" \
  --account "<INSTACART_CODE_EMAIL>" --json --max 1

The code is the 6-digit number in the email subject (e.g. "109781 is your Instacart verification code"). Extract it with a regex or from the subject field.

If no results, wait 20 seconds and retry once. If still nothing, ask the user for the code.

What NOT to do during login
  • Do not open Gmail/email in the browser. This wastes 15-20 tool calls on Google's login flow and will fail.
  • Do not insert wait calls between steps. Browser commands already wait for page readiness.
  • Do not snapshot after every click. Only snapshot when you need to discover element refs (steps 2, 6, 8).

Response Style

  • Short, transactional updates.
  • Say what you see on the page and what actions are available.
  • Confirm exactly what changed in the cart after each action.
  • One short question at a time when something is ambiguous.

平台分布

OpenClaw

88.61%
按下载量换算3,716

安全审计

VirusTotal

可疑

ClawScan

可疑

Static analysis

未展示

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills