“multi-step”的搜索结果

将规格或需求整理成实施计划，明确要修改的文件、函数定义、测试和验证步骤。在编写代码前拆分工作，形成可交给其他人员或代理执行的任务。

• 编写代码前规划功能实施
• 将需求拆成可测试的任务
• 向其他人员或代理交接实施工作

运用公开信息调查方法，将照片、视频和公开记录中的零散线索串联起来，核实地点、时间及组织关系，区分证据与假设，并整理反证和未确认事项。

• 核实公共场所照片的拍摄位置
• 从公开记录重建事件时间线
• 调查企业与域名的关系

让AI智能体按准确性、完整性、清晰度、可操作性和简洁性五个维度自评成果，给出有具体依据的分数、平均分，以及按优先级排列的改进建议。

• 多文件实现完成后的自评
• 检查多个工作阶段的最终成果
• 评估设计文档或分析报告

通过正则表达式和语法规则约束本地语言模型的输出，生成符合指定格式的JSON、日期和固定分类标签，并用Python构建可重复使用的多阶段生成流程。

• 生成符合指定结构的JSON
• 统一日期和电子邮箱地址格式
• 使用固定标签对文本分类

将代码修改和多步骤开发任务交给OpenHands CLI，选择LiteLLM支持的模型，无需交互界面即可单次或批量运行任务、查看进度并恢复历史会话。

• 将代码修改交给OpenHands
• 使用LiteLLM支持的模型开发
• 查看长时间开发任务的进度

在AI的建议、计划、估算或分析后附上两三个简短问题，帮助用户在行动前检查数字、逻辑与缺失信息。问题紧扣回答中的具体内容，可直接向代理追问。

• 检查费用与交付时间估算的前提
• 审视推荐计划的依据
• 检查提案或邮件草稿缺失的信息

为功能开发或多阶段开发任务调查官方资料和现有实现示例，从用户可见的完成目标出发，整理各阶段的工作、参考位置和验证方法，形成实施计划。

• 确定新增功能的实施步骤
• 实施前核实API的文档依据
• 将开发工作分到多个对话

Design and build AI agents for business, research, operations, or creative work. Define essential actions and knowledge, start with a simple execution loop, and add progress tracking or separate child agents as needs emerge.

中文简介正在准备中。当前显示英语简介。

• Designing an assistant for business tasks
• Building a research agent
• Choosing essential actions and knowledge

Fill web forms from a JSON list of field names and values, using saved templates or page labels to locate inputs. Optionally submit the form, save new field mappings, and redact values from recorded actions.

中文简介正在准备中。当前显示英语简介。

• Filling signup and contact forms
• Reusing templates for recurring submissions
• Saving field mappings for a new form

Automate Anthropic Admin tasks involving API keys, usage, workspaces, and organization management through Rube MCP, a connection to external tools. The skill guides tool discovery, connection checks, multi-step workflows, and bulk operations.

中文简介正在准备中。当前显示英语简介。

• Manage Anthropic Admin API keys
• Check Anthropic usage
• Handle workspace and organization management tasks

Review multi-step forms to find information users must enter more than once. Suggest automatic filling, reuse options, and preserving data between steps, while explaining relevant accessibility requirements and exceptions for security or validity.

中文简介正在准备中。当前显示英语简介。

• Checking repeated fields during checkout
• Reviewing signup and onboarding forms
• Reusing information in support flows

Adapt agent replies for readers with ADHD by putting the next action first, numbering steps, and restating progress across turns. Replies include concrete time estimates, visible completed work, and fewer tangents to make tasks easier to follow.

中文简介正在准备中。当前显示英语简介。

• Make the first action clear when starting a task
• Follow numbered steps through a task
• Keep track of progress across conversation turns

Manage WireMock from the command line to create mock HTTP responses, inspect and count received requests, change scenario states, and record real backend traffic. These operations support integration tests with controlled responses.

中文简介正在准备中。当前显示英语简介。

• Create mock API responses for integration tests
• Check request counts and unmatched requests
• Switch states in a multi-step test scenario

Run molecular property predictions, explore molecular shapes, and estimate how compounds bind to proteins through a cloud Python API. Process compound batches and connect modeling steps without maintaining local high-performance computing infrastructure.

中文简介正在准备中。当前显示英语简介。

• Predicting properties across a compound library
• Exploring molecular shapes and tautomeric forms
• Estimating how compounds bind to proteins

Audit an AI agent's completion claims using existing logs, artifacts, approvals, and deployment records. Reconstruct the run and assess each claim separately, distinguishing proven results, partial support, contradictory evidence, and missing proof without rerunning the work.

中文简介正在准备中。当前显示英语简介。

• Checking whether an agent's completion report has supporting evidence
• Reviewing tool calls, retries, and checkpoint recovery
• Matching approvals and deployments to the same artifact

Use natural-language requests with Aomi CLI to check crypto prices and balances or prepare swaps, staking, and transfers between chains. Review queued transactions and simulate multi-step batches before separately authorizing wallet signing.

中文简介正在准备中。当前显示英语简介。

• Check crypto prices and wallet balances
• Prepare and simulate a Uniswap swap
• Prepare an ETH stake with Lido

Add browser automation to Vercel apps by running agent-browser and headless Chrome in temporary, isolated virtual machines. Capture screenshots, inspect page structure, and complete multi-step forms while preserving browser state between commands.

中文简介正在准备中。当前显示英语简介。

• Add website screenshot capture to a Vercel app
• Fill and submit forms across multiple browser commands
• Run recurring browser tasks with Vercel Cron Jobs

Helps agents manage long tasks by checking for forgotten instructions and drifting conventions. It guides progress summaries, rule checks after context compaction, and handoff notes when work should continue in a fresh session.

中文简介正在准备中。当前显示英语简介。

• Keeping rules consistent during long agent sessions
• Rechecking decisions after context compaction
• Preparing handoff notes for a fresh session

Retrieve past operational experience from OpenViking before or during coding, configuration, deployment, or recovery work. It reads relevant records, checks whether their assumptions match the current environment, and applies suitable procedures and checks.

中文简介正在准备中。当前显示英语简介。

• Consulting past experience before code changes
• Checking prior procedures for configuration changes
• Finding diagnostic steps after a deployment failure

Use OpenViking MCP tools to retrieve relevant knowledge before substantial tasks and save lasting preferences, decisions, and lessons during or after work. Import documents as searchable resources so later tasks can draw on previous findings.

中文简介正在准备中。当前显示英语简介。

• Retrieve previous fixes before debugging
• Save lasting preferences and working conventions
• Record decisions and their rationale

Make Copilot CLI easier to follow with plain-English explanations of approval requests, errors, and command results. The skill adds color-coded risk levels, step-by-step plans, explanations of technical terms, and summaries of changes with undo guidance.

中文简介正在准备中。当前显示英语简介。

• Understand an approval request before deciding
• Interpret errors and command results
• Follow a multi-step task and compare options

Break a development task into small requirements with explicit acceptance criteria, then repeat implementation and verification while saving progress. Completion depends on evidence for each requirement and reviewer approval, with configurable reviewers and a final cleanup check.

中文简介正在准备中。当前显示英语简介。

• Carrying a multi-step implementation through verification
• Resuming work with recorded progress
• Tracking evidence for each requirement

Manage multi-step financial research with a clear objective, acceptance criteria, and linked evidence. Track findings across turns, audit whether each criterion is supported, and report unresolved gaps while keeping the workflow limited to research.

中文简介正在准备中。当前显示英语简介。

• Compare strategies or assets against shared criteria
• Review the evidence supporting an investment thesis
• Resume a research task across multiple turns

Review contact, lead capture, quote, survey, and checkout forms to identify obstacles to completion. Get prioritized recommendations for fields, layout, wording, mobile usability, and error messages, plus ideas for comparing alternative designs through tests.

中文简介正在准备中。当前显示英语简介。

• Reviewing where visitors abandon a contact form
• Reconsidering required fields in a demo request form
• Planning a multi-step quote request form