开发技能

Review Claude's assumptions before planning a development phase. The skill explains the proposed technical approach, implementation order, scope, risks, and dependencies so users can correct misunderstandings and decide how to proceed.

中文简介正在准备中。当前显示英语简介。

• Review technical assumptions before planning
• Check implementation order and dependencies
• Clarify scope boundaries and risks

Review code changed during a development phase for bugs, security vulnerabilities, and quality issues. Choose the analysis depth or specific files, then receive a report with findings grouped by severity and an inline summary.

中文简介正在准备中。当前显示英语简介。

• Review code changes at the end of a development phase
• Check specific files for bugs and security issues
• Analyze dependencies and calls across multiple files

Investigate bugs through symptom gathering, hypotheses, and testing, with session state saved to files so work can resume after context resets. Choose diagnosis alone for a root cause report, or continue through fixing the issue.

中文简介正在准备中。当前显示英语简介。

• Investigate a bug from its symptoms and reproduction steps
• Resume an interrupted debugging session
• Review hypotheses and evidence from active sessions

Coordinate changes with a Claude session working in another repository. Specify API inputs, outputs, authentication, and errors, then review the reply and verify the implementation before building work that depends on it.

中文简介正在准备中。当前显示英语简介。

• Requesting a backend endpoint needed by the frontend
• Coordinating a client change required by the backend
• Requesting a shared package update

Analyze which development phases must precede others by checking shared files, API usage, and data flow. Suggest dependency entries for ROADMAP.md and apply confirmed changes to clarify the order of work.

中文简介正在准备中。当前显示英语简介。

• Filling missing phase dependencies in ROADMAP.md
• Reviewing execution order before parallel development
• Ordering phases that modify the same files

Audit a development milestone before archiving it by comparing completed work with its original goals. Review requirement coverage, connections between phases, and complete user flows, while collecting technical debt and deferred gaps.

中文简介正在准备中。当前显示英语简介。

• Checking milestone completion before archiving
• Reviewing connections between development phases
• Collecting technical debt and deferred gaps

Handle repository change requests from another Claude session. Guides user approval, clarification of missing API details, negotiation of proposed changes, and precise replies listing changed files, agreed behavior, verification results, and deviations from the request.

中文简介正在准备中。当前显示英语简介。

• Handling an API change request from another session
• Clarifying missing request and response details
• Proposing an alternative before implementation

Audit registered projects, turn outstanding issues into tasks with clear dependencies, and assign work to parallel agents. Review their changes, check automated tests, and move pull requests through merging and deployment verification.

中文简介正在准备中。当前显示英语简介。

• Audit unfinished work across multiple projects
• Assign independent fixes to parallel agents
• Coordinate API changes across repositories

Run remaining project phases through discussion, planning, and execution. Track progress and phase documents, then audit and close the milestone. Select a phase range or use interactive discussion when decisions need your input.

中文简介正在准备中。当前显示英语简介。

• Run all remaining milestone phases
• Execute a selected phase or phase range
• Keep phase progress and planning documents updated

Review open GitHub pull requests across repositories, assign agents to fix automated checks and review feedback, and merge after confirmation. Also supports conflict handling, previewing planned actions, and optional synchronization between dev and main branches.

中文简介正在准备中。当前显示英语简介。

• Reviewing open pull requests across repositories
• Fixing failed automated checks and review feedback
• Previewing merge actions without making changes

Review issues from Sentry, Linear, and GitHub Issues against code and deployment evidence. Close confirmed fixes after approval, organize remaining issues by priority, and dispatch agents to prepare fixes and pull requests.

中文简介正在准备中。当前显示英语简介。

• Review issues across Sentry, Linear, and GitHub Issues
• Close issues whose fixes are confirmed deployed
• Prioritize unresolved issues

Turn a codebase into Markdown guides for operation, deployment, architecture, and maintenance, plus an index.json for AI agents. Operational claims include source references and confidence labels, while unverified deployment details remain documented gaps.

中文简介正在准备中。当前显示英语简介。

• Understand an unfamiliar codebase
• Create source-backed operation and deployment guides
• Prepare documentation and an index for an AI agent handoff

Query and update Supabase data, inspect table structures, and generate TypeScript types through Rube MCP. Review project settings and service health, and inspect storage bucket lists and Edge Functions, code hosted by Supabase.

中文简介正在准备中。当前显示英语简介。

• Filtering and sorting table data
• Inspecting table structures before SQL updates
• Generating TypeScript types from database structures

Plan system architecture and APIs, compare design patterns, and record decisions with their context and rationale. Check scalability, reliability, monitoring, and security to choose a simple structure that meets the project's requirements.

中文简介正在准备中。当前显示英语简介。

• Comparing system architecture options
• Recording design decisions and their rationale
• Reviewing scalability and reliability

Create or update Claude skills that package specialized knowledge, repeatable workflows, and tool instructions. Work from concrete usage examples to organized instructions and reusable resources, then validate the skill and package it as a distributable ZIP.

中文简介正在准备中。当前显示英语简介。

• Create a skill for a recurring workflow
• Package company knowledge and API instructions
• Organize scripts, references, and templates

Create structured Claude skill packages, validate their files and metadata, and bundle them as ZIP archives. Share the new skill’s description and file links in a Slack channel through Rube so teammates can discover it.

中文简介正在准备中。当前显示英语简介。

• Create a template for a new Claude skill
• Validate a skill before distribution
• Package a skill and its supporting files as a ZIP

Understand a TypeScript codebase using Ontoly's Software Graph, a map of code elements and their relationships. Explain architecture, trace request flows, inspect dependencies, and assess change impact while distinguishing reported facts from inference.

中文简介正在准备中。当前显示英语简介。

• Understand a repository's architecture
• Trace a login request through the code
• Assess the impact of changes and refactoring

Investigate past agent runs using recorded tool calls, exit codes, and file changes. Replay failures, distinguish observed evidence from inferred causes, and compare models from the same saved point when requested.

中文简介正在准备中。当前显示英语简介。

• Find why an agent overwrote a file
• Identify which step broke a build
• Replay a previously recorded failure

Build MCP servers, which expose external services as tools for AI agents. Follow a structured process covering API research, tool design, Python or TypeScript implementation, code review, and realistic evaluations of how agents use the tools.

中文简介正在准备中。当前显示英语简介。

• Connecting external APIs to AI agents
• Building an MCP server in Python or TypeScript
• Improving tool inputs, responses, and error messages

Analyze recent Claude Code conversations to identify coding habits, strengths, and specific learning gaps. Receive prioritized improvement suggestions and relevant HackerNews reading recommendations in a personalized report delivered to Slack direct messages.

中文简介正在准备中。当前显示英语简介。

• Review development habits from recent work
• Identify recurring technical difficulties
• Choose learning resources tied to actual projects

Delegate reviews, research, and other tasks from Claude Code to Codex or Antigravity CLI. Claude evaluates the responses, explains agreements and disagreements, and preserves separate conversations for follow-up questions on each topic.

中文简介正在准备中。当前显示英语简介。

• Getting a second opinion before merging code
• Challenging an architecture with another model
• Delegating a research side task

Manage long or interruption-prone Claude Code tasks with saved progress, clear next actions, and limited retries. A local checkpoint records verification results and requires a passing check before the task can be marked complete.

中文简介正在准备中。当前显示英语简介。

• Resuming a migration across multiple sessions
• Limiting retries while fixing failing tests
• Recording verification before declaring completion

Build complex, interactive HTML artifacts for claude.ai using React, Tailwind CSS, and shadcn/ui. It supports interfaces with shared state and multiple views, then packages the code and styles into one HTML file for sharing.

中文简介正在准备中。当前显示英语简介。

• Creating an interactive interface with multiple views
• Building screens with shadcn/ui components
• Packaging a React app into a single HTML file

Address issues identified in a QA code review by investigating causes, fixing critical problems first, and updating tests. The skill prepares a fix report that records changes and explains any deferred warnings for re-verification.

中文简介正在准备中。当前显示英语简介。

• Fixing critical issues after a code review
• Updating tests alongside review fixes
• Documenting reasons for deferred warnings