Coding standards, frameworks, refactoring, and code review (18.2K skills)
Add human review and approval points to LangGraph workflows. Configure when execution pauses, how workflows resume after input, and how pending approvals are handled through timeouts, escalation paths, and notifications.
Turn a multi-step software specification into a practical implementation plan before coding. It breaks work into small tasks, tracks dependencies, and gives each task a test-first sequence with expected results and a handoff path for execution.
Investigate bugs, failing tests, and unexpected behavior by tracing data flow, comparing working examples, and testing one hypothesis at a time. It guides a focused fix and verification, and prompts you to reconsider the architecture after repeated failed fixes.
Use this skill to carry out implementation plans with independent tasks in the current session. It assigns a fresh subagent to each task and adds separate checkpoints for checking whether the work meets the request and whether the code is well built.
Evaluate code review feedback against the actual codebase before making changes. Clarify the requirement, check technical fit, respond with acknowledgment or a reasoned objection, then implement valid suggestions one at a time.
Decide what to do with a development branch after implementation is complete and tests pass. The workflow checks the base branch, presents four integration choices, carries out the selected option, and cleans up the worktree when required.
Execute a written implementation plan in a separate session, handling tasks in batches and reporting for review between them. It supports resuming saved tasks, setting up a worktree, and finishing the branch after the plan is complete.
Use this skill to split independent engineering problems across parallel agents. It helps coordinate separate investigations, check their solutions for conflicts, and verify the combined result with the full test suite.
Turn a technical plan into small development tasks with acceptance criteria, effort estimates, dependencies, and links to specification requirements. The skill also identifies work that can happen in parallel, maps the critical path, and includes testing tasks.
Convert feature requests into user-focused requirements and stories with testable acceptance criteria. The skill clarifies business value, scope, edge cases, assumptions, and open questions so a specification can be reviewed before planning.
Carry out feature development from a specification by preparing project files, implementing tasks in dependency order, and checking acceptance criteria with tests. It also validates requirement coverage and includes integration testing and a human review step.
Define and document project-wide principles for how a team works, including development practices, code quality, testing, UX, performance, security, and architecture. Use it to formalize existing standards or set governance before planning a new feature area.
Plan and work through complex coding tasks in stages, using tools, correcting mistakes, and checking the proposed solution. The source describes autonomous engagement without human intervention, but provides no implementation details beyond that overview.
Clarify goals and requirements before coding begins. The process reviews project context, explores design options and tradeoffs, then prepares a design document for approval before implementation planning continues.
Turn specifications and project constraints into a technical architecture and implementation strategy. Compare technology options, map components and integrations, plan milestones and testing, assess risks, and prepare decisions for human review before breaking work into tasks.
Sets up an isolated Git worktree and new branch for feature work, detects the project setup command, and checks the existing test baseline. It helps prepare a separate workspace before carrying out an implementation plan.
Set up LangSmith tracing for LLM applications built with LangChain or LangGraph. Capture nested execution traces, inspect run history and metadata, and compare runs to help debug agent behavior. Requires a LangSmith account and API key.
Request a code review after finishing work, implementing a major feature, or preparing to merge. The review checks the changes against requirements and returns feedback grouped by urgency, so you can address critical and important issues before proceeding.
Build custom tools for LangChain agents and connect existing tools or APIs. Define structured inputs, descriptions, return types, and error handling so an LLM can understand how to call each tool; the skill also covers tool chains and pipelines.
Build routing for LangGraph workflows so the next node depends on the current state or a routing decision. The skill covers conditional edges, router nodes, multiple branches, and fallback paths, helping structure workflows that handle different outcomes and errors.
Guides the creation and editing of agent skill documents and checks whether they work before deployment. It uses pressure scenarios with subagents to see whether a skill changes agent behavior, then refines the document to close gaps.
Coordinate multiple agents working in parallel by choosing a communication structure, assigning roles, and setting a consensus method. Use checkpoints to check that work stays aligned, then support a shared final decision.
Use this skill to carry software tasks through four separate stages: implementation, independent validation, adversarial review, and commit. It checks work against a specification and requires evidence of a passing review before committing.
Route tasks according to complexity, from quick code transformations to multi-agent coordination. It uses Q-learning, a method that improves choices from past outcomes, and selects among a WebAssembly fast path and Mixture-of-Experts models for different workloads.