Coding standards, frameworks, refactoring, and code review (18.4K skills)
Use this skill to shape a feature, component, or system before implementation. It asks focused questions, compares design options, and helps produce an agreed specification so the work starts with clearer requirements and decisions.
Review a completed feature, work session, or incident using concrete evidence such as failed checks and rework. Identify costly friction, trace likely causes, and propose durable fixes, prioritizing automation and keeping only a few high-impact findings.
Plan and guide structural refactors across files or modules while preserving behavior and public contracts. The skill maps dependencies and consumers, ranks changes by priority, and lays out staged moves with validation steps and compatibility considerations.
Investigate bugs by clarifying expected behavior, reproducing the problem, and testing possible causes. The skill helps prepare a fix plan with risks and validation steps before code changes begin.
Analyze existing code for complexity, duplication, coupling, and unnecessary layers, then propose prioritized simplifications that improve readability and maintainability. The workflow checks whether code can be removed or replaced with existing platform features before suggesting changes.
Understand a code entry point by examining its purpose, behavior, and dependencies up to three levels deep. The skill produces a structured Markdown knowledge document, with an optional HTML overview when requested.
Clarify a feature’s problem, users, scope, success criteria, and open questions, then prepare its requirements, design, and testing documents. It helps teams compare approaches and review requirements before moving into design work.
Turn feature requirements, design notes, and test scenarios into an ordered implementation plan. Track task status, dependencies, blockers, scope changes, and next steps in the configured planning documents.
Track development and debugging work with the ai-devkit task CLI. Record phases, progress, next steps, blockers, ownership, and fresh validation evidence so later readers can see what changed, what remains, and what supports the result.
Turn a long AI coding session into a structured handoff or resume artifact in Markdown or JSON. The Jev-backed CLI extracts useful continuation facts and possible memory candidates; review its output before relying on its claims.
Check whether AI DevKit is ready on the local machine, identify setup gaps it can address, and run the narrowest relevant setup for supported agent harnesses. Recheck status afterward and report remaining manual actions.
Coordinate a feature’s development phases, from requirements and design through planning, implementation, testing, and review. It checks project setup, routes work to the relevant phase skill, and uses documented progress and verification to guide what comes next.
Publish a committed, reviewed feature branch by fetching and rebasing it onto the latest base branch, pushing safely, and opening or updating a review request on a supported Git host. Report the URL, commit, validation, and risks.
Prepare or resume an isolated workspace for feature work by selecting and verifying a branch and Git worktree, then bootstrapping dependencies. It reports the active work directory and any workspace risks or skipped setup steps.
Carry out planned development tasks and check whether the code matches the feature’s design and requirements. The workflow also keeps implementation and testing documents current, tracks task progress, and summarizes deviations, gaps, and follow-up work.
Prepare a focused Git commit by reviewing repository changes, staging only files related to the requested work, and choosing a concise conventional commit message. It also reports validation and any remaining unstaged files.
Review a feature before pushing by comparing its code with the design, tracing affected callers, and checking contracts, edge cases, security, and integration risks. Get findings ordered by severity with file references and recommended fixes.
Review a feature design against its requirements. Check architecture, components, data models, API contracts, trade-offs, and non-functional needs, then clarify decisions and record reusable architecture choices before implementation begins.
Update the top Unreleased section of CHANGELOG.md using commits since the latest release. Add one concise linked line per commit, using a pull request link when evidence is clear and a commit link otherwise, while preserving the file’s existing style.
Explore a local map of code symbols and their relationships to answer structural questions about a repository. Find definitions, callers, imports, and nearby dependencies, then inspect the source files for details. The graph refreshes automatically after agent turns.
Browse an automatically refreshed, AST-derived map of a repository through the Deeplake mount. Search symbols, inspect file neighborhoods, trace dependency paths, and estimate change impact, then read the actual source for confirmation. Supports TypeScript, JavaScript, and Python.
Explore an automatically refreshed code graph to locate symbols and inspect calls, imports, inheritance, and dependencies. Use it to trace flows, identify likely change impact, or map repository subsystems, then verify findings in the source files.
Plan a bounded, five-layer profiling run for one model and prepare the executor prompt. The workflow pauses for human approval before GPU work, verifies the agent’s completion, and routes evidence to local run artifacts and the shared methodology document.
Investigate a slow sglang-omni stage by pairing a stack-enabled mapping trace with a production-like formal trace. The analysis attributes GPU kernel time to Python source lines and separates code location from the time users actually experience.