Coding standards, frameworks, refactoring, and code review (18.5K skills)
Help Claude Code carry long coding tasks across session resets. The skill filters command output, records significant actions and pending tasks in `session.log`, and uses that log to announce what remains and continue work when a new session starts.
Create a clear CONTRIBUTING.md for an open-source project, covering local setup, ways to contribute, the change and pull request workflow, project standards, review expectations, and where to ask for help.
Simplify working code by removing unused abstractions, needless indirection, and confusing control flow while preserving behavior. The result includes verification evidence and a ledger explaining each removal and the future flexibility it gives up.
Make unfamiliar code easier to understand with a one-line summary and a step-by-step walkthrough tailored to the reader. It also points out non-obvious behavior, assumptions, edge cases, and bugs or code smells noticed while reading.
Reviews a pull request or code diff for correctness, security, design, tests, and readability. Findings are ranked from blocking issues to optional nits, with reasons and concrete suggestions, alongside specific strengths and a clear approve, request-changes, or comment verdict.
Creates a focused pull request review checklist tailored to the programming language, change type, and risk level. It covers relevant correctness and testing checks, common pitfalls, review depth, and concrete conditions for approval or requesting changes.
Helps prepare a project context file such as CLAUDE.md for an AI coding agent. It captures the project structure, conventions, exact build and test commands, and safety guardrails so the agent can navigate the codebase and verify its work.
Guide Claude through a four-stage coding workflow: confirm a written plan, isolate changes on a feature branch, write tests before implementation, and review the result for correctness and code quality before presenting it.
Turn an unclear bug report into an actionable triage record with reproducible steps, environment details, justified severity and priority, a likely component and owner queue, and a duplicate check. It flags missing information instead of filling gaps with guesses.
Investigate software bugs through reliable reproduction, narrowing the faulty area, and testing ranked cause hypotheses. Get a structured diagnosis with evidence for the root cause, a fix, and a regression test to help prevent the defect from returning.
Turn a real commit range into release notes people can understand. The skill groups changes by user impact, combines commits for the same feature, explains breaking changes and upgrade steps, and produces a changelog ready to paste into a release.
Turn a vague software problem into a reproducible bug report. It organizes a clear title, exact steps, expected and actual results, environment details, severity and priority, and evidence, while marking assumptions and keeping suspected causes separate from observed facts.
Turn raw git commits, a commit list, or developer release notes into a user-facing changelog entry. Organize changes under a version header and relevant categories, call out breaking changes with migration guidance, and omit sections that have no entries.
Create a versioning strategy for an API or service, tailored to its API type, consumers, and constraints. The document sets a version scheme and lifecycle, classifies breaking changes, and includes a deprecation timeline, migration guide structure, and client communication templates.
Turn a system description into a renderable Mermaid architecture diagram. Group services, clients, data stores, and external dependencies into logical layers, show connection direction and sync or async flows, then add a component legend and notes.
Document a technical or architectural choice in a structured ADR, a record of why a decision was made. Capture its context, options, chosen approach, consequences, tradeoffs, risks, and conditions that could prompt a review.
Review AI-written or AI-assisted code for subtle logic errors, invented or misused APIs, weak tests, duplicate utilities, unnecessary complexity, leftover scaffolding, and security shortcuts. Get severity-ranked findings, practical verification steps, and a checklist tailored to the codebase.
Prepare an existing codebase for an evo optimization run. Explore its goals and evaluations, suggest promising optimization targets when needed, set up a benchmark and instrumentation in a baseline worktree, and launch the first experiment so results can be compared against a recorded baseline.
Turn the best valid result from an evo run into a focused change that maintainers can review. Rebuild the minimal diff, check that tests and project conventions remain intact, verify the result, and prepare a mergeability report; use a pull request for repositories with a remote or merge locally otherwise.
Resolve an Octane or Octane library defect at the repository that owns it. When the fix is clear, add regression coverage and prepare a pull request; otherwise, file an actionable issue with reproduction evidence and affected versions.
Review Octane changes for risks to frequently executed code, including unstable V8 object shapes, forced DOM layout, and extra scheduling steps. The skill scans changes, checks findings against runtime behavior, and records the evidence needed before a pull request is ready.
Use this skill to classify an unfamiliar Octane issue, failure, test result, or proposed task. It identifies the likely owning area, issue type, severity, and confidence, then suggests evidence-based next steps and a focused validation command.
Route React library work in Octane to the appropriate workflow: use update-bindings for maintenance of existing bindings, or octane-react-library-port for a new port or copied and rewritten React implementation.
Assess Octane performance changes across runtime, compiler output, server rendering, hydration, scheduling, and bundle size. The skill helps choose a relevant benchmark, compare equivalent work against a baseline, investigate regressions, and report evidence and remaining uncertainty.