Coding standards, frameworks, refactoring, and code review (18.5K skills)
Review an implementation by comparing its plan, change record, research, and validation evidence. Record scoped findings with severity, supporting evidence, resolution conditions, and suggested routes for accepted follow-up work.
Check an implementation plan against supplied requirements and evidence without changing the plan. Receive one consolidated assessment with actionable findings, severity, evidence, and the smallest useful corrections; follow-up reviews also track whether earlier findings were resolved.
Choose an HVE-Core installation method based on the user's environment, team needs, and update preferences. Install through the VS Code extension or a repository clone, then validate the setup and optionally add selected components.
Document existing or planned software architecture with C4 diagrams for system context, containers, components, and, when supported by evidence, deployment. It produces Mermaid diagrams by default and can update existing diagrams while preserving unaffected content.
Review Python changes against practical standards for style, idioms, type hints, error handling, security, and maintainability. It helps an agent spot risky patterns and keep new code consistent with the existing project structure.
Create a lasting architecture review record that captures confirmed scope, evidence, assumptions, pillar findings, trade-offs, recommendations, and reviewer decisions. It keeps analysis traceable and links significant accepted decisions to ADRs while leaving decision authority with the user.
Create Architecture Decision Records that capture scope, decision makers, drivers, options, rationale, and consequences. Choose a concise Y-Statement or MADR v4 format, start from a planner handoff, or adopt an existing project template while preserving supersession history.
Prepare a review-ready GitHub pull request from the current branch by inspecting its committed changes, drafting a concise description, and running focused local checks. With separate approval, it can push changes and create or update the pull request.
Create or update GitHub Copilot agents, prompts, instructions, and skills using the repository's current formats and workflows. It guides artifact selection, frontmatter, package placement, documentation, delegation, tracking, and applicable validation checks.
Use a Python command-line tool to inspect and update GitLab merge requests, comments, pipelines, jobs, and logs on GitLab.com or self-managed instances. It supports OAuth login by default and an explicit legacy-token mode for compatible setups.
Review pull requests, branch diffs, or local changes through functional, standards, accessibility, security, and readiness perspectives. The workflow builds context, adjusts review depth to risk, and presents evidence-based findings with consistent severity levels.
Draft a new agent skill or improve an existing one by clarifying its purpose, writing instructions, and evaluating example runs. It can also benchmark results and refine the skill description so the skill is more likely to trigger for relevant requests.
Install, inspect, list, or upgrade Python packages with `uv pip` in a uv-managed environment. It helps resolve missing-import errors by mapping module names to package names and checking whether the dependency is already installed.
Investigate bugs by turning observed symptoms into specific, testable hypotheses, then designing focused experiments to confirm or reject them. It helps isolate failing components, interpret errors and test results, and narrow the search toward a root cause.
Narrow down where a bug begins by tracing data and control flow, isolating components, and building minimal reproductions. The skill also covers git bisect for regressions, strategic logging, conditional breakpoints, and watch expressions in TypeScript, SQL, and bash.
Organize an unclear incident or complex bug into independent investigation threads, each testing a different possible cause. Share evidence at planned sync points, decide whether to continue, redirect, or converge, and combine findings into a root cause report.
Review code or pull requests through five focused stages: requirements, correctness, maintainability, testing, and security or performance. The skill groups findings by severity and explains why each issue matters, with a concrete suggestion for addressing it.
Set validation checkpoints between project phases, with acceptance criteria, automated checks, manual sign-offs, and conditional reviews. This helps teams catch issues before moving forward and decide when work is ready for the next step.
Investigate bugs, crashes, and unexpected behavior through reproduction, evidence gathering, ranked hypotheses, and systematic tests. Use methods such as the Five Whys, stack trace analysis, and Git history review to document a confirmed cause before proposing a fix.
Plan a feature or system before coding by clarifying requirements, comparing implementation options, and documenting architecture, data models, interfaces, and steps. It helps turn complex or unclear work into a design that can be reviewed and refined during implementation.
Rebuild eligible creative websites by first capturing a local mirror, then tracing and porting their client-side behavior and assets, and checking the result against the original. The workflow covers sites with WebGL, scroll animations, or portfolio layouts and sets explicit verification steps.
Work through an undecided technical or product design choice, compare options, and recommend an approach. The skill writes or updates a build specification in `docs/specs/`, helping development proceed from recorded decisions.
After a change, compare repository evidence with its `AGENTS.md` guidance, scope tracking, and linked spec statuses. It makes narrowly defined updates and reports stale specs or conflicts that need a person or another workflow to resolve.
Investigate a failing test or incorrect behavior by reproducing it, narrowing down where it occurs, and testing one root cause hypothesis at a time. Apply a minimal fix, verify the result, and provide a regression test or guidance for one.