claude-skills

Reviews Python code changes for quality, correctness, and adherence to project conventions before commits or PRs. Checks for SOLID violations, DRY/YAGNI issues, missing tests, type safety, security concerns, and style compliance. Use when reviewing diffs, preparing PRs, after implementing features, or when the user asks to review code. Python-specific — covers ruff, pyright, pytest, and modern Python 3.13+ idioms by default (falls back to the project's pyproject.toml `requires-python` floor if it pins something lower). For deep silent-failure hunting or type-invariant design, use code-review's targeted agents instead.

Turn clear requirements into an ordered, verifiable implementation plan. Use when a spec or well-understood goal needs decomposition before coding — when the task feels too large to start, the order of work isn't obvious, or scope must be communicated. Trigger phrases include: "plan this", "break this down", "make an implementation plan", "how should we implement this", "create a task list", "what order should we build this in". This skill defines HOW to build and in what order — it assumes WHAT to build is already pinned down. If requirements are loose, unverified, or missing, first run the task-scoping skill (task-scope-and-spec: scope a loose engineering task into a tight, testable spec list) or gather inputs per the Input Gate below.

Scope a loose engineering task into a tight, testable spec list. Use when the user gives background for a piece of work and wants the requirements narrowed down before implementation. Trigger phrases include: "scope this", "what are the specs", "what's involved", "what would it take", "what's left for this task", "port this to", "bring this logic to", and "spec this ticket/issue". This skill defines WHAT must be true when the work is done — it does not produce an implementation plan, task ordering, or execution steps. For sequencing the specs into ordered, verifiable tasks, use planning-and-task-decomposition instead.

Bootstrap the .hac/ (Human-Agent Context) directory in any project. Use when the user says "set up hac", "add .hac", "set up working memory", "initialize hac", or when beginning multi-session work on a project that lacks a .hac/ directory. Language-agnostic — works for Python, TypeScript, Go, Rust, or any project type. Also trigger when migrating existing plan files into .hac/ or when resetting a stale .hac/ directory.

Write focused pytest tests covering critical paths, edge cases, and error handling. Avoids redundant tests that bloat codebase. Use for creating minimal but comprehensive test coverage of complex logic, public APIs, and failure scenarios.

Generates a minimal, clear PR description from the current branch's diff and the repo's PR template. Use at the end of any task when the user wants to open a PR or needs a PR description written.

Extracts what the user actually wants instead of what they think they should want. Achieves this through one-question-at-a-time interview until ~95% confidence about the underlying intent. Use when an ask is underspecified ("build me X" without "for whom" or "why now"), when the user explicitly invokes ("interview me", "grill me", "are we sure?", "stress-test my thinking"), or when you catch yourself silently filling in ambiguous requirements before any plan, spec, or code exists.

Implement a spec fully, verify each requirement with parallel agents, and produce a report. Use whenever the user runs /goal-workflow, mentions "implement spec", "verify against spec", or wants to go from requirements to verified implementation in one shot. Even if the request seems simple, invoke this skill — it handles both small and large specs.

Use when the user asks "what can I ask you to do?", "what skills do you have?", "show me a usage guide", or types /help. Presents a user-facing cheat sheet of all personal skills grouped by category.

Scaffold a new Python project with production-ready tooling and structure. Use when the user says "new project", "set up a project", "initialize a Python project", "create a repo", "project template", or asks for boilerplate setup. Generates pyproject.toml, Makefile, pre-commit config, ruff/pyright config, test structure, HAC (.hac/), and directory layout following modern Python conventions with uv, ruff, pyright, and pytest.

Draw polished, hand-drawn-style Excalidraw diagrams for any technical or conceptual topic. Use this skill whenever the user asks to "draw", "diagram", "visualise", "make an Excalidraw", or "sketch" something. Triggers include: system architecture, data flows, pipelines, concept maps, process flows, comparisons, and any request for a visual explanation. Always run the full three-phase approach: ASCII preview → wait for approval → draw. Never skip to drawing without the ASCII step first.

Explains code with visual diagrams, analogies, and structured walkthroughs. Use when explaining how code works, teaching about a codebase, when the user asks "how does this work?", "what does this do?", "explain this code", or "walk me through this". Distinct from codebase-research which answers WHERE something is — this skill answers HOW and WHY something works.

Expert Python development workflow following SOLID principles, OOP design patterns, and modern best practices. Use when architecting Python applications, refactoring code, implementing complex features, or making design decisions requiring careful planning. Applies Factory Method, Strategy, Dependency Injection patterns. Creates modular architecture with pytest testing. For new project setup, see the python-project-scaffold skill instead.

Navigate, search, and map codebases efficiently to answer structural and locational questions. Use this skill whenever the user asks where something is in a codebase, how the code is organized, where a pattern or symbol is used or defined, what depends on a module, or whether something is still in use. Trigger for any of these signals: "where is this logic?", "find all usages of X", "where is Y defined?", "how is the codebase structured?", "what imports Z?", "is this used anywhere?", "which files handle X?", "give me an overview of the codebase", or any architectural question about an existing project. Always prefer this skill over ad-hoc file reading — it produces ranked, synthesized answers rather than raw tool output.

Expert guidance for Large Language Model development — architectures, training, inference, and production applications. Use when working with transformer internals, pretraining/fine-tuning pipelines, RAG systems, agentic workflows, evals, tool use, structured outputs, reasoning models, MCP integrations, or shipping LLM features to production. Combines deep-learning expertise with modern engineering practices (Python 3.13+, pytest, eval-driven development).

Language-agnostic, four-phase debugging methodology with root cause analysis. Use when investigating bugs, fixing test failures, diagnosing unexpected behavior, or troubleshooting any runtime/logic error. Trigger this skill whenever the user says "debug", "fix this bug", "why is this failing", "unexpected behavior", "broken", "not working", or presents a stack trace / error log. Emphasizes NO FIXES WITHOUT ROOT CAUSE FIRST.

Create a structured handover document so the next agent or engineer can resume work with zero prior context. Use this skill when a session is getting too long, the context window is under pressure, the user asks to "wrap up", "hand off", "create a summary for the next agent", or says "let's continue in a new chat". Also trigger proactively when a multi-subtask session has been running for a long time and meaningful progress has been made. Produces a single HANDOVER.md file — not a chat summary — that survives session boundaries and enables true cold-start resumption.