arch-design
Design architecture docs for new features, refactors, or redesigns. Produces implementation-ready docs with complete file impact analysis.
Best use case
arch-design is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Design architecture docs for new features, refactors, or redesigns. Produces implementation-ready docs with complete file impact analysis.
Teams using arch-design should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/arch-design/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How arch-design Compares
| Feature / Agent | arch-design | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Design architecture docs for new features, refactors, or redesigns. Produces implementation-ready docs with complete file impact analysis.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Architecture Design Design an architecture doc for a new feature, refactor, or redesign. Produces a doc in `docs/architecture/pending/` with complete file impact analysis. ## Inputs The user provides one of: - A feature name or concept (e.g., "behavior mutations") - A path to an issue doc (e.g., `docs/issues/002-behavior-mutations.md`) - A description of the problem to solve ## Process ### Checkpoint 1: Problem + Solution (wait for user approval) 1. **Load context:** - `docs/CAPABILITIES.md` — current capability status - `docs/architecture/README.md` — architecture overview + reading order - If user provided an issue path, read that too 2. **Write Problem and Solution sections:** - Problem: What's wrong or missing. Include a concrete example (config snippet, error, limitation). - Solution: High-level approach. One paragraph + diagram or YAML snippet. 3. **Present to user for approval.** Do not continue until they confirm the direction is right. ### Automated Phase (after direction is approved) 4. **Load relevant architecture docs** per README.md reading order for the area being designed. 5. **Read existing source code** that the design touches. Use Grep/Glob/find_references to locate all relevant files. 6. **Write remaining sections** using the template at `.claude/skills/arch-design/template.md`: - **Semantics** — behavioral rules, edge cases, ordering, timing - **Configuration** — YAML examples if the feature has educator-facing config - **Interface Contracts** — function signatures with full docstrings - **Validation Rules** — parse-time (Pydantic) and business rules - **Implementation Impact** — the key section (see below) - **What Doesn't Change** — explicit scope boundaries 7. **Build the Implementation Impact section:** - Use Grep to find every import of types/functions being moved or modified - Use find_references for symbols being renamed or relocated - Use Glob to find test files that cover affected modules - For each file: action (Create/Modify), path, one-line description of change - Separate into: Source Files, Test Files, Import Updates - The Import Updates table must show current import → new import for every file affected by moves/renames 8. **Write the complete doc** to `docs/architecture/pending/<name>.md` 9. **Present the full doc to user** with a summary of: - Total files affected (create/modify counts) - Import sites found - Any design decisions that could go either way (flag for user) ## Quality Rules - **No invented scenario values** (Principle #7) — contracts must not introduce defaults for educator-specified parameters - **No future scaffolding** (Principle #8) — design only what this feature needs, not extensibility for hypothetical future work - **Breaking changes are fine** (Principle #9) — don't add compatibility shims - **Complete import tracing** — every file that imports a moved/renamed symbol must appear in the Import Updates table. Use Grep for `from module import symbol` and `import module` patterns. - **Concrete contracts** — every function signature includes Args, Returns, Raises. No `...` bodies in the doc; show the signature and docstring only. - **Testable sections** — Semantics section should use tables (Condition | Result) so sprint specs can derive test cases directly. - **No implementation code** — design docs contain signatures and docstrings, never implementation bodies. Describe behavior in prose and tables, not code blocks with for-loops or if-statements. Wrong: showing the literal code to insert into `processor.py`. Right: "After `execute_action()` returns non-empty decisions, apply mutations and record history if tracked." - **No duplication with sprint specs** — the design doc provides rationale, semantics, and constraints (the WHY). The sprint spec provides contracts, phases, and test cases (the WHAT). If a sprint spec will be written from this design doc, the sprint planner extracts contracts from here — don't write content that forces the planner to duplicate or filter. ## Output Location `docs/architecture/pending/<feature-name>.md` The `pending/` directory holds designs that haven't been implemented yet. After sprint completion, move to `docs/architecture/` if the design becomes a permanent reference, or delete if it was consumed by the sprint spec.
Related Skills
role-architect
System architect mode for designing interfaces, contracts, and architecture decisions.
arch-review
Review architecture documents against code implementation and principles.
verify-sprint
Spec-fidelity verification tracing requirements through code.
session
Analyze Claude Code session transcripts — search, summarize, list, or inspect how a session went.
role-educator
Course designer mode for creating exercises, configs, and QA criteria.
review-tests
Comprehensive test review using parallel test-reviewer agents.
review-sprint
Post-implementation mechanical audit of sprint deliverables.
issue
Create, list, and resolve review issues. Critical issues get individual files for research; warnings and gaps go to a quick-fix checklist.
implement-sprint
Automated sprint implementation with context discipline and quality gates.
get-started
Interactive setup guide for configuring CLAUDE.md and CAPABILITIES.md.
eval-sprint
Adversarial evaluation of sprint spec before implementation.
create-sprint
Guide sprint planning from scope assessment to spec artifacts.