Codebase Review

Codebase Review

Conduct a thorough senior-architect-level review of a codebase. Evaluate 10 quality dimensions, assign a letter grade, and deliver a prioritized improvement roadmap with effort estimates. Designed to run after a major update or before a release milestone.

When to Activate

Manual triggers:

• "Code review" / "review the codebase" / "review this repo"
• "Architecture review" / "architecture audit"
• "Deep dive review"
• "Audit this code"
• "Give me a health check on this codebase"
• "Grade this codebase"

Auto-detect triggers:

• User says "we just finished a major refactor — can you take a look?"
• User asks "how is the overall quality of this project?"
• User is preparing for a release or handoff and wants a quality assessment

Phase 1: Orient (5-8 tool calls)

Read the project's structural foundation before sampling any code.

1.1 Project Manifest

Read these files if they exist (prioritize what's present):

• package.json / pyproject.toml / Cargo.toml / go.mod — deps, scripts, metadata
• tsconfig.json / tsconfig.*.json — TypeScript strictness settings
• .eslintrc* / biome.json / .prettierrc — linting and formatting config
• jest.config.* / vitest.config.* / pytest.ini — test framework and coverage config
• .env.example / docker-compose.yml / Dockerfile — environment and deploy config
• CLAUDE.md / README.md — documented conventions and onboarding story

1.2 Directory Tree

Use a shell command to get a two-level directory tree:

find . -maxdepth 2 -not -path '*/node_modules/*' -not -path '*/.git/*' \
  -not -path '*/dist/*' -not -path '*/.next/*' | sort

Identify the architectural layers present (e.g., main process, renderer, API, data layer, UI components, tests, build scripts).

1.3 First Impressions

Before reading any code, note:

• What type of project is this? (Electron app, web app, CLI, library, API, etc.)
• What's the rough scale? (file count, apparent complexity)
• What framework/stack is in use?
• Are tests present at all?
• Does a CI/CD pipeline exist?

Phase 2: Sample the Codebase (15-20 tool calls)

Read representative files from each architectural layer. The goal is breadth, not depth — you're looking for patterns, not every line.

Sampling Strategy

Entry points and orchestration:

• Main process / app entry (e.g., src/main.ts, main.py, cmd/main.go)
• Root router or controller
• App bootstrap / initialization code

Data layer:

• Schema definitions (Zod, Yup, Prisma, SQL, Pydantic, etc.)
• Data access / repository layer
• File I/O or database query code
• Any data parsing or transformation utilities

Business logic:

• The most complex service or module (largest file by lines, or most-imported)
• A utility module that is imported broadly
• Any state management (Redux, Zustand, MobX, Context, etc.)

UI / presentation layer (if applicable):

• A complex component with meaningful logic
• A form with validation
• Any component that handles user input

Security surface:

• IPC handlers (Electron), API routes, WebSocket handlers
• Authentication or session management code
• Any code that touches the filesystem, shell, or network
• Content Security Policy config (if web-facing)

Tests:

• One unit test file
• One integration test file (if present)
• One e2e test file (if present)

Build and deploy:

• CI/CD config (.github/workflows/, Jenkinsfile, etc.)
• Build script or bundler config
• Release / packaging config (e.g., electron-builder.yml)

Cap total tool calls for reading at 30. If the codebase is large, prefer reading more files briefly over reading fewer files completely.

Phase 3: Evaluate 10 Dimensions

Score each dimension A through F using the rubric below. Be honest — a C is normal for a production codebase under active development.

---

Dimension 1: Data Integrity & Validation

What to look for:

• Are external inputs (user input, file reads, IPC messages, API responses)

validated before use?

• Is there a schema validation library in use (Zod, Yup, Joi, Pydantic, etc.)?
• Are parse failures handled gracefully (no silent swallowing of errors)?
• Is there protection against data shape mismatches at runtime?

Grade guide:

• A: Schema validation at all external boundaries; parse errors surface clearly
• B: Most boundaries validated; occasional gaps
• C: Validation present in some layers; inconsistent
• D: Minimal validation; frequent unchecked assumptions
• F: No validation; data used as-is from external sources

---

Dimension 2: Test Coverage & Quality

What to look for:

• Is a test framework configured?
• Are tests present and non-trivial?
• Are unit, integration, and e2e tests all represented?
• Is code coverage tracked (even informally)?
• Do tests cover edge cases and failure paths, or only happy paths?

Grade guide:

• A: Solid unit + integration coverage; e2e for critical flows; coverage >70%
• B: Unit tests present; some integration; coverage 40–70%
• C: Some tests exist; mostly happy path; coverage 20–40%
• D: Minimal tests; mostly smoke tests or empty test files
• F: No tests, or tests are broken/skipped entirely

---

Dimension 3: Security Posture

What to look for:

• IPC handlers (Electron): contextIsolation, nodeIntegration settings;

preload script scope; ipcMain/ipcRenderer input validation

• Web apps: CSP headers, XSS vectors, CSRF protection
• Input sanitization: SQL injection, path traversal, shell injection
• Secrets management: hardcoded credentials, env var exposure
• Dependency hygiene: known CVEs in package-lock.json / lockfile

Grade guide:

• A: Defense-in-depth; all IPC/API inputs validated; no known CVEs; CSP present
• B: Most surfaces secured; minor gaps; no critical CVEs
• C: Basic security practices; some unsanitized inputs or missing CSP
• D: Known vectors present; nodeIntegration: true without guards; hardcoded secrets
• F: Severe vulnerabilities; no input validation; credentials in source

---

Dimension 4: Error Handling & Resilience

What to look for:

• Are async operations wrapped in try/catch or .catch()?
• Are errors logged with enough context to debug?
• Does the app degrade gracefully on failure (shows error UI vs. crashes)?
• Is there crash recovery or restart logic?
• Are unhandled promise rejections caught at the process level?

Grade guide:

• A: Consistent error handling; graceful degradation; errors logged with context
• B: Most async paths handled; occasional silent failures
• C: Inconsistent; some paths unhandled; errors swallowed in places
• D: Try/catch mostly absent; app crashes on common errors
• F: No error handling; frequent unhandled rejections

---

Dimension 5: Code Organization & DRY

What to look for:

• Is there a clear separation of concerns (presentation, logic, data)?
• Are shared utilities centralized, or is code copy-pasted across modules?
• Are module boundaries respected?
• Is there obvious dead code or commented-out blocks?
• Is naming consistent and descriptive?

Grade guide:

• A: Clear layers; DRY utilities; minimal dead code; logical module structure
• B: Mostly organized; some duplication; minor boundary leaks
• C: Mixed concerns in places; noticeable duplication; some dead code
• D: Significant duplication; unclear module responsibilities; large monolithic files
• F: No structure; everything in a few files; copy-paste throughout

---

Dimension 6: Type Safety

What to look for:

• Is TypeScript (or equivalent) in use?
• Is strict: true enabled in tsconfig.json?
• Are any types used frequently?
• Are type assertions (as SomeType) used to bypass the type system?
• Are runtime type checks present where TypeScript can't verify (e.g., parsed JSON)?

Grade guide:

• A: strict: true; minimal any; runtime type validation at boundaries
• B: TypeScript in use; some any or loose casts; mostly typed
• C: TypeScript present but strict: false; frequent any
• D: TypeScript in name only; any used broadly; types ignored
• F: No type system; or completely unenforced

---

Dimension 7: Performance & Scalability

What to look for:

• Are large lists virtualized (e.g., react-window, @tanstack/virtual)?
• Are expensive operations debounced or throttled?
• Is data loaded lazily where possible?
• Are there obvious memory leaks (event listeners not cleaned up, intervals not cleared)?
• Are synchronous blocking operations on the main thread?

Grade guide:

• A: Virtualization for large data; debouncing; lazy loading; no obvious leaks
• B: Performance considered; minor gaps; no blocking main-thread operations
• C: Functional but no optimization; potential memory leaks; no virtualization
• D: Blocking operations; re-renders on every keystroke; obvious leaks
• F: Unusable at scale; app hangs under normal load

---

Dimension 8: Accessibility

What to look for:

• Are ARIA attributes used on interactive elements?
• Is keyboard navigation supported (Tab order, focus management)?
• Are form inputs labeled (`, aria-label, aria-labelledby`)?
• Are color contrast ratios adequate?
• Are screen reader announcements provided for dynamic content?

Grade guide:

• A: ARIA throughout; full keyboard nav; labels on all inputs; dynamic announcements
• B: Most interactive elements accessible; some gaps
• C: Basic HTML semantics; minimal ARIA; keyboard nav partially works
• D: Custom UI components with no ARIA; keyboard nav broken
• F: Completely inaccessible; no semantic markup; mouse-only

---

Dimension 9: Build & Deploy Pipeline

What to look for:

• Is there a CI/CD pipeline (GitHub Actions, CircleCI, etc.)?
• Does the pipeline run tests on every PR/push?
• Is code signing configured for desktop/mobile releases?
• Is there an auto-update mechanism?
• Are build artifacts reproducible and versioned?

Grade guide:

• A: CI runs tests + lint on every PR; signed releases; auto-update; versioned artifacts
• B: CI present; some steps missing (e.g., signing); mostly automated
• C: Basic CI; tests not always run; manual release steps
• D: No CI; manual build and release process
• F: No build pipeline; ad-hoc releases

---

Dimension 10: Documentation & Onboarding

What to look for:

• Does the README explain what the project is, how to install, and how to run?
• Are non-obvious architectural decisions documented (ADRs, inline comments)?
• Is there a CONTRIBUTING.md or developer setup guide?
• Are complex functions/modules commented where the logic isn't self-evident?
• Could a new developer be productive within a day?

Grade guide:

• A: Excellent README; architecture docs; ADRs; meaningful inline comments
• B: Good README; some architecture context; partial inline docs
• C: Basic README; sparse comments; some setup gaps
• D: Minimal README; no inline docs; onboarding requires tribal knowledge
• F: No documentation; undiscoverable project

---

Phase 4: Compute the Overall Grade

Convert letter grades to a GPA-style scale:

• A = 4.0, A- = 3.7
• B+ = 3.3, B = 3.0, B- = 2.7
• C+ = 2.3, C = 2.0, C- = 1.7
• D+ = 1.3, D = 1.0
• F = 0.0

Average the 10 dimension scores. Convert back to a letter grade:

• 3.7–4.0 = A, 3.3–3.6 = A-
• 3.0–3.2 = B+, 2.7–2.9 = B, 2.3–2.6 = B-
• 2.0–2.2 = C+, 1.7–1.9 = C, 1.3–1.6 = C-
• 1.0–1.2 = D+, 0.7–0.9 = D
• Below 0.7 = F

Phase 5: Prioritized Improvement Plan

Identify the top 10 improvements. Rank by combined impact × urgency.

For each item, provide:

• Priority: Critical / High / Medium / Low
• Effort: X days (be specific — 0.5, 1, 2, 3, 5, 8, 13)
• Dimension: Which of the 10 dimensions this addresses
• What: One sentence describing the concrete change
• Why it matters: One sentence on the consequence of not fixing it

Prioritization heuristics:

• Security vulnerabilities → always Critical
• Missing test coverage on core paths → High
• Missing error handling on user-visible paths → High
• Accessibility blockers → High (legal risk + user impact)
• Performance issues under realistic load → High or Medium
• Documentation gaps → Medium (unless blocking onboarding)
• Code style / DRY issues → Low (unless blocking future work)

Output Format

Write the review in this exact structure:

---

# Codebase Review: [Project Name]

**Date:** [today's date]
**Reviewer:** Claude (Senior Architect Mode)
**Project Type:** [Electron app / Web app / CLI / API / Library / etc.]
**Stack:** [primary languages, frameworks, key libraries]

---

## Overall Grade: [LETTER]

> One paragraph executive summary: what this codebase does well, what its
> biggest risks are, and what kind of team/maturity level it reflects.

---

## Dimension Scores

| # | Dimension | Grade | Summary |
|---|-----------|-------|---------|
| 1 | Data Integrity & Validation | [grade] | [one line] |
| 2 | Test Coverage & Quality | [grade] | [one line] |
| 3 | Security Posture | [grade] | [one line] |
| 4 | Error Handling & Resilience | [grade] | [one line] |
| 5 | Code Organization & DRY | [grade] | [one line] |
| 6 | Type Safety | [grade] | [one line] |
| 7 | Performance & Scalability | [grade] | [one line] |
| 8 | Accessibility | [grade] | [one line] |
| 9 | Build & Deploy Pipeline | [grade] | [one line] |
| 10 | Documentation & Onboarding | [grade] | [one line] |

---

## Dimension Deep Dives

### 1. Data Integrity & Validation — [grade]

[2–4 sentences. What you found. Specific file/pattern examples. Concrete gap.]

### 2. Test Coverage & Quality — [grade]

[2–4 sentences...]

[... repeat for all 10 dimensions ...]

---

## Top 10 Improvement Plan

| # | Priority | Effort | Dimension | What | Why It Matters |
|---|----------|--------|-----------|------|----------------|
| 1 | Critical | X days | [dim] | [what] | [why] |
| 2 | High | X days | [dim] | [what] | [why] |
| ... | | | | | |

---

## Appendix: Files Reviewed

- `path/to/file.ts` — [why sampled]
- ...

---

Tone and Standards

• Write as a senior architect doing a genuine peer review, not as a tool running

a checklist. Use judgment.

• Be specific: name the files, patterns, and line-level examples you found.
• Be honest about grades. A C is not an insult — it is the median production codebase.
• Avoid generic advice. Every recommendation should be actionable with a clear

next step.

• If a dimension cannot be evaluated due to missing files or context, grade it

I (Incomplete) and explain what information is missing.

• Do not pad the review with praise. Be direct about both strengths and gaps.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ryankolean
• Source: ryankolean/summit-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.