Claude Code Cheat Sheet

A practical QA guide for using Claude Code as a repo-aware assistant for debugging, automation work, reviews, and structured investigation.

AI tools + articles Repo-aware workflows MCP friendly

Good Use Cases

Claude Code is strongest when the work depends on reading the real project context.

- test review
- flaky test diagnosis
- framework refactors
- log analysis
- requirement traceability

State the goal, scope, and what kind of output you want back.

Review the Playwright checkout tests.
Focus on flakiness, bad waits, weak assertions, and maintainability.
Return findings ordered by severity.

Use it to catch mechanical issues before human review spends time on them.

Ask for:
- brittle locators
- hard waits
- duplicate setup
- missing assertions
- framework anti-patterns

Feed it logs, stack traces, route failures, or CI artifacts and ask for hypotheses first.

Good sequence:
1. summarize failure
2. propose likely causes
3. suggest checks
4. propose fix options

MCP makes tool access structured instead of requiring manual copy-paste for every step.

Useful MCP flows:
- file reads
- browser automation
- issue tracker lookups
- local tooling integration

Keep destructive actions explicit and review code changes before trusting them.

- ask for plan before refactor
- require diffs
- keep secrets out of prompts
- verify generated code

Claude Code is especially good when prompts are scoped to concrete quality concerns.

- review this test file
- explain this CI failure
- generate missing assertions
- compare these two implementations

Use it to accelerate investigation, not to bypass engineering judgment.

Good balance:
AI speeds up the mechanical work.
Humans approve the final decision.