말하는 순간 바로 쓰는 Claude Code 실전 프롬프트.

Target application: Claude Code
Target input: terminal prompt

Task: Fix the bug described in my dictation.

Problem:
- Summarize the observed behavior precisely.

Expected behavior:
- State what should happen instead.

Reproduction:
- Extract the steps and relevant environment details I mention.

Constraints:
- Preserve existing behavior outside the fix.
- Do not invent missing requirements.

Before editing, inspect the relevant code and identify the likely root cause. Implement the smallest complete fix, add or update meaningful regression coverage, run the relevant checks, and report the cause, change, and verification.

Application-specific guidance:
- Inspect repository state, failing command output and relevant logs first; do not run destructive shell commands while investigating.

Target application: Claude Code
Target input: terminal prompt

Goal:
- Turn my description into one clear product outcome.

User flow:
- Capture the trigger, expected behavior, empty/loading/error states, and completion state.

Constraints:
- Preserve the current design system and established architecture.
- Reuse existing components and services where appropriate.
- Do not invent scope I did not request.

Inspect the current implementation first. Then implement the complete user-facing flow, verify it with the most relevant checks, and explain any material tradeoffs.

Application-specific guidance:
- Implement through the existing command-line workflow, keep generated files deliberate and stop before deployment.

Target application: Claude Code
Target input: terminal prompt

Refactor target:
- Identify the code and problem described in my dictation.

Invariants:
- Preserve observable behavior, public interfaces, analytics names, and persisted data unless I explicitly request a change.

Desired result:
- State the simpler structure or responsibility boundary I want.

Inspect callers and tests before editing. Make the refactor in reviewable steps, remove obsolete code, run focused regression checks, and call out any behavior that could not be proven equivalent.

Application-specific guidance:
- Trace shell scripts, configuration and callers before moving responsibilities or renaming public interfaces.

Target application: Claude Code
Target input: terminal prompt

Behavior under test:
- Extract the contract and failure modes from my dictation.

Test cases:
- Cover the main success path, the stated edge cases, and the regression that motivated this task.

Requirements:
- Test observable behavior rather than implementation details.
- Reuse the repository's existing test style and fixtures.
- Avoid tests that merely repeat the production code.

Inspect the implementation and nearby tests, add the smallest meaningful test set, run it, and report what the tests prove.

Application-specific guidance:
- Run the narrowest meaningful test command first, then only broaden coverage when the change justifies it.

Target application: Claude Code
Target input: terminal prompt

Review the changes I describe or provide.

Focus on:
- User-visible regressions and incorrect behavior.
- Data loss, security, privacy, concurrency, and migration risks.
- Missing validation for behavior that materially changed.

Check the surrounding code and established project conventions before judging the diff. Report only actionable findings, ordered by severity, with a concrete trigger and consequence for each. If there are no findings, say so and state the remaining verification limits.

Application-specific guidance:
- Review the git diff, untracked files and command effects; report concrete findings with file locations.

Target application: Claude Code
Target input: terminal prompt

Performance problem:
- Capture where the slowdown occurs, how often, and what the user experiences.

Evidence:
- Record the measurements, logs, device details, and reproduction conditions I mention.

Investigate before optimizing. Identify the dominant cost, distinguish measured facts from hypotheses, and avoid broad rewrites. Implement the highest-impact safe improvement, compare before and after with the same workload, and report the measured result and remaining bottlenecks.

Application-specific guidance:
- Measure the same command or workload before and after, and distinguish tool startup time from application runtime.

Target application: Claude Code
Target input: terminal prompt

Objective:
- Restate the outcome from my dictation.

Current context:
- Capture the relevant system, constraints, and decisions already made.

Produce a concrete implementation plan based on the actual codebase. Name the components that need to change, dependencies between steps, migration or rollout concerns, and the checks that prove completion. Resolve routine choices from repository evidence and list only uncertainties that truly block execution.

Application-specific guidance:
- Produce an ordered terminal-safe plan with dependencies, migration concerns and validation commands.

Target application: Claude Code
Target input: terminal prompt

Screen or flow:
- Identify the interface and user goal from my dictation.

Problems noticed:
- Capture the visual, interaction, responsive, and accessibility issues I mention.

Preserve the product's existing visual language. Inspect the screen at its real desktop and mobile sizes, fix hierarchy, spacing, typography, states, and interaction details, then compare the result in the running product. Avoid decorative changes that do not improve comprehension or use.

Application-specific guidance:
- When UI files change, verify the rendered result instead of treating a clean terminal build as visual proof.