Give a coding agent a reproducible problem and review its actual edits and test output.
Coding & development · Documented · Documentation checked 2026-10-04
Claude Code in terminals and supported development surfaces; Claude chat is not a local shell.
Authentication, installed version, surface, subscription or API billing, and organization policy matter.
The worked example is illustrative and has not been tested in a signed-in product account.
What it does and how it connects
Give a coding agent a reproducible problem and review its actual edits and test output.
Claude Code operates in a development environment with files and tools. A local coding agent is different from attaching a source file to Claude chat. Its output should be a scoped change plus evidence from the actual environment, not only a plausible explanation.
Walkthrough
Use the product named in the access panel. The steps below combine documented controls with the editorial example.
- Install and authenticate through the official setup for your surface.
- Open an authorized sample repository and preserve the current working-tree state.
- Describe a reproducible problem and ask for inspection before editing.
- Review the diff and actual checks before committing or sharing the branch.
Worked example
Fix a sample form that submits twice on a double click. Provide the exact reproduction and expected single submission. The agent should inspect the existing event flow, change the relevant code, and verify the reported behavior.
Reproduce this sample double-submit bug: [steps]. Inspect the event flow first. Implement a scoped fix, preserve unrelated edits, and verify the original case plus a normal submission. Report changed files and actual check results. Do not push, merge, or deploy.
Inspect and verify the result
Inspect event handlers and the diff, repeat the reported scenario, and confirm the check actually executed. A successful edit does not prove the failure was understood or the existing behavior preserved.
Limits and troubleshooting
Dependencies and permissions can block checks. Keep blocked verification visible. A large ambiguous request invites unnecessary changes; split it into a plan and one testable implementation slice.
Data, permissions, and actions
Code and tool output enter the task context. Commands execute with allowed host access. Keep sensitive configuration and production services outside the sample exercise and review authority before expanding it.
Related reading
CLAUDE.md and skills
Keep project context distinct from a reusable task procedure loaded when needed.
A local change with explicit permission review
Implement a small sample fix and verify its behavior.
A read-only extension-assisted review
Use approved issue context and delegated review without external side effects.
Evaluations: does the system do the job?
Create representative tests for quality, tool behavior, refusal, escalation, and the effects of a change.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Claude Code overview · Anthropic
Configure permissions · Anthropic