Implement a small sample fix and verify its behavior.
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.
An illustrative recipe. This website does not run the automation.
The goal and the right kind of system
Implement a small sample fix and verify its behavior.
An interactive agent fits the coding loop; the person retains command and change review.
Inputs and connected services
Inputs: authorized sample repo, reproduction, CLAUDE.md, and configured tool permissions.
Services: Claude Code and local test tools; no external write connection needed.
The steps
An illustrative sequence to adapt and test before using real data or enabling ongoing work.
- Read project context and inspect the reported behavior.
- Plan the narrow fix within allowed tools.
- Review commands and run the relevant checks.
- Inspect the diff and reproduce the original case.
Outputs and human checkpoints
Outputs: scoped patch, actual check outcomes, and remaining uncertainty.
Human checkpoints: inspect commands and final changes before commit.
Exceptions, permissions, and stopping conditions
Permission boundary: honor configured deny / ask rules. A denied tool is not an invitation to use another path.
Stop on missing reproduction, failed checks, broad unexpected edits, or blocked necessary access.
Adaptable prompt
Implement [sample fix] within existing permissions. Read project guidance, preserve unrelated edits, and verify the reported behavior. Explain blocked tools without bypassing them. No push, merge, deployment, or external writes.
Related reading
Local repository work and review
Give a coding agent a reproducible problem and review its actual edits and test output.
CLAUDE.md and skills
Keep project context distinct from a reusable task procedure loaded when needed.
Permission modes and stopping rules
Use enforceable tool policies rather than relying on prose instructions as access controls.
Designing an AI workflow
Make a useful task repeatable with clear inputs, steps, outputs, exceptions, and human checkpoints.
Permissions and meaningful approvals
Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Claude Code overview · Anthropic
Configure permissions · Anthropic
Features and extension relationships · Anthropic
Skills: instructions and resources · Anthropic