AI Agent CLI Guide for a First Local Coding Task
See how a terminal AI coding agent reads, edits and tests a local project. Practice on a disposable folder with scoped permissions and a small reviewable diff.
What a CLI coding agent can do
Read project context
The agent can inspect files available to it to find the implementation and existing tests.
Propose or make changes
A tool may offer planning and editing modes. Read the permission settings before granting write access.
Run and report checks
A local test run gives you concrete output to compare with the expected behavior and file diff.
Practice on a local toy project
Greeting behavior
Change one blank-input case in a tiny Python project that has a unit test suite.
README correction
Update one usage example and compare it with the existing local script.
Test-only exercise
Add an edge-case test for a reversible local task and review the failure message before editing the code.
Run one scoped edit
Prepare the scratch folder
Create or copy a tiny repo containing greeting.py and its tests. Leave credentials and production files outside it.
Review permissions
Check the CLI’s docs for read, write, command and network scope. Begin with a plan before allowing the intended edit.
Check the evidence
Read git diff, run python3 -m unittest and compare the valid-name and blank-name outcomes. Record a clean screen demo only if useful.
Frequently Asked Questions
What complete prompt or practical brief can I copy?
Work only in this disposable local Git repository. Read greeting.py and the existing tests. First propose a plan for using “Hello, friend!” when the name is blank or whitespace. Then update only greeting.py and its test file, keeping the existing behavior for a valid name. Run python3 -m unittest locally. Use no network, credentials, installs or external services. Show the changed files, test result and Git diff for review before any commit.
What is an AI agent CLI?
Is a prompt the same as a permission rule?
Do I need dependencies or network access for this test?
What should I verify before committing?
Turn the local review into a short walkthrough
Use CapCut to trim a clean recording of the test result and diff after reviewing every frame for private information.