Week 45, updated 2 Nov 2026

AI use cases Coding agents AI pair programming in the terminal

AI pair programming in the terminal

Work with an AI coding agent from your terminal: describe a change, review the diff, commit. Example prompts, a workflow and the best-scored tools.

Tools compared
5
Open source
4of 5
Reviewed
16 Oct

For: Developers

Best-scored tools for this jobscore, safety
  1. 1Gemini CLIAI tool70Caution
  2. 2OpenCodeAI tool70No issues found
  3. 3Claude CodeAI tool65No dependencies scanned
  4. 4CodexAI tool58Caution
  5. 5AiderAI tool31Caution
What the score and shapes mean

Score = public maintenance, adoption and docs signals, as a percentile of 52 tools. Nobody here tested these hands-on. Shapes are the separate safety label: No issues found, No dependencies scanned, Caution.

What it does for you

Developers who are comfortable with git and a command line and want to hand off routine edits without leaving the terminal.

  • Explain an unfamiliar codebase and find where a feature lives
  • Implement a small feature or bug fix across several files
  • Run the test suite, read the failures and fix them
  • Write commit messages and pull request descriptions from the diff
  • Draft one-off scripts and shell commands, then explain them
More about this job

A terminal coding agent reads your repository, proposes edits across several files and can run your build or tests. You stay in charge: you describe the task, read the diff and decide what gets committed.

Workflow in 5 steps

  1. Commit or stash your work so the agent starts from a clean tree, ideally on a new branch.
  2. Start the agent in the repository root and ask it to explain the relevant code first.
  3. Give one small, testable task with clear acceptance criteria rather than a vague goal.
  4. Let it edit, then read the full diff yourself and run the tests.
  5. Ask for follow-up fixes in the same session, then commit with a message you have read.
Full steps
  1. Commit or stash your work so the agent starts from a clean tree, ideally on a new branch.
  2. Start the agent in the repository root and ask it to explain the relevant code first; correct it if it misreads the project.
  3. Give one small, testable task with clear acceptance criteria rather than a vague goal.
  4. Let it edit, then read the full diff yourself and run the tests.
  5. Ask for follow-up fixes in the same session, then commit with a message you have read.

Prompts to copy

Starting points we wrote, not tested results.

Read the repository and explain how a request flows from the router to the database. List the files involved and do not change anything.

2 more prompts

Add input validation to the signup handler so empty and malformed emails return a 400 with a clear message. Update or add tests, run them and show me the diff.

The test tests/test_billing.py::test_refund fails. Find the cause, propose the smallest fix and explain why it is correct before you edit.

Watch out for

  • Treat AI changes like a pull request from a new colleague: read the diff, run the tests and keep a human sign-off before anything ships.
  • Agents that can run commands can delete or overwrite files.
  • Keep API keys, tokens and customer data out of prompts and out of files the agent can read.
  • Agent loops can burn through paid API tokens quickly.
  • Large tasks in one prompt tend to wander.
  • Licence to check before commercial use: Claude Code (Proprietary (Anthropic Commercial Terms)).
  • CautionGemini CLI9 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • No issues foundOpenCodeNo issues found in this week’s scan
  • No dependencies scannedClaude CodeCode checks clean, no lockfile to check
  • CautionCodex4 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • CautionAider18 critical-severity known vulnerabilities in runtime dependencies (OSV)
Why these warnings?
  • Treat AI changes like a pull request from a new colleague: read the diff, run the tests and keep a human sign-off before anything ships.
  • Agents that can run commands can delete or overwrite files. Work in a branch or sandbox, commit first, and approve destructive commands by hand.
  • Keep API keys, tokens and customer data out of prompts and out of files the agent can read. Use a separate, low-privilege key for any agent.
  • Agent loops can burn through paid API tokens quickly. Set spending limits with your model provider and watch the first few runs.
  • Large tasks in one prompt tend to wander. Split work into steps you can review one at a time.
  • Gemini CLI: 9 critical-severity known vulnerabilities in runtime dependencies (OSV); Skill/agent file instructs piping a remote script into a shell: .gemini/skills/agent-tui/SKILL.md — curl -fsSL https://raw.githubusercontent.com/pproenca/agent-tui/master/install.sh | sh
  • Claude Code: code and instruction-file checks found nothing, but there is no runtime lockfile or manifest, so dependencies were not checked
  • Codex: 4 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • Aider: 18 critical-severity known vulnerabilities in runtime dependencies (OSV)

Recommended tools

From our ranked data, highest score first. A high score does not mean a clean scan.

  • 70Gemini CLIAI tool · Apache-2.0 · #4 overall
  • 70OpenCodeAI tool · MIT · #5 overall
  • 65Claude CodeAI tool · Proprietary (Anthropic Commercial Terms) · #11 overall
  • 58CodexAI tool · Apache-2.0 · #16 overall
  • 31AiderAI tool · Apache-2.0 · #45 overall

Related jobs

Sources for tool facts

google-gemini/gemini-cli · anomalyco/opencode · anthropics/claude-code · openai/codex · Aider-AI/aider. Data fetched 2 Nov 2026. Scores and labels are automated, relative and not a hands-on test.

Get the board every Monday

One email a week: the new top 10, the biggest score changes and every safety label that changed.

Free. Double opt-in: you confirm by email first, unsubscribe in one click.