AI coding tools
Aider Operating Card
Use Aider as a Git-aware terminal pair programmer with an explicit model, file set, test loop, and recovery checkpoint.
- Format
- Reference
- Level
- Beginner
- Audience
- Practitioner, Developer
- Owner
- Project42 Editorial
- Review cadence
- Every 30 days
- Prerequisites
- A supported Python environment; A version-controlled working copy; An approved model-provider credential path
Use it for a Git-centered terminal pairing loop
Aider is a terminal pair-programming tool that edits files in a local Git repository and can connect to multiple model providers. Choose it when you want a conversational editing loop with visible repository changes and locally controlled tests.
Install through a supported isolated Python path, choose the provider and model deliberately, and keep credentials in the supported environment or configuration mechanism rather than prompts or committed files.
Bound files, commits, and validation
Begin from a reviewed Git status and add only the files needed for the task. State the outcome, conventions, test command, and files that must not change. Inspect Aider's edits and commits as evidence, not as proof that behavior is correct.
Run the smallest relevant lint or test during the loop and the repository's release gate before accepting. Recover from the last known Git checkpoint when changes cross scope or verification fails.
Task: [one observable code outcome]
Scope: [repository, branch, and files allowed]
Permissions: [local edit, model access, shell/test limits]
Model: [approved provider and model]
Conventions: [repository instructions]
Verification: [diff review and exact checks]
Stop conditions: [scope drift, secret request, failing invariant]
Recovery: [Git checkpoint and safe retry]Expected evidence and verification
Expected evidence includes the starting Git state, selected model, files placed in context, reviewed diff, and exact test output. Verify that unrelated files, secrets, and dependency caches were not added.
If verification fails, preserve the failure, return only the affected files to a reviewed checkpoint, narrow the file set or prompt, and rerun the deterministic check.