Aider

Aider is a terminal-native AI pair-programming harness. This entry covers its repository context, conventions, edit, and verification workflows, not the behavior of the model selected for a session.

• Add only the files that the task needs to edit, and use /read or --read for conventions and other context that must remain read-only.
• Keep repository guidance in a concise file such as CONVENTIONS.md, then load it explicitly or list it under read in the repository's .aider.conf.yml.
• Let the repository map supply related symbols and dependencies, but inspect and add the concrete files that must be changed when a task crosses module boundaries.
• Configure a project-appropriate --lint-cmd and --test-cmd; enable --auto-test when repeated post-edit verification is worth the runtime.
• Select an edit format that fits the model and file complexity. Prefer the default unless observed edit failures justify an explicit --edit-format or architect/editor split.
• Keep project configuration reviewable and avoid placing credentials in committed configuration.

• Assuming Aider automatically loads a conventional instruction filename; conventions are supplied through /read, --read, or the read configuration option.
• Adding the whole repository to the chat when a focused file set and the repository map provide sufficient context.
• Treating the repository map as a substitute for reading the implementation files that will be edited.
• Treating lint or test commands as harmless: they execute project tooling after edits and should be reviewed for side effects.
• Attributing context assembly, edit-format selection, or test-loop behavior to the model.

Aider separates editable chat files from read-only context. Its repository map supplies a compact, ranked view of related files and symbols, while /add or command-line file arguments make selected files editable. .aider.conf.yml can centralize read-only conventions, model-independent edit settings, and lint/test commands; repository-root configuration should be reviewed with the same care as other project automation.

Repository-map coverage is constrained by the configured token budget, and edit-format reliability varies by model and task. Verify the actual files in context and run the project’s validation commands after changes rather than relying on the map or a successful edit parse alone.