Aider looping, repeating lint fixes or ignoring Ctrl-C?
This guide separates repeated Aider actions, recurring lint or test errors, and ineffective interrupts. Capture the repeated action, stop condition, and last useful repository change before starting again.
What the failure looks like.
Aider continues exploring, proposing, or applying similar changes without reaching a stable result or satisfying a stated completion condition.
Why it happens.
A broad goal without a stop condition can leave progress difficult to judge. For repeated lint/test feedback, use the lint-specific section below.
Fix it without losing the task.
What the next AI should receive.
The original goal, the repeated action and its output, current diff, decisions that still hold, approaches not to repeat, exact completion criterion, and one smallest next check.
First safe action
Stop after the repeated action yields no new result. Preserve the last useful diff and write down what the next attempt must prove.
Evidence boundary
Issue #3614 is one Aider report and is also referenced by the lint-loop page for context. It is not an independent second incident or evidence of prevalence.
Diagnostic checkpoint
If the same linter output is driving each edit, use the lint-specific section below; broader non-converging actions require a stated stop condition.
Source trail.
Aider issue #3614 and Aider usage documentation.
FAQ.
Does a long Aider run prove it is stuck?
No. Compare observable progress, changed files, and the stated completion condition. A run is not necessarily stuck simply because it takes longer than expected.
When should I use the lint-loop section?
Use that guide when the same lint or test result is repeatedly returned after edits. The lint-loop section distinguishes that case from broader non-converging actions.
Scope note: This is a task-control workflow for an Aider run that is not converging. It does not claim every long run is a product defect; the evidence cited is a single report.
Aider repeats the same lint or test failure.
Aider edits a file, runs a lint or test command, then receives the same failing result and makes another change without producing a new verification result.
- Record the check command, complete output, exit code, and whether the same diagnostic appears on each cycle.
- Compare diffs in order. Identify the first edit that repeats or reverses a previous change.
- Run the check independently only if it is safe and deterministic. If it fails without Aider, inspect the check configuration or environment before asking for another code edit.
- Resume with one targeted correction and a stated passing condition. Keep the lint/test enabled unless evidence shows the check itself is misconfigured.
Issues #1090 and #3614 are individual reports. They do not establish how often lint loops occur or make disabling lint the correct fix for other projects.
Aider ignores Ctrl-C or a child process stays active.
The terminal does not return to a prompt after Ctrl-C, or Aider appears to continue while a lint, test, or other child process remains active. These are different states and should be checked separately.
- Press Ctrl-C once and note whether Aider returns to its prompt, prints a cancellation, or leaves a child command in the foreground.
- Inspect the terminal process and child command separately. Stop only the confirmed active process using the controls for that terminal and OS.
- Reopen Aider only after checking the worktree. Verify whether the interrupted command changed files before rerunning it.
Issue #945 documents one reported case. It does not establish a general Aider defect or guarantee that terminating the process preserves unsaved terminal input.
Additional symptom sources.
How ShardStitch helps.
ShardStitch can turn the current goal, observed changes, failed approaches, and next check into a bounded handoff. It cannot cap Aiders runtime or decide whether an open-ended plan is complete.