A headless Aider run that never reaches the model — every API call fails, every retry is exhausted — still exits with status code 0. The same code a successful run reports.
What happened
Filed as Aider-AI/aider#5552 by the affected user (noelsaw1) on 2026-08-11, with an exact reproduction: point AIDER_OPENAI_API_BASE at an invalid or unreachable host and invoke aider --message "test". Aider catches the underlying litellm connection exception, retries up to its configured limit, logs the failure to its own output — and then exits cleanly with code 0.
That's the whole bug. No edits are made, no API call succeeds, and the process still reports success to whatever invoked it.
Why it matters more than it looks
Exit code is the standard contract between a CLI tool and anything that orchestrates it — CI jobs, cron wrappers, agent pipelines chaining tool calls. Aider breaks that contract silently: a total connection failure is indistinguishable, from the caller's side, from a clean no-op run that had nothing to change.
This is also not an isolated slip. It's the third distinct report within the same week of an Aider code path swallowing a failure signal that should have reached the user or caller — following a valid diff silently discarded as "no tracked changes" (#5486) and a partial-hunk-application warning that fails to fire (#5573). Three different functions, same shape: the tool fails, and nothing downstream is told.
What the source establishes — and what it doesn't
The issue is filed by the person who hit it, with a command anyone can run to reproduce it, and it reproduces as described. That's verified and reproducible, which is why it's in this database at all.
What it does not establish: no maintainer response had landed as of publication, so there's no confirmed fix timeline and no confirmation this is intended retry behavior versus an oversight. No specific downstream incident — a bad deploy, a silently skipped pipeline stage — was reported alongside the filing. The damage here is the mechanism, demonstrated directly, not a documented production failure yet.
Status
Open, unconfirmed by maintainers, as of this writing. Any headless Aider invocation that gates its next step on the exit code is exposed to this today.
Full incident record and severity scoring (4.8 / medium): STUPID-2026-0081
This is one of 81+ verified, severity-scored AI agent incidents documented at StupidLLM, an open incident database for AI coding agent failures.
Top comments (0)