Known Issues in Published Releases
As of August 19, 2026, npm ships 0.3.0, which includes every fix listed below — this page is now a historical record, not an active caveat. If you're on 0.1.0 or 0.2.0, upgrade (npm install ai-review-bot@latest) rather than building from main.
Bugs present in the published releases, fixed on main
These four bugs were in code that already existed at the 0.2.0 tag — anyone who installed 0.1.0 or 0.2.0 was exposed to them.
| Severity | Symptom | Fixed by |
|---|---|---|
| P1 | Review posts nothing and the webhook still returns 202 when every agent throws — a total silent failure with no visible error anywhere | PR #31 |
| P1 | Codex bot produces no reviews at all — AI_NoObjectGeneratedError on every agent | PR #19 |
| P1 | No idempotency claim before agents run — concurrent triggers could post duplicate reviews on the same PR | PR #18 |
| P1 | Review body rendered as one giant Markdown heading (setext-heading corruption) instead of the intended sections | PR #48 |
Bugs in features added after 0.2.0 (never shipped in a published release)
These six were introduced and fixed entirely on main, in code that didn't exist yet at the 0.2.0 tag (the QStash scheduler, the reaction/feedback signal system, and the incremental-review triage gate were all added afterward). No published version was ever exposed to them — listed here for completeness and because they explain a lot of main's recent history, not as a reason to avoid 0.2.0.
| Severity | Symptom | Fixed by |
|---|---|---|
| P1 | QStash deduplicationId contained a :, so every scheduled-review publish failed and the scheduler silently fell back to the legacy inline path on every PR | PR #30 |
| P0 | Codex bot misreported "out of credits" as a generic rate limit (both arrive as HTTP 429) — masked a billing problem as a transient retry-later error | PR #36 |
| P1 | The 😕 "confused" reaction on a finding was silently dropped from the signal model instead of being tracked as a distinct verdict | PR #32 |
| P1 | Blocking (REQUEST_CHANGES) reviews rendered no reason on the incremental re-review path — a review could block a merge with no visible explanation | PR #54 |
| P1 | Summary prose could contradict the "Still open" carry-forward table across re-reviews — misleading about whether a finding was actually resolved | PR #61 |
| P1 | Stuck-loop / incremental-review-state issue — the bot could keep re-flagging findings across re-reviews instead of recognizing they'd already been addressed | PR #21, PR #22 |
See also the dedicated post-mortems for the two most severe historical incidents:
- Post-mortem: OpenSSL PKCS#1 — total loss of review output on first deploy
- Post-mortem: Reviewer Hallucinations
Why there's no newer tag yet
This list exists because publishing was tracked separately from fixing — bugs got fixed and merged to main without a corresponding version bump + npm publish. Cutting a new release (v0.3.0+) is the real fix for this page needing to exist at all; this table is a stopgap until that happens.