Repository navigation
fix(automation): stop watcher alerts on availability and doc churn - #108
Merged
Merged
Conversation
The 12-hourly free-model check treated any non-2xx answer as a model failure, so an upstream 402 "Endpoint is unavailable" or a 503 blip filed an issue that needed manual triage and closed itself hours later. The daily upstream watcher filed on release tags, docs edits and free-model listing churn, none of which is actionable. The health check now classifies every sample: only a model-level answer (400/401/403/404 without the shared availability body, 410) alerts. Rate limits, server faults, network errors and the endpoint-unavailable answer are retried, reported in the run summary and never open an issue; a probe-cap timeout is reported as inconclusive. An existing alert issue is edited in place instead of collecting a comment every run. The upstream watcher now alerts only on session-schema changes, keeps informational changes in the run summary, advances that baseline every run so nothing repeats, and offers a manual re-baseline dispatch input.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What was wrong
The two scheduled watchers filed issues that needed manual triage but described nothing actionable:
longcat-2.5-preview-free(HTTP 402,Upstream request failed: Endpoint is unavailable.) andflint-40b-free(a 503 that cleared on the next sample). Ten such issues in eight days, each closed once upstream recovered.cat-jam-freeandnemotron-3-ultra-free, while both answer 200.What changed
Health check classifies every sample and only alerts on a model-level answer (400/401/403/404 without the shared availability body, 410). Rate limits, server faults, network errors and the endpoint-unavailable answer are retried (three samples), reported in the run summary with a
::warning::annotation, and never open an issue. A probe-cap timeout is reported as inconclusive, not as a verdict. An existing alert issue is edited in place instead of collecting a comment every 12 hours.Upstream watcher alerts only on a session-schema change (
identifier.ts/session-id.ts), which is the one signal that can break compatibility. Release tags, docs edits and listing churn land in the run summary as informational, and the baseline advances on every successful read so nothing repeats daily. Arebaselinedispatch input re-establishes the baseline and closes an open alert.Verification
npm run typecheckclean; full suite1160 pass / 0 fail(1161 tests).longcat402 retried three times and classified unavailable — this exact state filed 🚨 OpenCode Free Tier Model Disruption (2026-10-22 10:57 UTC) #107 under the old code.v1.19.3→v2.1.3, listing changed (four models dropped, two added) → informational report, exit 0, no issue.Closes #107 (evidence comment follows).