What happened
On PR #528, the review agent ran 30+ times between Jun 3 and Jul 3, 2026. Of the 18 most recent fullsend.yaml runs (Jul 1-3), exactly half (9) were triggered by pull_request_review events — the review bot's own submitted reviews triggering new dispatch cycles. Issue #1420 was filed on May 20 and closed on May 25 proposing to add a bot filter for pull_request_review events in the shim, mirroring the existing issue_comment bot filter. However, the fullsend.yaml shim in build-service (and at least 4 other enrolled repos: konflux-ci, multi-platform-controller, tekton-integration-catalog, mobster) still lacks this filter. The dispatch job's if condition only filters bots on issue_comment, not pull_request_review.
What could go better
The feedback loop doubles the number of review dispatches per PR. On this long-lived bot PR, that contributed to 30+ review cycles repeating the same findings, wasting tokens and creating noise. Confidence is high: the shim YAML shows the missing filter, and the 9 pull_request_review-triggered runs in 3 days confirm the loop is active. The fix proposed in #1420 was correct but was either never implemented or never propagated to enrolled repo shims. This may indicate the shim update mechanism (template sync or manual update) has a gap — the issue was closed without the fix reaching production.
Proposed change
Update the fullsend.yaml shim template (the canonical source in fullsend-ai/fullsend that enrolled repos copy) to add a bot filter for pull_request_review events. The dispatch job's if condition should change from:
if: >-
github.event_name != 'issue_comment'
|| github.event.comment.user.type != 'Bot'
to:
if: >-
(github.event_name != 'issue_comment'
|| github.event.comment.user.type != 'Bot')
&& (github.event_name != 'pull_request_review'
|| github.event.review.user.type != 'Bot')
Then propagate the updated shim to all enrolled repos. Also consider reopening or referencing #1420 to track that the fix was not actually deployed.
Validation criteria
After the shim update is deployed to build-service and other enrolled repos: (1) The review bot posting a review should NOT trigger a new fullsend.yaml run. Verify on the next 10 PRs that pull_request_review events from bot accounts are filtered out. (2) Per-PR fullsend.yaml run count should drop by ~50% (eliminating the review-triggered runs). (3) Spot-check at least 3 enrolled repos to confirm the shim was propagated.
Generated by retro agent from konflux-ci/build-service#528
What happened
On PR #528, the review agent ran 30+ times between Jun 3 and Jul 3, 2026. Of the 18 most recent fullsend.yaml runs (Jul 1-3), exactly half (9) were triggered by
pull_request_reviewevents — the review bot's own submitted reviews triggering new dispatch cycles. Issue #1420 was filed on May 20 and closed on May 25 proposing to add a bot filter forpull_request_reviewevents in the shim, mirroring the existingissue_commentbot filter. However, the fullsend.yaml shim in build-service (and at least 4 other enrolled repos: konflux-ci, multi-platform-controller, tekton-integration-catalog, mobster) still lacks this filter. The dispatch job'sifcondition only filters bots onissue_comment, notpull_request_review.What could go better
The feedback loop doubles the number of review dispatches per PR. On this long-lived bot PR, that contributed to 30+ review cycles repeating the same findings, wasting tokens and creating noise. Confidence is high: the shim YAML shows the missing filter, and the 9
pull_request_review-triggered runs in 3 days confirm the loop is active. The fix proposed in #1420 was correct but was either never implemented or never propagated to enrolled repo shims. This may indicate the shim update mechanism (template sync or manual update) has a gap — the issue was closed without the fix reaching production.Proposed change
Update the fullsend.yaml shim template (the canonical source in fullsend-ai/fullsend that enrolled repos copy) to add a bot filter for
pull_request_reviewevents. The dispatch job'sifcondition should change from:to:
Then propagate the updated shim to all enrolled repos. Also consider reopening or referencing #1420 to track that the fix was not actually deployed.
Validation criteria
After the shim update is deployed to build-service and other enrolled repos: (1) The review bot posting a review should NOT trigger a new fullsend.yaml run. Verify on the next 10 PRs that
pull_request_reviewevents from bot accounts are filtered out. (2) Per-PR fullsend.yaml run count should drop by ~50% (eliminating the review-triggered runs). (3) Spot-check at least 3 enrolled repos to confirm the shim was propagated.Generated by retro agent from konflux-ci/build-service#528