Problem
The review agent's intent-coherence or correctness sub-agent raises premature-issue-closure findings when a PR closes an issue that has blocking references (e.g., blocked by #N) in its body. However, the agent relies solely on textual analysis of the issue body and does not verify the current state of the referenced blocking issues via the GitHub API.
This caused a false positive on fullsend PR #6434: the agent flagged the PR for closing fullsend#5115, claiming it was blocked by #5109, #5110, and #6133. All three had been closed since mid-August 2026, over a week before the PR was created. The finding was raised three times across review iterations 2, 4, and 6, consuming at least one full fix iteration of wasted rework.
Proposed change
In the review sub-agent that produces premature-issue-closure findings (likely the intent-coherence sub-agent):
- When the agent identifies blocking/prerequisite references (blocked by #N, depends on #N) in the linked issue body, it must query the GitHub API to check the current state of those referenced issues before raising a finding.
- If all referenced blockers are closed, suppress the premature-closure finding.
- If some blockers are still open, raise the finding naming specifically which ones remain open.
This could be implemented as a prompt instruction to the sub-agent to verify issue state before raising such findings, or as a pre-check step that resolves blocking references and injects their status into the sub-agent context.
Related issues
- agents#726 - correctness sub-agent verifying state of external issues/PRs referenced in code (different trigger: code comments vs issue body parsing)
- agents#1093 - fix and review agents checking if the linked issue itself is still open (different scope: obsolescence detection vs false positive suppression)
Validation criteria
On the next 5 PRs where the review agent encounters blocking issue references in the linked issue body: (1) if all referenced blockers are closed, no premature-issue-closure finding is raised; (2) if some blockers are still open, the finding names the specific open blockers.
Problem
The review agent's intent-coherence or correctness sub-agent raises premature-issue-closure findings when a PR closes an issue that has blocking references (e.g., blocked by #N) in its body. However, the agent relies solely on textual analysis of the issue body and does not verify the current state of the referenced blocking issues via the GitHub API.
This caused a false positive on fullsend PR #6434: the agent flagged the PR for closing fullsend#5115, claiming it was blocked by #5109, #5110, and #6133. All three had been closed since mid-August 2026, over a week before the PR was created. The finding was raised three times across review iterations 2, 4, and 6, consuming at least one full fix iteration of wasted rework.
Proposed change
In the review sub-agent that produces premature-issue-closure findings (likely the intent-coherence sub-agent):
This could be implemented as a prompt instruction to the sub-agent to verify issue state before raising such findings, or as a pre-check step that resolves blocking references and injects their status into the sub-agent context.
Related issues
Validation criteria
On the next 5 PRs where the review agent encounters blocking issue references in the linked issue body: (1) if all referenced blockers are closed, no premature-issue-closure finding is raised; (2) if some blockers are still open, the finding names the specific open blockers.