[JacobStephens2]

Summary

After a Self-merge, thirdshift no longer tries to predict whether GitHub will close the issue. It closes the issue itself unless the issue is already closed.

 after_merge
   delete Issue branch on origin
-  close_unless_merge_does
-    if merge_closes_issue (gh pr view --json baseRefName,closingIssuesReferences + gh repo view)
-       or issue is closed: return
+  close_unless_closed
+    if issue is closed: return
     gh issue close --comment "Closed by #<pr>, merged into <base> by a thirdshift Merge run."

• github::merge_closes_issue is removed. Its closingIssuesReferences field doesn't exist in gh 2.45.
• In the fake gh, pr merge no longer closes the issues a PR body names, since GitHub may not close them either. --json now rejects unknown fields with gh's Unknown JSON field error, on pr view (with and without --jq), pr list and issue view. The fake's repo view is gone because nothing calls it any more.
• The Self-merge entry in CONTEXT.md now says: "the issue closed by thirdshift, unless it is already closed."

Evidence

• Before: the new tests failed on main's code:

  a_merge_into_the_default_branch_closes_the_issue_github_leaves_open
    comments["7"]: Null != ["Closed by #1, merged into main by a thirdshift Merge run."]
  no_gh_call_asks_for_the_prs_closing_issue_references
    gh pr view issue-7 --json baseRefName,closingIssuesReferences asks for a field gh 2.45 does not have
  

  After: cargo test passes every suite, and cargo clippy --all-targets and cargo fmt --check are clean. New or changed tests in tests/merge_run.rs:

  merge into main, PR says "Closes #7", fake leaves #7 open  -> #7 CLOSED with thirdshift's comment, no warning
  merge, after-merge hook closes #7 first                    -> no gh issue close, no comment
  every gh call in a Merge run                               -> none mentions closingIssuesReferences
  

  The existing tests for a merge into a non-default base and for a failed gh issue close (exit 0 plus the "could not close issue Numbered Issue branches after a merged or closed PR #7…" warning) pass unchanged.

Merge Danger

Door: two-way

A revert brings back the old behaviour. The only lasting effect is a duplicate close comment, in the accepted race where GitHub closes the issue at the moment thirdshift checks.

Blast Radius: small

It only affects the post-merge close step of a Merge run, which never fails the Run. Issues that GitHub closes may occasionally get an extra thirdshift comment as well.

Unaddressed findings

Standards

• The explicit closingIssuesReferences test duplicates the fake's rejection. Kept on purpose: the spec asks that "a test fails if one does", and a named test says why better than a warning buried in stderr.
• "default branch" in test names now suggests a distinction the code no longer makes. Kept: the names match the acceptance criteria and pin the exact scenario from Release script resumes an interrupted release #158 and One Run notification per Spec run #148.
• The closing comment for main is written inline instead of through a helper. It's a single use, moved from a deleted test, so a helper isn't worth it.
• "gh 2.45" is pinned in both the fake's docstring and the test message. It documents the real version seen in the report. Out of scope: the spec rules out setting a minimum gh version.

Spec

None unaddressed. The fake's partial rejection of unknown fields was fixed across pr view --jq, pr list and issue view.

Closes #181

🤖 Generated with Claude Code