Fix stale approved follow-up issue creation

README.md

@@ -195,9 +195,10 @@ include small, localized, low-risk current-PR cleanup under:
 Same-PR follow-ups are sent back to the coder in the existing PR and require a
 new review round. They should stay narrowly scoped to files already touched by
 the PR or directly adjacent code; larger redesigns and independent work belong
-under Future follow-ups. Future follow-ups are retained and processed only after
-final approval. The issue modes create at most three follow-up issues per
-approved round to avoid issue noise.
+under Future follow-ups. Future follow-ups are processed only from the final
+approval round, so reviewers should restate any still-relevant future work after
+same-PR updates. The issue modes create at most three follow-up issues per final
+approval round to avoid issue noise.
 
 By default, `--approved-followups=ignore` asks reviewers not to include
 approved-review follow-up sections. Reviewers should mark the review blocking

docs/local_agent_loop.md

@@ -387,9 +387,10 @@ under:
 Same-PR follow-ups are sent back to the coder in the existing PR and require a
 new review round. They should stay narrowly scoped to files already touched by
 the PR or directly adjacent code; larger redesigns and independent work belong
-under Future follow-ups. Future follow-ups are retained and processed only
-after the final approval round. Without a `fix-and-*` mode, reviewers should
-mark same-PR cleanup blocking instead.
+under Future follow-ups. Future follow-ups are processed only from the final
+approval round, so reviewers should restate any still-relevant future work after
+same-PR updates. Without a `fix-and-*` mode, reviewers should mark same-PR
+cleanup blocking instead.
 
 By default, `--approved-followups=ignore` asks reviewers not to include these
 sections. Reviewers should mark the review blocking instead when cleanup should

src/coding_review_agent_loop/orchestrator.py

@@ -367,7 +367,6 @@ def run_pr_loop(
         # Backward-compatible single-reviewer resume support: older callers
         # pass one reviewer session, so attach it to the first configured reviewer.
         reviewer_session_ids[configured_reviewers[0]] = reviewer_session_id
-    pending_future_followups: list[ApprovedFollowup] = []
 
     for round_number in range(1, config.max_rounds + 1):
         coder_name = agent_display_name(config.coder)
@@ -424,7 +423,7 @@ def run_pr_loop(
                         )
 
         if not blocking_reviews and not same_pr_followups:
-            approved_followups = [*pending_future_followups, *round_future_followups]
+            approved_followups = round_future_followups
             if config.approved_followups in ("summarize", "fix-and-summarize") and approved_followups:
                 body = _format_approved_followup_summary(pr_number, approved_followups)
                 post_pr_comment(runner, config=config, pr_number=pr_number, body=body)
@@ -451,7 +450,6 @@ def run_pr_loop(
             )
 
         if same_pr_followups and not blocking_reviews:
-            pending_future_followups.extend(round_future_followups)
             combined_review = _format_same_pr_followups(same_pr_followups)
             followup_prompt = build_same_pr_followup_prompt(
                 pr_number,

tests/test_agent_loop.py

@@ -1084,7 +1084,7 @@ def test_pr_loop_caps_approved_followup_issues(tmp_path):
     assert issue_summary.endswith("-- coding-review-agent-loop")
 
 
-def test_pr_loop_fix_and_summarize_sends_same_pr_followups_to_coder_then_rereviews(tmp_path):
+def test_pr_loop_fix_and_summarize_ignores_future_followups_before_same_pr_rereview(tmp_path):
     runner = FakeRunner(
         codex_outputs=[
             "Codex approves with cleanup.\n\n"
@@ -1103,20 +1103,18 @@ def test_pr_loop_fix_and_summarize_sends_same_pr_followups_to_coder_then_rerevie
 
     agent_commands = [cmd[:2] for cmd, _cwd in runner.commands if cmd[:1] in (["claude"], ["codex"])]
     assert agent_commands == [["codex", "exec"], ["claude", "--print"], ["codex", "exec"]]
-    assert len(runner.comments) == 4
+    assert len(runner.comments) == 3
     followup_prompt = next(
         cmd[-1] for cmd, _cwd in runner.commands if cmd[:1] == ["claude"] and "Same-PR follow-ups" in cmd[-1]
     )
     assert "Rename the helper before merge." in followup_prompt
     assert "small, localized cleanup for the\ncurrent PR" in followup_prompt
     assert "Keep the change narrowly scoped to the listed items" in followup_prompt
     assert "Do not take on\nlarger redesigns or unrelated future work" in followup_prompt
-    summary = runner.comments[-1]
-    assert summary.startswith("Approved-review future follow-ups for PR #77:")
-    assert "- Add broader integration coverage later. (Codex)" in summary
+    assert runner.comments[-1] == "Codex approves final pass.\n<!-- AGENT_STATE: approved -->\n-- OpenAI Codex"
 
 
-def test_pr_loop_fix_and_issue_retains_future_followups_across_same_pr_round(tmp_path):
+def test_pr_loop_fix_and_issue_ignores_stale_future_followups_after_same_pr_rereview(tmp_path):
     runner = FakeRunner(
         codex_outputs=[
             "Codex approves with cleanup.\n\n"
@@ -1133,14 +1131,13 @@ def test_pr_loop_fix_and_issue_retains_future_followups_across_same_pr_round(tmp
 
     assert run_pr_loop(runner, pr_number=77, config=config) == 0
 
-    assert len(runner.issues) == 1
-    assert runner.issues[0]["title"] == "Follow up future review note: Add a separate migration dry-run command."
-    assert "- https://github.com/OWNER/REPO/issues/99" in runner.comments[-1]
+    assert runner.issues == []
+    assert runner.comments[-1] == "Codex approves final pass.\n<!-- AGENT_STATE: approved -->\n-- OpenAI Codex"
     commands = [cmd[:3] for cmd, _cwd in runner.commands]
-    assert commands.count(["gh", "issue", "create"]) == 1
+    assert commands.count(["gh", "issue", "create"]) == 0
 
 
-def test_pr_loop_fix_and_summarize_retains_future_followups_from_multiple_reviewers(tmp_path):
+def test_pr_loop_fix_and_summarize_only_uses_final_round_future_followups(tmp_path):
     runner = FakeRunner(
         codex_outputs=[
             "Codex approves with cleanup.\n\n"
@@ -1149,14 +1146,20 @@ def test_pr_loop_fix_and_summarize_retains_future_followups_from_multiple_review
             "### Future follow-ups\n"
             "- Add Codex's larger follow-up later.\n"
             "<!-- AGENT_STATE: approved -->\n-- OpenAI Codex",
-            "Codex approves final pass.\n<!-- AGENT_STATE: approved -->\n-- OpenAI Codex",
+            "Codex approves final pass.\n\n"
+            "### Future follow-ups\n"
+            "- Add Codex's final follow-up later.\n"
+            "<!-- AGENT_STATE: approved -->\n-- OpenAI Codex",
         ],
         claude_outputs=[
             "Claude approves.\n\n"
             "### Future follow-ups\n"
             "- Add Claude's larger follow-up later.\n"
             "<!-- AGENT_STATE: approved -->\n-- Anthropic Claude",
-            "Claude approves final pass.\n<!-- AGENT_STATE: approved -->\n-- Anthropic Claude",
+            "Claude approves final pass.\n\n"
+            "### Future follow-ups\n"
+            "- Add Claude's final follow-up later.\n"
+            "<!-- AGENT_STATE: approved -->\n-- Anthropic Claude",
         ],
         gemini_outputs=["Added assertion.\n<!-- AGENT_STATE: blocking -->\n-- Google Gemini"],
     )
@@ -1178,8 +1181,10 @@ def test_pr_loop_fix_and_summarize_retains_future_followups_from_multiple_review
         ["claude", "--print"],
     ]
     summary = runner.comments[-1]
-    assert "- Add Codex's larger follow-up later. (Codex)" in summary
-    assert "- Add Claude's larger follow-up later. (Claude)" in summary
+    assert "- Add Codex's final follow-up later. (Codex)" in summary
+    assert "- Add Claude's final follow-up later. (Claude)" in summary
+    assert "Add Codex's larger follow-up later." not in summary
+    assert "Add Claude's larger follow-up later." not in summary
 
 
 def test_pr_loop_reruns_all_reviewers_when_any_reviewer_blocks(tmp_path):