update

GiggleLiu · GiggleLiu · commit 8623328e95de · 2026-03-19T23:23:38.000+08:00
diff --git a/.claude/skills/issue-to-pr/SKILL.md b/.claude/skills/issue-to-pr/SKILL.md
@@ -76,7 +76,7 @@ For `[Rule]` issues, `ISSUE_JSON` already includes `source_problem`, `target_pro
 
 ### 4. Research References
 
-Use `WebSearch` and `WebFetch` to look up the reference URL provided in the issue. This helps:
+Use web search to look up the reference URL provided in the issue. This helps:
 - Clarify the formal problem definition and notation
 - Understand the reduction algorithm in detail (variable mapping, penalty terms, proof of correctness)
 - Resolve any ambiguities in the issue description without bothering the contributor
@@ -104,8 +104,8 @@ Include the concrete details from the issue (problem definition, reduction algor
 - Otherwise, ensure the information provided is enough to implement a solver.
 
 **Example rules:**
-- Implement the user-provided example instance as an example program in `examples/`.
-- Run the example; verify JSON output against user-provided information.
+- Implement the user-provided example instance in the canonical `example_db` path for the issue (`src/example_db/model_builders.rs` or `src/example_db/rule_builders.rs`, as appropriate).
+- Run the relevant export and fixture regeneration steps; verify the generated example data against the user-provided information.
 - Present in `docs/paper/reductions.typ` in tutorial style with clear intuition (see KColoring->QUBO section for reference).
 
 ### 6. Create PR (or Resume Existing)
@@ -180,7 +180,7 @@ If execution fails, leave the PR open with the plan commit only — the user can
 
 Structural and quality review is handled by the `review-pipeline` stage, not here. The run stage just needs to produce working code.
 
-**Commit all changes** (implementation):
+Ensure all implementation changes are committed before cleanup. A small coherent commit stack is acceptable, especially when resuming an existing PR or integrating subagent work; do not rewrite history just to collapse commits. If there are still uncommitted implementation changes, commit them now:
 ```bash
 git add -A
 git commit -m "Implement #<number>: <title>"
diff --git a/.claude/skills/run-pipeline/SKILL.md b/.claude/skills/run-pipeline/SKILL.md
@@ -167,7 +167,7 @@ This handles the full pipeline: fetch issue, verify Good label, research, write
 
 ### 4. Move to "Review pool"
 
-After `issue-to-pr` succeeds, move the issue to the `Review pool` column:
+After `issue-to-pr` fully succeeds, move the issue to the `Review pool` column. "Fully succeeds" means the implementation work is committed, the temporary plan file has been deleted, the PR implementation summary comment has been posted, the branch has been pushed, and the working tree is clean aside from ignored/generated files:
 
 ```bash
 python3 scripts/pipeline_board.py move <ITEM_ID> review-pool
diff --git a/scripts/pipeline_pr.py b/scripts/pipeline_pr.py
@@ -49,17 +49,73 @@ def is_bot_login(login: str) -> bool:
 
 
 def extract_linked_issue_number(title: str | None, body: str | None) -> int | None:
+    issue_numbers = extract_linked_issue_numbers(title, body)
+    return issue_numbers[0] if issue_numbers else None
+
+
+def extract_linked_issue_numbers(title: str | None, body: str | None) -> list[int]:
+    issue_numbers: list[int] = []
+
+    def append_unique(number: int) -> None:
+        if number not in issue_numbers:
+            issue_numbers.append(number)
+
     for text in [body or "", title or ""]:
-        match = _CLOSING_ISSUE_RE.search(text)
-        if match:
-            return int(match.group(1))
+        for match in _CLOSING_ISSUE_RE.finditer(text):
+            append_unique(int(match.group(1)))
 
     for text in [body or "", title or ""]:
-        match = _GENERIC_ISSUE_RE.search(text)
-        if match:
-            return int(match.group(1))
+        for match in _GENERIC_ISSUE_RE.finditer(text):
+            append_unique(int(match.group(1)))
+
+    return issue_numbers
+
+
+def extract_linked_issue_number_from_pr_data(pr_data: dict | None) -> int | None:
+    issue_numbers = extract_linked_issue_numbers_from_pr_data(pr_data)
+    return max(issue_numbers) if issue_numbers else None
+
+
+def extract_linked_issue_numbers_from_pr_data(pr_data: dict | None) -> list[int]:
+    issue_numbers: list[int] = []
+
+    def append_unique(number: int) -> None:
+        if number not in issue_numbers:
+            issue_numbers.append(number)
+
+    if pr_data:
+        for issue in pr_data.get("closingIssuesReferences") or []:
+            number = issue.get("number")
+            if number is not None:
+                append_unique(int(number))
+
+    if issue_numbers:
+        return issue_numbers
+
+    if not pr_data:
+        return []
+
+    return extract_linked_issue_numbers(pr_data.get("title"), pr_data.get("body"))
+
+
+def _normalized_match_text(text: str | None) -> str:
+    return " ".join((text or "").lower().split())
 
-    return None
+
+def score_linked_issue_candidate(pr_data: dict, issue: dict) -> tuple[int, int]:
+    score = 0
+    pr_title = _normalized_match_text(pr_data.get("title"))
+    pr_body = _normalized_match_text(pr_data.get("body"))
+    issue_title = _normalized_match_text(issue.get("title"))
+    issue_number = int(issue.get("number") or -1)
+
+    if issue_title:
+        if issue_title in pr_title:
+            score += 100
+        if issue_title in pr_body:
+            score += 40
+
+    return score, issue_number
 
 
 def normalize_issue_thread_comment(comment: dict) -> dict:
@@ -292,10 +348,7 @@ def build_snapshot(
     codecov_summary: dict | None = None,
 ) -> dict:
     if linked_issue_number is None:
-        linked_issue_number = extract_linked_issue_number(
-            pr_data.get("title"),
-            pr_data.get("body"),
-        )
+        linked_issue_number = extract_linked_issue_number_from_pr_data(pr_data)
 
     labels = [label.get("name") for label in pr_data.get("labels", []) if label.get("name")]
     files = [
@@ -428,7 +481,11 @@ def build_context_result(
 
 def build_pr_context(repo: str, pr_number: int) -> dict:
     snapshot = build_pr_snapshot(repo, pr_number)
-    comments = build_comments_summary(repo, pr_number)
+    comments = build_comments_summary(
+        repo,
+        pr_number,
+        linked_issue_number=snapshot.get("linked_issue_number"),
+    )
     linked_issue_result = build_linked_issue_context(
         repo,
         pr_number,
@@ -478,7 +535,8 @@ def fetch_pr_data(repo: str, pr_number: int) -> dict:
         "--json",
         (
             "number,title,body,labels,files,additions,deletions,commits,"
-            "headRefName,baseRefName,headRefOid,url,state,mergeable,author"
+            "headRefName,baseRefName,headRefOid,url,state,mergeable,author,"
+            "closingIssuesReferences"
         ),
     )
 
@@ -562,10 +620,12 @@ def fetch_check_runs(repo: str, head_sha: str) -> dict:
 
 
 def fetch_linked_issue_bundle(repo: str, pr_data: dict) -> tuple[int | None, dict | None]:
-    issue_number = extract_linked_issue_number(pr_data.get("title"), pr_data.get("body"))
-    if issue_number is None:
+    issue_numbers = extract_linked_issue_numbers_from_pr_data(pr_data)
+    if not issue_numbers:
         return None, None
-    return issue_number, fetch_issue_data(repo, issue_number)
+    issues = [fetch_issue_data(repo, issue_number) for issue_number in issue_numbers]
+    best_issue = max(issues, key=lambda issue: score_linked_issue_candidate(pr_data, issue))
+    return int(best_issue["number"]), best_issue
 
 
 def fetch_ci_summary(repo: str, pr_number: int, pr_data: dict | None = None) -> dict:
@@ -580,12 +640,16 @@ def fetch_ci_summary(repo: str, pr_number: int, pr_data: dict | None = None) ->
     return summary
 
 
-def build_comments_summary(repo: str, pr_number: int, pr_data: dict | None = None) -> dict:
+def build_comments_summary(
+    repo: str,
+    pr_number: int,
+    pr_data: dict | None = None,
+    *,
+    linked_issue_number: int | None = None,
+) -> dict:
     pr_data = pr_data or fetch_pr_data(repo, pr_number)
-    linked_issue_number = extract_linked_issue_number(
-        pr_data.get("title"),
-        pr_data.get("body"),
-    )
+    if linked_issue_number is None:
+        linked_issue_number = extract_linked_issue_number_from_pr_data(pr_data)
 
     summary = summarize_comments(
         inline_comments=fetch_inline_comments(repo, pr_number),
diff --git a/scripts/test_pipeline_pr.py b/scripts/test_pipeline_pr.py
@@ -15,6 +15,7 @@
     edit_pr_body,
     extract_codecov_summary,
     extract_linked_issue_number,
+    fetch_linked_issue_bundle,
     format_issue_context,
     post_pr_comment,
     parse_args,
@@ -88,6 +89,87 @@ def test_extract_linked_issue_number_falls_back_to_title(self) -> None:
         )
         self.assertEqual(linked_issue, 117)
 
+    @mock.patch("pipeline_pr.fetch_issue_data")
+    def test_fetch_linked_issue_bundle_prefers_closing_issue_references(
+        self,
+        fetch_issue_data: mock.Mock,
+    ) -> None:
+        fetch_issue_data.return_value = {
+            "number": 210,
+            "title": "[Model] Partition",
+        }
+
+        issue_number, issue = fetch_linked_issue_bundle(
+            "CodingThrust/problem-reductions",
+            {
+                "title": "Fix #19: stale title reference",
+                "body": "This PR still mentions Fixes #19 in copied text.",
+                "closingIssuesReferences": [
+                    {"number": 210, "title": "[Model] Partition"},
+                ],
+            },
+        )
+
+        self.assertEqual(issue_number, 210)
+        self.assertEqual(issue, {"number": 210, "title": "[Model] Partition"})
+        fetch_issue_data.assert_called_once_with("CodingThrust/problem-reductions", 210)
+
+    @mock.patch("pipeline_pr.fetch_issue_data")
+    def test_fetch_linked_issue_bundle_checks_all_candidates_and_prefers_title_match(
+        self,
+        fetch_issue_data: mock.Mock,
+    ) -> None:
+        issues = {
+            19: {"number": 19, "title": "[Rule] Coloring -> ILP"},
+            210: {"number": 210, "title": "[Model] Partition"},
+        }
+        fetch_issue_data.side_effect = lambda repo, number: issues[number]
+
+        issue_number, issue = fetch_linked_issue_bundle(
+            "CodingThrust/problem-reductions",
+            {
+                "title": "Fix #19: [Model] Partition",
+                "body": "Also closes #210.",
+                "closingIssuesReferences": [
+                    {"number": 19},
+                    {"number": 210},
+                ],
+            },
+        )
+
+        self.assertEqual(issue_number, 210)
+        self.assertEqual(issue, {"number": 210, "title": "[Model] Partition"})
+        self.assertEqual(fetch_issue_data.call_args_list, [
+            mock.call("CodingThrust/problem-reductions", 19),
+            mock.call("CodingThrust/problem-reductions", 210),
+        ])
+
+    @mock.patch("pipeline_pr.fetch_issue_data")
+    def test_fetch_linked_issue_bundle_uses_latest_issue_when_scores_tie(
+        self,
+        fetch_issue_data: mock.Mock,
+    ) -> None:
+        issues = {
+            19: {"number": 19, "title": "[Model] Old"},
+            210: {"number": 210, "title": "[Model] Newer"},
+        }
+        fetch_issue_data.side_effect = lambda repo, number: issues[number]
+
+        issue_number, issue = fetch_linked_issue_bundle(
+            "CodingThrust/problem-reductions",
+            {
+                "title": "Refresh implementation",
+                "body": "Touches multiple linked issues.",
+                "closingIssuesReferences": [
+                    {"number": 19},
+                    {"number": 210},
+                ],
+            },
+        )
+
+        self.assertEqual(issue_number, 210)
+        self.assertEqual(issue, {"number": 210, "title": "[Model] Newer"})
+
     def test_summarize_comments_splits_human_copilot_and_codecov_sources(self) -> None:
         summary = summarize_comments(
             inline_comments=[
@@ -205,6 +287,48 @@ def test_build_snapshot_includes_linked_issue_ci_and_codecov(self) -> None:
         self.assertEqual(snapshot["counts"]["files"], 1)
         self.assertEqual(snapshot["counts"]["commits"], 2)
 
+    def test_build_snapshot_prefers_closing_issue_references_when_inferring_link(self) -> None:
+        snapshot = build_snapshot(
+            {
+                "number": 664,
+                "title": "Fix #19: stale title reference",
+                "body": "Copied text still says Fixes #19.",
+                "state": "OPEN",
+                "url": "https://github.com/CodingThrust/problem-reductions/pull/664",
+                "headRefName": "issue-210-partition-v2",
+                "baseRefName": "main",
+                "mergeable": "CONFLICTING",
+                "headRefOid": "abc123",
+                "labels": [],
+                "files": [],
+                "commits": [],
+                "closingIssuesReferences": [{"number": 210}],
+            }
+        )
+
+        self.assertEqual(snapshot["linked_issue_number"], 210)
+
+    def test_build_snapshot_uses_latest_linked_issue_number_when_multiple_candidates_exist(self) -> None:
+        snapshot = build_snapshot(
+            {
+                "number": 700,
+                "title": "Refresh implementation",
+                "body": "References multiple issues.",
+                "state": "OPEN",
+                "url": "https://github.com/CodingThrust/problem-reductions/pull/700",
+                "headRefName": "refresh-linked-issues",
+                "baseRefName": "main",
+                "mergeable": "MERGEABLE",
+                "headRefOid": "def456",
+                "labels": [],
+                "files": [],
+                "commits": [],
+                "closingIssuesReferences": [{"number": 19}, {"number": 210}],
+            }
+        )
+
+        self.assertEqual(snapshot["linked_issue_number"], 210)
+
     def test_build_current_pr_context_includes_repo_and_pr_fields(self) -> None:
         current = build_current_pr_context(
             "CodingThrust/problem-reductions",
@@ -406,7 +530,11 @@ def test_build_pr_context_assembles_existing_helper_results(
         context = build_pr_context("CodingThrust/problem-reductions", 570)
 
         build_pr_snapshot.assert_called_once_with("CodingThrust/problem-reductions", 570)
-        build_comments_summary.assert_called_once_with("CodingThrust/problem-reductions", 570)
+        build_comments_summary.assert_called_once_with(
+            "CodingThrust/problem-reductions",
+            570,
+            linked_issue_number=117,
+        )
         build_linked_issue_context.assert_called_once()
         self.assertEqual(context["pr_number"], 570)
         self.assertEqual(context["issue_context_text"], "# [Model] GraphPartitioning")
