fix skill

Author: GiggleLiu

.claude/skills/add-model/SKILL.md

@@ -67,6 +67,16 @@ Read these first to understand the patterns:
 - **CLI creation:** `problemreductions-cli/src/commands/create.rs`
 - **Canonical model examples:** `src/example_db/model_builders.rs`
 
+## Pre-review Checklist
+
+Before implementing, make sure the plan explicitly covers these items that structural review checks later:
+- `ProblemSchemaEntry` metadata is complete for the current schema shape (`display_name`, `aliases`, `dimensions`, and constructor-facing `fields`)
+- `declare_variants!` is present with the correct `opt`/`sat` marker and exactly one `default` variant when multiple concrete variants exist
+- CLI discovery and `pred create <ProblemName>` support are included where applicable
+- A canonical model example is registered for example-db / `pred create --example`
+- `docs/paper/reductions.typ` adds both the display-name dictionary entry and the `problem-def(...)`
+- `src/unit_tests/trait_consistency.rs` is updated
+
 ## Step 1: Determine the category
 
 Choose the appropriate sub-module under `src/models/`:
@@ -107,6 +117,7 @@ Create `src/models/<category>/<name>.rs`:
 ```
 
 Key decisions:
+- **Schema metadata:** `ProblemSchemaEntry` must reflect the current registry schema shape, including `display_name`, `aliases`, `dimensions`, and constructor-facing `fields`
 - **Optimization problems:** `type Metric = SolutionSize<W::Sum>`, implement `OptimizationProblem` with `direction()`
 - **Satisfaction problems:** `type Metric = bool`, implement `SatisfactionProblem` (marker trait)
 - **Weight management:** use inherent methods (`weights()`, `set_weights()`, `is_weighted()`), NOT traits

.claude/skills/issue-to-pr/SKILL.md

@@ -223,8 +223,18 @@ EOF
 python3 scripts/pipeline_pr.py comment --repo "$REPO" --pr "$PR" --body-file "$COMMENT_FILE"
 rm -f "$COMMENT_FILE"
 
+# Repo verification may regenerate tracked exports (notably after `make paper`).
+# Inspect the tree once more before pushing.
+git status --short
+
+# If expected generated exports changed, stage them before push.
+git add docs/src/reductions/problem_schemas.json docs/src/reductions/reduction_graph.json
+
+# The issue plan file must be gone before push.
+test ! -e docs/plans/<plan-file>.md
+
 git push
-make copilot-review
+gh copilot-review "$PR"
 ```
 
 #### 7e. Done
@@ -280,3 +290,5 @@ Run /review-pipeline to process Copilot comments, fix CI, and run agentic tests.
 | Dirty working tree | Use `pipeline_worktree.py prepare-issue-branch` — it stops before branching if the worktree is dirty |
 | Bundling model + rule in one PR | Each PR must contain exactly one model or one rule — STOP and block if model is missing (Step 3.5) |
 | Plan files left in PR | Delete plan files before final push (Step 7c) |
+| `make paper` or export steps changed tracked JSON after verification | Run `git status --short`, stage expected generated exports, and STOP if unexpected files remain before push |
+| `make copilot-review` cannot find a resumed PR branch | Request review with `gh copilot-review "$PR"` so resumed local branch names do not matter |

.claude/skills/project-pipeline/SKILL.md

@@ -147,7 +147,17 @@ WORKTREE_DIR=$(printf '%s\n' "$WORKTREE" | python3 -c "import sys,json; print(js
 cd "$WORKTREE_DIR"
 ```
 
+If `worktree-for-issue` fails during the GraphQL-backed open-PR lookup, fall back to a REST search instead of creating a fresh branch blindly:
+
+```bash
+PR_JSON=$(gh api "search/issues?q=repo:$REPO+is:pr+is:open+%22Fix+%23$ISSUE%22")
+PR_NUMBER=$(printf '%s\n' "$PR_JSON" | python3 -c "import sys,json; items=json.load(sys.stdin).get('items', []); print(items[0]['number'] if items else '')")
+WORKTREE=$(python3 scripts/pipeline_worktree.py checkout-pr \
+  --repo "$REPO" --pr "$PR_NUMBER" --repo-root . --format json)
+```
+
 - `action == "resume-pr"`: existing PR checked out — `issue-to-pr` will skip plan creation and jump to execution
+- If the resumed branch later proves stale against `main` (for example, merge conflicts span registry/CLI/skill wiring or a merge helper reports `likely_complex: true`), STOP autonomous repair and move the issue to `Final review` for human triage instead of forcing a large migration inside project-pipeline
 - `action == "create-worktree"`: fresh branch from `origin/main`
 
 All subsequent steps run inside the worktree. This ensures the user's main checkout is never modified.
@@ -246,3 +256,4 @@ Completed: 2/4 | Review pool: 3 | Returned to Ready: 1
 | Working in main checkout | All work happens in `.worktrees/` — never modify the main checkout |
 | Missing items from project board | `gh project item-list` defaults to 30 items — always use `--limit 500` |
 | Creating a fresh branch when PR exists | Check `issue-context` action field first — use `checkout-pr` for existing PRs instead of `create-issue` |
+| Forcing a very stale resume PR through autonomous conflict repair | If a resumed PR has broad merge conflicts (`likely_complex: true`, registry/CLI/skill churn, etc.), stop and move it to `Final review` |

.claude/skills/review-pipeline/SKILL.md

@@ -99,6 +99,7 @@ The report is the Step 0 packet. It should already include:
 - Comment Summary
 - CI / Coverage
 - Merge Prep
+- PR head branch
 - Linked Issue Context
 
 Branch from the report:
@@ -118,7 +119,7 @@ The bundle already handled the mechanical claim step:
 - the selected reviewed PR is claimed through the review queue
 - explicit `--pr` matches on ambiguous cards are treated as deterministic disambiguation and claimed automatically
 
-When you need to take actions later, use the identifiers already printed in the report (`Board item`, `PR`, worktree path). If you need structured data, prefer parsing the existing packet or ensure Step 0 was generated as JSON once; do not rerun Step 0 just to switch formats inside the same invocation.
+When you need to take actions later, use the identifiers already printed in the report (`Board item`, `PR`, worktree path, `PR head branch`). If you need structured data, prefer parsing the existing packet or ensure Step 0 was generated as JSON once; do not rerun Step 0 just to switch formats inside the same invocation.
 
 All subsequent steps run inside the prepared worktree and should read facts from the report instead of re-fetching them by default.
 
@@ -132,7 +133,7 @@ All subsequent steps run inside the prepared worktree and should read facts from
 
 Read the merge result from the report's `Merge Prep` section.
 
-- If the report says the merge status is `clean`: push the merge commit and continue.
+- If the report says the merge status is `clean`: push the merge commit to the packet's `PR head branch` and continue.
 - If there are conflicts:
   1. Inspect the conflicting files listed in the report.
   2. Compare the current skill versions on main vs the PR branch to understand which patterns are current.
@@ -146,11 +147,13 @@ Use the report as the primary mechanical context:
 - `Comment Summary`
 - `CI / Coverage`
 - `Linked Issue Context`
+- `Merge Prep` (`PR head branch`)
 
 Inspect the report's Copilot comment count and linked issue context. If there are actionable comments: invoke `/fix-pr` to address them, then push:
 
 ```bash
-git push
+HEAD_REF_NAME=<from the review-pipeline packet>
+git push origin HEAD:"$HEAD_REF_NAME"
 ```
 
 If Copilot approved with no actionable comments: skip to next step.
@@ -196,7 +199,9 @@ Run agentic feature tests on the modified feature:
    - `[Model]` PRs: the new problem model name
    - `[Rule]` PRs: the new reduction rule (source -> target)
 
-2. **Invoke `/agentic-tests:test-feature`** with the identified feature. This simulates a downstream user exercising the feature from docs and examples. You MUST use the Skill tool to invoke `agentic-tests:test-feature`.
+2. **Invoke `/agentic-tests:test-feature`** with the identified feature. This simulates a downstream user exercising the feature from docs and examples.
+   - In environments with a Skill tool, use that tool.
+   - In Codex, open `~/.claude/commands/agentic-tests:test-feature.md` directly and follow it instead of assuming slash-command support.
 
 3. **If test-feature reports issues:** treat every reported issue as real until you have checked it in the **current PR worktree/branch**.
    - Reproduce each issue from the current PR branch/worktree before acting. If it does not reproduce there, classify it as `not reproducible in current worktree`.
@@ -238,7 +243,8 @@ print(mapping.get(state, 'FAILED'))
 
 3. **Push fixes:**
    ```bash
-   git push
+   HEAD_REF_NAME=<from the review-pipeline packet>
+   git push origin HEAD:"$HEAD_REF_NAME"
    ```
 
 4. Increment retry counter. If `< 3`, go back to step 1. If `= 3`, give up.

scripts/pipeline_board.py

@@ -1232,7 +1232,10 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
     ack_parser.add_argument("item_id")
 
     list_parser = subparsers.add_parser("list")
-    list_parser.add_argument("mode", choices=["ready", "in-progress", "review-pool", "review"])
+    list_parser.add_argument(
+        "mode",
+        choices=["ready", "in-progress", "review-pool", "review", "final-review"],
+    )
     list_parser.add_argument("--repo")
     list_parser.add_argument("--owner", default="CodingThrust")
     list_parser.add_argument("--project-number", type=int, default=8)
@@ -1338,6 +1341,13 @@ def main(argv: list[str] | None = None) -> int:
                 batch_pr_fetcher=batch_fetch_prs_with_reviews,
             )
             return print_candidate_list(args.mode, items, fmt=args.format)
+        if args.mode == "final-review":
+            items = status_items(
+                board_data,
+                STATUS_FINAL_REVIEW,
+                content_types={"Issue", "PullRequest"},
+            )
+            return print_candidate_list(args.mode, items, fmt=args.format)
         raise SystemExit(f"Unsupported list mode: {args.mode}")
 
     if args.mode in {"review", "final-review"} and not args.repo:

scripts/pipeline_checks.py

@@ -587,19 +587,42 @@ def fetch_issue(repo: str, issue_number: int) -> dict:
     )
 
 
+def fetch_existing_prs_via_search(repo: str, issue_number: int) -> list[dict]:
+    query = f'repo:{repo} is:pr is:open "Fix #{issue_number}"'
+    data = run_gh_json("api", "search/issues", "-f", f"q={query}")
+    items = data.get("items", [])
+    prs: list[dict] = []
+    for item in items:
+        number = item.get("number")
+        if number is None or "pull_request" not in item:
+            continue
+        pr = run_gh_json("api", f"repos/{repo}/pulls/{number}")
+        prs.append(
+            {
+                "number": pr.get("number"),
+                "headRefName": (pr.get("head") or {}).get("ref"),
+                "url": pr.get("html_url"),
+            }
+        )
+    return prs
+
+
 def fetch_existing_prs(repo: str, issue_number: int) -> list[dict]:
-    data = run_gh_json(
-        "pr",
-        "list",
-        "--repo",
-        repo,
-        "--state",
-        "open",
-        "--search",
-        f"Fix #{issue_number}",
-        "--json",
-        "number,headRefName,url",
-    )
+    try:
+        data = run_gh_json(
+            "pr",
+            "list",
+            "--repo",
+            repo,
+            "--state",
+            "open",
+            "--search",
+            f"Fix #{issue_number}",
+            "--json",
+            "number,headRefName,url",
+        )
+    except subprocess.CalledProcessError:
+        return fetch_existing_prs_via_search(repo, issue_number)
     if not isinstance(data, list):
         raise ValueError(f"Unexpected PR list payload for issue #{issue_number}: {data!r}")
     return data

scripts/pipeline_skill_context.py

@@ -225,6 +225,8 @@ def render_review_pipeline_text(result: dict) -> str:
     )
     if checkout.get("worktree_dir"):
         lines.append(f"- Worktree: `{checkout['worktree_dir']}`")
+    if checkout.get("head_ref_name"):
+        lines.append(f"- PR head branch: `{checkout['head_ref_name']}`")
     conflicts = merge.get("conflicts") or []
     if conflicts:
         lines.append("- Conflicts:")

scripts/test_pipeline_board.py

@@ -210,6 +210,46 @@ def fake_pr_state_fetcher(repo: str, pr_number: int) -> str:
             )
             self.assertIsNone(no_item)
 
+    def test_parse_args_accepts_final_review_list_mode(self) -> None:
+        import pipeline_board
+
+        args = pipeline_board.parse_args(["list", "final-review", "--repo", "CodingThrust/problem-reductions"])
+
+        self.assertEqual(args.command, "list")
+        self.assertEqual(args.mode, "final-review")
+
+    def test_main_lists_final_review_items(self) -> None:
+        import pipeline_board
+
+        board_data = {
+            "items": [make_issue_item("PVTI_30", 239, status="Final review", title="[Model] BalancedCompleteBipartiteSubgraph")]
+        }
+
+        with (
+            patch.object(pipeline_board, "fetch_board_items", return_value=board_data) as fetch_board_items,
+            patch.object(pipeline_board, "print_candidate_list", return_value=0) as print_candidate_list,
+        ):
+            rc = pipeline_board.main(
+                ["list", "final-review", "--repo", "CodingThrust/problem-reductions", "--format", "json"]
+            )
+
+        self.assertEqual(rc, 0)
+        fetch_board_items.assert_called_once()
+        print_candidate_list.assert_called_once_with(
+            "final-review",
+            [
+                {
+                    "number": 239,
+                    "issue_number": 239,
+                    "pr_number": None,
+                    "status": "Final review",
+                    "title": "[Model] BalancedCompleteBipartiteSubgraph",
+                    "item_id": "PVTI_30",
+                }
+            ],
+            fmt="json",
+        )
+
 
 class ReviewCandidateQueueTests(unittest.TestCase):
     def test_select_entry_from_entries_tracks_pending_until_ack(self) -> None:

scripts/test_pipeline_checks.py

@@ -1,6 +1,8 @@
 #!/usr/bin/env python3
+import subprocess
 import tempfile
 import unittest
+from unittest import mock
 from pathlib import Path
 
 import pipeline_checks
@@ -16,6 +18,46 @@
 
 
 class PipelineChecksTests(unittest.TestCase):
+    @mock.patch("pipeline_checks.run_gh_json")
+    def test_fetch_existing_prs_falls_back_to_rest_search_on_pr_list_failure(
+        self,
+        run_gh_json: mock.Mock,
+    ) -> None:
+        run_gh_json.side_effect = [
+            subprocess.CalledProcessError(1, ["gh", "pr", "list"]),
+            {
+                "items": [
+                    {
+                        "number": 223,
+                        "pull_request": {
+                            "url": "https://api.github.com/repos/CodingThrust/problem-reductions/pulls/223"
+                        },
+                    }
+                ]
+            },
+            {
+                "number": 223,
+                "head": {"ref": "issue-212-multiprocessor-scheduling"},
+                "html_url": "https://example.test/pull/223",
+            },
+        ]
+
+        prs = pipeline_checks.fetch_existing_prs(
+            "CodingThrust/problem-reductions",
+            212,
+        )
+
+        self.assertEqual(
+            prs,
+            [
+                {
+                    "number": 223,
+                    "headRefName": "issue-212-multiprocessor-scheduling",
+                    "url": "https://example.test/pull/223",
+                }
+            ],
+        )
+
     def test_detect_scope_reports_model_review_for_new_model_file(self) -> None:
         scope = detect_scope_from_paths(
             added_files=["src/models/graph/graph_partitioning.rs"],

scripts/test_pipeline_skill_context.py

@@ -178,7 +178,10 @@ def test_emit_result_prints_review_pipeline_text_report(self) -> None:
             },
             "prep": {
                 "ready": False,
-                "checkout": {"worktree_dir": "/tmp/review-pr-570"},
+                "checkout": {
+                    "worktree_dir": "/tmp/review-pr-570",
+                    "head_ref_name": "issue-117-graph-partitioning",
+                },
                 "merge": {"status": "conflicted", "conflicts": ["src/models/graph_partitioning.rs"]},
             },
         }
@@ -197,6 +200,7 @@ def test_emit_result_prints_review_pipeline_text_report(self) -> None:
         self.assertIn("- CI state: failure", rendered)
         self.assertIn("## Merge Prep", rendered)
         self.assertIn("- Worktree: `/tmp/review-pr-570`", rendered)
+        self.assertIn("- PR head branch: `issue-117-graph-partitioning`", rendered)
         self.assertIn("## Linked Issue Context", rendered)
 
     def test_emit_result_prints_review_implementation_text_report(self) -> None: