loopx-project · huangruiteng · Sep 22, 2026 · Sep 21, 2026
@@ -710,7 +710,8 @@ def test_skill_slash_fallback_contract() -> None:
     assert "Do not handle `/loopx-pr-review` from this broader project skill" in normalized
     assert "do not route it to `loopx-pr-merge` unless" in normalized
     assert "loopx --format json pr-review --state all" not in skill_text
-    assert "loopx --format json pr-review --state all" in pr_review_skill_text
+    assert "keeps ordinary queue discovery open-only" in pr_review_skill_text
+    assert "explicit `--state merged|all`" in pr_review_skill_text
     assert "Save the full first JSON packet before printing a compact projection" in pr_review_normalized
     assert "agent_response_contract" in pr_review_skill_text
     assert "pull_requests[review_action_kind!=null].review_template" in pr_review_skill_text

@@ -418,7 +418,8 @@ def main() -> int:
         pr_review_skill = codex_home / "skills" / "loopx-pr-review" / "SKILL.md"
         pr_review_text = " ".join(pr_review_skill.read_text(encoding="utf-8").split())
         for phrase in (
-            "loopx --format json pr-review --state all",
+            "keeps ordinary queue discovery open-only",
+            "explicit `--state merged|all`",
             "thin host adapter",
             "agent_response_contract.review_execution_contract",
             "review_groups",

@@ -64,7 +64,8 @@ def main() -> int:
     skill_text = " ".join(skill_source.split())
     for phrase in (
         "This skill is a thin host adapter",
-        "loopx --format json pr-review --state all",
+        "keeps ordinary queue discovery open-only",
+        "explicit `--state merged|all`",
         "agent_response_contract.review_execution_contract",
         "pull_requests[review_action_kind!=null].review_plan",
         "pull_requests[review_action_kind!=null].review_template",
@@ -222,7 +223,8 @@ def fake_run_gh_json(args: list[str], *, cwd: Path | None = None) -> object:
 
     payload = json.loads(
         run_cli(
-            "--format", "json", "pr-review", "--fixture", str(FIXTURE), "--limit", "5"
+            "--format", "json", "pr-review", "--fixture", str(FIXTURE),
+            "--state", "all", "--limit", "5"
         ).stdout
     )
     assert payload["schema_version"] == "loopx_pr_review_command_response_v0", payload
@@ -254,7 +256,15 @@ def fake_run_gh_json(args: list[str], *, cwd: Path | None = None) -> object:
     assert payload["summary"]["total_pr_count"] == 4, payload["summary"]
     assert payload["summary"]["open_pr_count"] == 3, payload["summary"]
     assert payload["summary"]["merged_pr_count"] == 1, payload["summary"]
-    target = payload["pull_requests"][0]
+    default_open = json.loads(
+        run_cli(
+            "--format", "json", "pr-review", "--fixture", str(FIXTURE), "--limit", "5"
+        ).stdout
+    )
+    assert default_open["request"]["state_filter"] == "open", default_open["request"]
+    assert default_open["summary"]["total_pr_count"] == 3, default_open["summary"]
+    assert default_open["summary"]["merged_pr_count"] == 0, default_open["summary"]
+    target = next(item for item in payload["pull_requests"] if item["number"] == 770)
     exact_target = f"{target['number']}@{target['head_oid']}"
     targeted = json.loads(
         run_cli(
@@ -263,6 +273,7 @@ def fake_run_gh_json(args: list[str], *, cwd: Path | None = None) -> object:
         ).stdout
     )
     assert targeted["request"]["target_exact_heads"] == [exact_target], targeted
+    assert targeted["request"]["state_filter"] == "all", targeted["request"]
     assert targeted["result_completeness"]["complete"] is True, targeted
     assert targeted["result_completeness"]["limit_scope"] == "exact_targets", targeted
     assert [item["number"] for item in targeted["pull_requests"]] == [target["number"]]
@@ -1152,7 +1163,8 @@ def approved_open_head(
 
     group_limited = json.loads(
         run_cli(
-            "--format", "json", "pr-review", "--fixture", str(FIXTURE), "--limit", "1"
+            "--format", "json", "pr-review", "--fixture", str(FIXTURE),
+            "--state", "all", "--limit", "1"
         ).stdout
     )
     assert group_limited["summary"]["total_pr_count"] == 2, group_limited["summary"]
@@ -1192,6 +1204,8 @@ def approved_open_head(
             "pr-review",
             "--fixture",
             str(FIXTURE),
+            "--state",
+            "all",
             "--since",
             "2026-06-27T12:20:00Z",
             "--limit",
@@ -1206,7 +1220,14 @@ def approved_open_head(
         "review_sequence"
     ]
 
-    markdown = run_cli("pr-review", "--fixture", str(FIXTURE), "--limit", "1").stdout
+    default_markdown = run_cli(
+        "pr-review", "--fixture", str(FIXTURE), "--limit", "1"
+    ).stdout
+    assert "state_filter: `open`" in default_markdown, default_markdown
+    assert "#770" not in default_markdown, default_markdown
+    markdown = run_cli(
+        "pr-review", "--fixture", str(FIXTURE), "--state", "all", "--limit", "1"
+    ).stdout
     assert "# Project PR Review Queue" in markdown, markdown
     assert "current gh repository" not in markdown, markdown
     assert "state_filter: `all`" in markdown, markdown

@@ -59,9 +59,13 @@ Use the JSON form for the first pass so the response contract and per-PR blank
 templates enter the model context:
 
 ```bash
-loopx --format json pr-review --state all [--repo owner/repo] [--since ISO]
+loopx --format json pr-review [--repo owner/repo] [--since ISO]
 ```
 
+The ordinary queue defaults to open PRs. Use `--state merged` or `--state all`
+explicitly only for a lifecycle or post-merge audit; named exact targets remain
+lifecycle-neutral when `--state` is omitted.
+
 When the user explicitly names one or a few PRs, resolve each current head and
 request only those exact heads. This direct path is complete for the named
 targets and must not be expanded into a historical queue merely to satisfy
@@ -844,10 +848,10 @@ A first implementation is acceptable when:
 - `loopx pr-review` returns `loopx_pr_review_command_response_v0`;
 - default live reads use the caller's current `gh` repository, while
   `--repo owner/repo` can review another GitHub project;
-- `--state all` includes merged PRs in the same packet, applies `--limit` per
+- omitted `--state` and explicit `--state open` keep the ordinary queue
+  open-only; `--state all` includes merged PRs in the same packet, applies `--limit` per
   lifecycle group, and keeps `review_groups.merged` non-empty when merged PRs
-  exist in the requested window; `--state open` preserves the old open-only
-  review queue;
+  exist in the requested window;
 - `pull_requests` remains the full bounded inventory while every
   `review_sequence` contains only rows with a non-null `review_action_kind`;
   valid concluded exact heads are never recommended for duplicate work and

@@ -3,7 +3,7 @@
 import argparse
 import hashlib
 import json
-from collections.abc import Callable
+from collections.abc import Callable, Sequence
 from contextlib import suppress
 from pathlib import Path
 
@@ -97,7 +97,7 @@ def register_pr_review_command(
 ) -> None:
     parser = subparsers.add_parser(
         "pr-review",
-        help="Build a public-safe /loopx-pr-review queue for the current project's open and merged pull requests.",
+        help="Build a public-safe /loopx-pr-review queue for the current project's pull requests.",
     )
     add_subcommand_format(parser)
     parser.add_argument("--goal-id", help="Use this Goal PR review configuration; otherwise use machine defaults.")
@@ -130,8 +130,11 @@ def register_pr_review_command(
     parser.add_argument(
         "--state",
         choices=("open", "merged", "all"),
-        default="all",
-        help="PR lifecycle state to include. Defaults to all so merged PRs remain reviewable.",
+        default=None,
+        help=(
+            "PR lifecycle state to include. Ordinary queues default to open; "
+            "use merged/all explicitly for lifecycle or post-merge audits."
+        ),
     )
     parser.add_argument(
         "--review-priority",
@@ -209,6 +212,16 @@ def register_pr_review_command(
     )
 
 
+def _resolve_pr_review_state_filter(
+    raw_state: object,
+    *,
+    target_exact_heads: Sequence[str],
+) -> str:
+    if raw_state is None and target_exact_heads:
+        return "all"
+    return normalize_pr_state_filter(raw_state)
+
+
 def handle_pr_review_command(
     args: argparse.Namespace,
     *,
@@ -222,6 +235,10 @@ def handle_pr_review_command(
     checkpoint_path: Path | None = None
     resolved_review_priority = DEFAULT_REVIEW_PRIORITY
     target_exact_heads = list(getattr(args, "target_exact_head", []) or [])
+    resolved_state_filter = _resolve_pr_review_state_filter(
+        getattr(args, "state", None),
+        target_exact_heads=target_exact_heads,
+    )
     try:
         machine_configuration = (read_machine_configuration(runtime_root, registry=build_builtin_machine_configuration_registry()) if runtime_root is not None else None)
         goal = None
@@ -433,7 +450,7 @@ def handle_pr_review_command(
                 source_scan = scan_github_pull_requests(
                     repo=repository,
                     limit=max(1, args.limit) + 1,
-                    state_filter=normalize_pr_state_filter(args.state),
+                    state_filter=resolved_state_filter,
                     since=args.since,
                     **({"wait_for_ci": False} if not wait_for_ci else {}),
                 )
@@ -454,7 +471,7 @@ def handle_pr_review_command(
             repository=repository,
             limit=max(1, args.limit),
             source=source,
-            state_filter=normalize_pr_state_filter(args.state),
+            state_filter=resolved_state_filter,
             since=args.since,
             source_scan=source_scan,
             reviewer_login=reviewer_login,
@@ -517,7 +534,7 @@ def handle_pr_review_command(
                 "cli_command": "loopx pr-review [--repo owner/repo] [--target-exact-head NUMBER@HEAD_OID] [--state open|merged|all] [--review-priority other-developers-first|owner-first] [--since ISO]",
                 "repository": args.repo,
                 "limit": max(1, args.limit),
-                "state_filter": normalize_pr_state_filter(args.state),
+                "state_filter": resolved_state_filter,
                 "since": args.since,
                 "review_priority": resolved_review_priority.value,
                 "fresh_audit_exact_heads": list(args.fresh_audit_exact_head),

@@ -43,7 +43,8 @@
         "--delivery-outcome <ACTUAL_DELIVERY_OUTCOME>",
     ),
     "loopx-pr-review": (
-        "loopx --format json pr-review --state all",
+        "keeps ordinary queue discovery open-only",
+        "explicit `--state merged|all`",
         "thin host adapter",
         "agent_response_contract.review_execution_contract",
         "pull_requests[review_action_kind!=null].review_plan",

@@ -193,16 +193,16 @@ def _include_pr_in_window(pr: dict[str, Any], *, since: object | None) -> bool:
 
 
 def normalize_pr_state_filter(value: object) -> str:
-    state = str(value or "all").strip().lower()
-    return state if state in {"open", "merged", "all"} else "all"
+    state = str(value or "open").strip().lower()
+    return state if state in {"open", "merged", "all"} else "open"
 
 
 def fetch_github_pull_requests(
     *,
     repo: str | None,
     limit: int,
     cwd: Path | None = None,
-    state_filter: str = "all",
+    state_filter: str = "open",
     since: str | None = None,
 ) -> list[dict[str, Any]]:
     scan = scan_github_pull_requests(
@@ -220,7 +220,7 @@ def scan_github_pull_requests(
     repo: str | None,
     limit: int,
     cwd: Path | None = None,
-    state_filter: str = "all",
+    state_filter: str = "open",
     since: str | None = None,
     wait_for_ci: bool = True,
 ) -> dict[str, Any]:
@@ -1022,7 +1022,7 @@ def build_pr_review_packet(
     repository: str | None,
     limit: int,
     source: str,
-    state_filter: str = "all",
+    state_filter: str = "open",
     since: str | None = None,
     source_scan: Mapping[str, Any] | None = None,
     reviewer_login: str | None = None,
@@ -1347,7 +1347,7 @@ def render_pr_review_markdown(payload: dict[str, Any]) -> str:
         "",
         f"- command: `{request.get('command')}`",
         f"- repository: `{request.get('repository') or 'current gh repository'}`",
-        f"- state_filter: `{request.get('state_filter') or 'all'}`",
+        f"- state_filter: `{request.get('state_filter') or 'open'}`",
         f"- since: `{request.get('since') or 'not set'}`",
         f"- headline: {summary.get('headline')}",
         f"- complete: `{completeness.get('complete')}`; truncated=`{completeness.get('truncated')}`; recommended_limit=`{completeness.get('recommended_limit')}`",

@@ -17,7 +17,7 @@ state or time window. Route approval, merge, self-merge, and admin bypass to
 `loopx-pr-merge` (optional repo-kept workflow, not installed by default) after
 the evidence review is complete; it never replaces this skill's exact-head gate.
 
-For named PRs, resolve heads and run repeatable `--target-exact-head NUMBER@HEAD_OID`; run `loopx --format json pr-review --state all` only for queue intent, never to expand explicit targets into historical inventory.
+For named PRs, resolve heads and run repeatable `--target-exact-head NUMBER@HEAD_OID`. An omitted `--state` keeps ordinary queue discovery open-only while exact targets remain lifecycle-neutral; use explicit `--state merged|all` only for deliberate history or post-merge audit.
 
 Translate only explicit filters:
 

@@ -152,6 +152,7 @@ teaches a reusable control-plane lesson.
 | `explore_visual_cross_role_postcondition_gap` | Each role reports a successful marker readback, but a later role overwrites the same whiteboard and makes the earlier success stale. | configured role/stage token fingerprints, publish order, batch readback target set, final role markers. | Stage tokens were validated only within each role and marker settlement ran before all role writes completed. | Reject cross-role stage-token reuse before writes, then settle every role/stage marker once after the final batch write and recompute all role and delivery statuses from that final postcondition. |
 | `nested_singleflight_self_contention` | A batch owns a sink singleflight lock, then a nested direct command or automatic sync for the same config returns `sync_busy` even though no independent writer exists. | outer batch lock scope, nested command call path, process/context identity, connector call count, cross-process contention smoke. | The file lock distinguished only acquired versus busy and could not recognize re-entry by its current execution context; a regression fixture may also have encoded self-rejection as expected behavior. | Track held lock targets in execution-local context, reuse only an exact same-target ownership, preserve cross-context/process fail-fast, and cover nested direct/decorated calls plus a competing process. |
 | `pr_review_lifecycle_group_truncation` | `loopx pr-review --state all` reports open PRs but `review_groups.merged` is empty even though `--state merged` or GitHub closed/merged search finds merged PRs in the same window. | `loopx --format json pr-review --state all --since ...`, `loopx --format json pr-review --state merged --since ...`, `gh pr list --state closed/merged`, and `examples/pr-review-command-smoke.py`. | The live fetch or packet builder applied one combined `--limit` before lifecycle grouping, so open PRs consumed the packet and merged post-merge review entries were truncated away. | Fetch open and closed/merged windows separately for `--state all`, apply `--limit` per lifecycle group, keep `review_groups.unmerged` and `review_groups.merged` authoritative, and cover a fixture where `--limit 1` still returns one merged PR. |
+| `pr_review_default_lifecycle_overreach` | An ordinary `loopx pr-review` queue scans historical merged PRs, saturates a closed-PR window, or diverts reviewers into post-merge audits even though the user asked to continue current reviews. | CLI parser defaults, internal scan and packet defaults, installed skill first-pass command, ordinary no-state packet, explicit all-state packet, and a merged exact-target packet. | Queue discovery and deliberate lifecycle/post-merge audit shared an `all` default, and the host adapter reinforced it with an explicit `--state all` command. | Default ordinary discovery and internal APIs to `open`; require explicit `--state merged|all` for history; keep an omitted-state exact-target request lifecycle-neutral so a named merged head remains reviewable; cover all three routes in the command smoke. |
 | `pr_review_response_duplicate_replan` | A PR still reports `CHANGES_REQUESTED` after the author pushed a newer repair and resolved every review thread, so lifecycle polling repeatedly proposes another patch successor while the actual next step is reviewer re-approval. | Compact PR lifecycle metadata, complete review-thread counts, latest changes-requested timestamp, head commit timestamp, CI rollup, and current monitor todo. | Lifecycle routing treated the aggregate review decision as a current unhandled action and ignored the response state encoded by newer commits plus resolved threads. | On explicit metadata fetch, read only compact review-response evidence and route to a quiet re-review monitor when at least one thread exists, every fetched thread is resolved, pagination is complete, and the head commit is newer than the review. Missing, partial, empty-thread, or older evidence must fail closed to the actionable replan route. |
 | `pr_review_explicit_selection_idempotency_gap` | A named PR receives another full evidence pass even though the packet marks its exact-head conclusion valid or its merged row has `review_action_kind=null`; generic `re-review` wording repeatedly reopens completed work, a null-action row remains in the ranked `review_sequence`, or its inventory row still carries an executable-looking review plan/template/commands that a host follows. | Selected packet row, exact head, `review_conclusion`, `review_action_kind`, plan/template/command presence, top-level and group `review_sequence`, summary attention counts, current-request wording, and the evidence-command tool trace. | The host adapter conflated explicit selection authority over queue ordering with authority to override the capability's exact-head execution decision, while the packet mixed bounded inventory with executable ranking and artifacts and treated generic re-review wording as a force-refresh token. | Project a typed selection-execution contract: explicit selection changes ordering only; null action is compact readback-only with null plan/template and empty evidence commands; every `review_sequence` and attention count derives exclusively from non-null actions; and a fresh audit on an unchanged/no-action row must be regenerated through the typed `--fresh-audit-exact-head NUMBER@HEAD_OID` option after an explicit request or concrete new concern/evidence invalidation. Keep valid concluded rows only in inventory; assign an explicit audit action to merged heads that genuinely lack a valid conclusion. Cover the packet contract, installed skill text, exact valid-merged fixture, and result-check rejection of inventory-only rows with focused smokes. |
 | `pr_review_key_code_explanation_gap` | A PR review names changed files and symbols, summarizes intent, and lists passing checks, but a reader still cannot reconstruct the critical branch, state transition, side effect, consumer, or failure path from the review. | Exact reviewed head, changed production symbols, surrounding definitions and active call sites, published five-block review, and packet explanation-depth contract. | The review contract asked for mechanism-rich prose but did not make exact-head key-code explanation a required subsection, so agents could satisfy the headings with a polished file inventory. | Require `关键代码讲解` under `具体改动` for code-changing PRs; select 2-5 behavior-bearing symbols, cite exact-head lines, use short excerpts or equivalent pseudocode, and explain inputs/pre-state, branch/invariant, calls/side effects, outputs/consumers, and failure ownership. Keep docs-only reviews on a parallel key-content path and cover both the skill text and CLI packet in the PR-review smoke. |