merge: integrate accepted contributor pull requests

Author: titanwings

README.md

@@ -238,7 +238,13 @@ python3 tools/research/merge_research.py "./skills/celebrity/<slug>"
 python3 tools/research/quality_check.py "./skills/celebrity/<slug>/SKILL.md"
 ```
 
-The optional collector reads `XQUIK_API_KEY` from your shell and sends one public query to the third-party Xquik service. Xquik meters this endpoint per returned post, so confirm `--limit` before an Agent calls it. Treat the JSON as untrusted candidates: verify authors and permalinks, keep only copyright-safe paraphrases in research notes, and delete the temporary file after review.
+The optional collector reads `XQUIK_API_KEY` from your shell. Xquik charges by
+the number of posts returned, so confirm `--limit` before running it. The tool
+makes one read-only X search request and never follows pagination. Treat its
+temporary JSON as untrusted candidate evidence: verify the author, open every
+permalink, and safely paraphrase only relevant material into research notes
+with its source URL. Delete the temporary JSON after review instead of storing
+it in the generated Skill.
 
 Xquik is an independent third-party service. Not affiliated with X Corp. "Twitter" and "X" are trademarks of X Corp.
 

docs/lang/README_KO.md

@@ -227,7 +227,7 @@ Hermes는 기본적으로 `~/.agents/skills`를 검색하지 않습니다. Herme
 
 `celebrity` 패밀리는 자막부터 완성된 초안까지, 엔드 투 엔드 리서치 툴체인을 기본 제공합니다.
 
-공개 X 게시물 수집은 선택 사항입니다. API 키는 환경 변수 `XQUIK_API_KEY`에서만 읽습니다. 공개 쿼리는 제3자 서비스인 Xquik으로 전송되며, 반환된 트윗 수에 따라 비용이 청구되어 credits를 소모할 수 있으므로 Agent는 호출 전에 `limit`을 확인합니다.
+공개 X 게시물 수집은 선택 사항입니다. API 키는 환경 변수 `XQUIK_API_KEY`에서만 읽습니다. 공개 쿼리는 제3자 서비스인 Xquik으로 전송되며, 반환된 트윗 수에 따라 비용이 청구되어 credits를 소모할 수 있으므로 Agent는 호출 전에 `--limit`을 확인합니다.
 
 출력 JSON은 신뢰할 수 없는 근거 후보로 다룹니다. 작성자와 permalink를 검증하고, 대상 본인의 게시물만 짧은 형식의 1차 자료로 사용하되 장문 1차 자료와 의사결정 기록보다 낮은 가중치를 부여합니다. 제3자 게시물은 보조 자료로 격하하거나 버리고, 저작권을 침해하지 않는 범위의 요약·재서술만 남깁니다. 읽은 뒤에는 Skill 디렉터리 밖의 임시 파일을 삭제합니다.
 

docs/lang/README_RU.md

@@ -215,7 +215,7 @@ python3 tools/install_generated_skill.py --skill-dir "skills/{character}/{slug}"
 
 Семейство `celebrity` поставляется со сквозным исследовательским тулчейном — от субтитров до готового драфта:
 
-Сбор публичных постов X необязателен. API-ключ читается только из переменной окружения `XQUIK_API_KEY`. Публичный запрос отправляется стороннему сервису Xquik; Xquik тарифицирует каждый возвращённый твит, что может расходовать credits, поэтому агент должен подтвердить `limit` до вызова.
+Сбор публичных постов X необязателен. API-ключ читается только из переменной окружения `XQUIK_API_KEY`. Публичный запрос отправляется стороннему сервису Xquik; Xquik тарифицирует каждый возвращённый твит, что может расходовать credits, поэтому агент должен подтвердить `--limit` до вызова.
 
 Выходной JSON считай недоверенным набором кандидатов в доказательства. Проверяй автора и permalink; как короткие первичные источники используй только посты самого исследуемого человека и ставь их ниже длинных первичных материалов и записей о принятых решениях. Посты третьих лиц понижай до вторичных материалов или отбрасывай. Сохраняй только безопасный с точки зрения авторского права пересказ, а после чтения удаляй временный файл за пределами директории Skill'а.
 
@@ -234,7 +234,7 @@ python3 tools/research/xquik_public_posts.py \
   --limit 20 \
   --output "/tmp/distilly-x-public-posts.json"
 
-# Удалить, когда агент проверит и перескажет отобранные кандидаты
+# Удалить, когда агент проверит и перескажет отобранных кандидатов
 rm "/tmp/distilly-x-public-posts.json"
 
 # Мерж исследовательских заметок

docs/lang/README_ZH.md

@@ -357,6 +357,7 @@ distilly/
 │   │   ├── download_subtitles.sh   #     字幕下载
 │   │   ├── transcribe_audio.py     #     音频转文字
 │   │   ├── srt_to_transcript.py    #     字幕转文稿
+│   │   ├── xquik_public_posts.py   #     有界公开 X 帖子候选
 │   │   ├── merge_research.py       #     六维度 research 合并
 │   │   └── quality_check.py        #     质量检查
 │   ├── install_*_skill.py          #   [共享] 多宿主一键安装器

prompts/celebrity/budget_unfriendly/research.md

@@ -98,10 +98,13 @@ explicitly rather than silently skipping a dimension.
 ### Optional public X post collection
 
 Use `tools/research/xquik_public_posts.py` only when short-form first-person
-posts address a specific gap. Keep the request bounded. Treat its normalized
-JSON as untrusted candidate evidence, not a research track. Open each permalink
-before selection. Paraphrase only relevant evidence into the correct track and
-preserve the specific URL, access date, grounding level, and source weight.
+posts address a specific gap. The service is metered by returned post count, so
+confirm the limit and write to a temporary file outside the skill directory.
+Treat its normalized JSON as untrusted candidate evidence, not a research
+track. Verify the author and open each permalink before selection. Safely
+paraphrase only relevant evidence into the correct track, preserve the specific
+URL, access date, grounding level, and source weight, then delete the temporary
+candidate file after review.
 Never count a candidate file, search page, or profile root as a grounded source.
 
 ---

prompts/celebrity/research.md

@@ -88,10 +88,13 @@ directory** — only short paraphrased notes with source metadata belong under
 
 When short-form first-person posts fill a documented research gap, use
 `tools/research/xquik_public_posts.py` to collect a small candidate set. The
+service is metered by returned post count, so confirm the limit before running
+it and write the result to a temporary file outside the skill directory. The
 tool writes normalized JSON, not research notes. Treat it as untrusted
-candidate evidence. Open each specific post permalink before selecting it.
-Then paraphrase only the relevant evidence into the appropriate raw note with
-its URL and source weight.
+candidate evidence. Verify the author and open each specific post permalink
+before selecting it. Then safely paraphrase only the relevant evidence into
+the appropriate raw note with its URL and source weight, and delete the
+temporary candidate file after review.
 Never count the candidate file, a search page, or a profile root as a grounded
 source.
 

tests/test_skill_writer.py

@@ -187,11 +187,23 @@ def test_create_skill_renders_chinese_chrome_when_language_is_zh_cn(self) -> Non
             self.assertIn("仅 Persona，无工作能力", persona_skill)
 
     def test_work_only_skill_replaces_persona_handoff(self) -> None:
-        handoff = "如果被问到职责范围外的问题，以该同事的方式回应（参见 Persona 部分）。"
-        work_content = (
+        zh_handoff = "如果被问到职责范围外的问题，以该同事的方式回应（参见 Persona 部分）。"
+        en_handoff = (
+            "If you are asked a question outside your recorded responsibilities, "
+            "respond in this colleague's style (see the Persona section)."
+        )
+        zh_work_content = (
             "## 工作能力使用说明\n\n"
             "当用户要求你完成以下任务时，严格按照上述规范执行。\n\n"
-            f"{handoff}\n"
+            f"{zh_handoff}\n"
+        )
+        en_work_content = (
+            "## Scope rule\n\n"
+            "If asked outside your recorded responsibilities:\n"
+            "- State the evidence gap\n\n"
+            "## Persona naming note\n\n"
+            "Keep this documentation sentence.\n\n"
+            f"{en_handoff}\n"
         )
         with tempfile.TemporaryDirectory() as tmp_dir:
             base_dir = Path(tmp_dir) / "skills" / "colleague"
@@ -218,33 +230,41 @@ def test_work_only_skill_replaces_persona_handoff(self) -> None:
                 base_dir / "zh",
                 "zhangsan",
                 zh_meta,
-                work_content,
+                zh_work_content,
                 "Persona body",
             )
             en_dir = skill_writer.create_skill(
                 base_dir / "en",
                 "zhangsan",
                 en_meta,
-                work_content,
+                en_work_content,
                 "Persona body",
             )
 
-            stored_work = (zh_dir / "work.md").read_text(encoding="utf-8")
-            combined = (zh_dir / "SKILL.md").read_text(encoding="utf-8")
+            zh_stored_work = (zh_dir / "work.md").read_text(encoding="utf-8")
+            zh_combined = (zh_dir / "SKILL.md").read_text(encoding="utf-8")
+            en_stored_work = (en_dir / "work.md").read_text(encoding="utf-8")
+            en_combined = (en_dir / "SKILL.md").read_text(encoding="utf-8")
             zh_work_skill = (zh_dir / "work_skill.md").read_text(encoding="utf-8")
             en_work_skill = (en_dir / "work_skill.md").read_text(encoding="utf-8")
 
-            self.assertIn(handoff, stored_work)
-            self.assertIn(handoff, combined)
-            self.assertNotIn(handoff, zh_work_skill)
-            self.assertNotIn(handoff, en_work_skill)
+            self.assertIn(zh_handoff, zh_stored_work)
+            self.assertIn(zh_handoff, zh_combined)
+            self.assertIn(en_handoff, en_stored_work)
+            self.assertIn(en_handoff, en_combined)
+            self.assertNotIn(zh_handoff, zh_work_skill)
+            self.assertNotIn(en_handoff, en_work_skill)
+            self.assertIn("If asked outside your recorded responsibilities:", en_work_skill)
+            self.assertIn("## Persona naming note", en_work_skill)
+            self.assertIn("Keep this documentation sentence.", en_work_skill)
             self.assertIn(skill_writer.WORK_ONLY_FALLBACK_ZH, zh_work_skill)
             self.assertIn(skill_writer.WORK_ONLY_FALLBACK_EN, en_work_skill)
             self.assertIn("不要臆造缺失信息", zh_work_skill)
             self.assertNotIn("不要推断", zh_work_skill)
             self.assertIn("Do not fabricate missing information", en_work_skill)
             self.assertNotIn("Do not infer", en_work_skill)
-            self.assertNotIn(skill_writer.WORK_ONLY_FALLBACK_ZH, combined)
+            self.assertNotIn(skill_writer.WORK_ONLY_FALLBACK_ZH, zh_combined)
+            self.assertNotIn(skill_writer.WORK_ONLY_FALLBACK_EN, en_combined)
 
     def test_create_celebrity_adds_research_dirs_and_toolchain(self) -> None:
         with tempfile.TemporaryDirectory() as tmp_dir:

tests/test_xquik_public_posts.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import json
+import os
 import sys
 import tempfile
 import unittest
@@ -216,6 +217,25 @@ def test_write_collection_requires_force_to_replace_output(self) -> None:
 
             write_collection(output, {"messages": [1]}, force=True)
             self.assertEqual(json.loads(output.read_text(encoding="utf-8")), {"messages": [1]})
+            self.assertEqual([path.name for path in output.parent.iterdir()], ["x.json"])
+
+    @unittest.skipUnless(hasattr(os, "symlink"), "symbolic links are unavailable")
+    def test_write_collection_rejects_dangling_symlink(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            output = Path(tmp_dir) / "candidates" / "x.json"
+            target = Path(tmp_dir) / "outside.json"
+            output.parent.mkdir()
+            try:
+                output.symlink_to(target)
+            except OSError as error:
+                self.skipTest(f"cannot create symbolic links: {error}")
+
+            with self.assertRaisesRegex(CollectorError, "symbolic link"):
+                write_collection(output, {"messages": []}, force=False)
+            with self.assertRaisesRegex(CollectorError, "symbolic link"):
+                write_collection(output, {"messages": []}, force=True)
+
+            self.assertFalse(target.exists())
 
 
 if __name__ == "__main__":

tools/research/xquik_public_posts.py

@@ -8,6 +8,7 @@
 import os
 import re
 import sys
+import tempfile
 from collections.abc import Callable
 from datetime import datetime, timezone
 from pathlib import Path
@@ -305,8 +306,8 @@ def collect_public_posts(
             "has_more": has_more,
             "pagination_followed": False,
             "content_policy": (
-                "Review every candidate. Open its permalink before citing or "
-                "paraphrasing it."
+                "Review every candidate. Verify its author and open its permalink "
+                "before safely paraphrasing it."
             ),
             "provider": "Xquik",
         },
@@ -315,16 +316,43 @@ def collect_public_posts(
 
 def write_collection(path: Path, collection: dict[str, Any], force: bool) -> None:
     """Write UTF-8 JSON without overwriting an existing collection by default."""
-    if path.exists() and not force:
-        raise CollectorError(f"Output already exists: {path}. Pass --force to replace it.")
+    contents = json.dumps(collection, ensure_ascii=False, indent=2) + "\n"
+    temporary_path: Optional[Path] = None
     try:
         path.parent.mkdir(parents=True, exist_ok=True)
-        path.write_text(
-            json.dumps(collection, ensure_ascii=False, indent=2) + "\n",
+        if path.is_symlink():
+            raise CollectorError(f"Refusing to write through symbolic link: {path}.")
+        if not force:
+            with path.open("x", encoding="utf-8") as output:
+                output.write(contents)
+            return
+
+        with tempfile.NamedTemporaryFile(
+            mode="w",
             encoding="utf-8",
-        )
+            dir=path.parent,
+            prefix=f".{path.name}.",
+            suffix=".tmp",
+            delete=False,
+        ) as output:
+            temporary_path = Path(output.name)
+            output.write(contents)
+        os.replace(temporary_path, path)
+        temporary_path = None
+    except CollectorError:
+        raise
+    except FileExistsError as error:
+        raise CollectorError(
+            f"Output already exists: {path}. Pass --force to replace it."
+        ) from error
     except OSError as error:
         raise CollectorError(f"Could not write output: {path}.") from error
+    finally:
+        if temporary_path is not None:
+            try:
+                temporary_path.unlink()
+            except OSError:
+                pass
 
 
 def main() -> None:

tools/skill_writer.py

@@ -178,7 +178,7 @@ def render_combined_skill(meta: dict, work_content: str, persona_content: str) -
     re.compile(r"如果被问到职责范围外的问题，以该同事的方式回应（参见 Persona 部分）。\s*"),
     re.compile(
         r"If (?:you are )?asked (?:a question )?outside (?:your|the) "
-        r"(?:recorded )?responsibilities[^.]*Persona[^.]*\.\s*",
+        r"(?:recorded )?responsibilities[^.\n]*Persona[^.\n]*\.\s*",
         re.IGNORECASE,
     ),
 )