[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C2HJS8MARX96829QBTF0WK":3,"skill-same:01M7C2HJS8MARX96829QBTF0WK":45},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7C2HJS8MARX96829QBTF0WK","academic-pdf-redaction",null,"Redact text from PDF documents for blind review anonymization","linyouxue\u002Fskill-repair-benchmark","linyouxue","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F216151464?v=4","01M7BJEVCXKD74KH6FG9CBVSQB","林宥学","01M7BJEVE02NP217V93BKZFC0J",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002FWML-Opus47-15Tasks-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Facademic-pdf-redaction\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: academic-pdf-redaction\ndescription: Redact text from PDF documents for blind review anonymization\n---\n\n# PDF Redaction for Blind Review\n\nRedact identifying information from academic papers for blind review.\n\n## CRITICAL RULES\n\n1. **PRESERVE References section** - Self-citations MUST remain intact\n2. **ONLY redact specific text matches** - Never redact entire pages\u002Fregions\n3. **VERIFY output** - Check that 80%+ of original text remains\n\n## Common Pitfalls to AVOID\n\n```python\n# ❌ WRONG - This removes ALL text from the page:\nfor block in page.get_text(\"blocks\"):\n    page.add_redact_annot(fitz.Rect(block[:4]))\n\n# ❌ WRONG - Drawing rectangles over text:\npage.draw_rect(fitz.Rect(0, 0, 600, 100), fill=(0,0,0))\n\n# ✅ CORRECT - Only redact specific search matches:\nfor rect in page.search_for(\"John Smith\"):\n    page.add_redact_annot(rect)\n```\n\n## Patterns to Redact (Before References Only)\n\n**IMPORTANT: Use FULL names\u002Fphrases, not partial matches!**\n- ✅ \"John Smith\" (full name)\n- ❌ \"Smith\" (partial - would incorrectly match \"Smith et al.\" citations in References)\n\n1. **Author names** - FULL names only (e.g., \"John Smith\", not just \"Smith\")\n2. **Affiliations** - Universities, companies (e.g., \"Duke University\")\n3. **Email addresses** - Pattern: `*@*.edu`, `*@*.com`\n4. **Venue names** - Conference\u002Fworkshop names (e.g., \"ICML 2024\", \"ICML Workshop\")\n5. **arXiv identifiers** - Pattern: `arXiv:XXXX.XXXXX`\n6. **DOIs** - Pattern: `10.XXXX\u002F...`\n7. **Acknowledgement names** - Names in \"Acknowledgements\" section (also grant numbers \u002F funder codes that appear alongside them and can identify authors, e.g., \"W911NF-...\", \"NSF 1234567\")\n8. **Equal contribution footnotes** - canonical fixed-string variants that MUST be probed for every paper, regardless of whether they were noticed on first scan:\n   - `Equal contribution`\n   - `* Equal contribution`\n   - `*Equal contribution`\n   - `Equal Contribution`\n   - `Equal contributions`\n   - `†Equal contribution` \u002F `‡Equal contribution` (and other footnote-marker variants)\n\n### MANDATORY: per-document category coverage checklist\n\nBefore invoking the redactor for a paper, walk the 8 categories above in order and, for each one, EITHER:\n\n- add at least one concrete search string to that paper's `patterns` list, OR\n- record a short justification that the category is genuinely not present in the pre-References pages of this document (e.g., \"no DOI printed on first page\").\n\nCategories that are usually fixed phrases rather than document-specific tokens (notably #8 Equal contribution footnotes, and generic markers like `arXiv:`, `@`, `doi.org\u002F`) SHOULD be added by default to every paper's `patterns` list, because they are cheap to search and easy to overlook during ad-hoc reading of page 1. A category must never be silently skipped just because no instance was noticed while browsing.\n\n## PyMuPDF (fitz) - Recommended Approach\n\n```python\nimport fitz\nimport os\n\ndef redact_with_pymupdf(input_path: str, output_path: str, patterns: list[str]):\n    \"\"\"Redact specific patterns from PDF using PyMuPDF.\"\"\"\n    doc = fitz.open(input_path)\n    original_len = sum(len(p.get_text()) for p in doc)\n\n    # Find References page - stop redacting there\n    references_page = None\n    for i, page in enumerate(doc):\n        if \"references\" in page.get_text().lower():\n            references_page = i\n            break\n\n    for page_num, page in enumerate(doc):\n        if references_page is not None and page_num >= references_page:\n            continue  # Skip References section\n\n        for pattern in patterns:\n            # ONLY redact exact search matches\n            for rect in page.search_for(pattern):\n                page.add_redact_annot(rect, fill=(0, 0, 0))\n        page.apply_redactions()\n\n    os.makedirs(os.path.dirname(output_path), exist_ok=True)\n    doc.save(output_path)\n    doc.close()\n\n    # MUST verify after saving\n    verify_redaction(input_path, output_path)\n```\n\n## REQUIRED: Verification Function\n\n**Always run this after ANY redaction to catch errors early.**\n\nVerification MUST do two independent things:\n\n1. Structural \u002F retention checks (page count, character retention thresholds).\n2. **Category-taxonomy probes** derived directly from the 8 categories in \"Patterns to Redact\", executed on the pre-References pages of the redacted PDF. These probes are intentionally independent of whatever `patterns` list the planner happened to build for this paper, so that a category dropped during planning surfaces here as a leak instead of a false \"clean\" pass.\n\nOnly when both the structural checks pass AND every category probe returns no hit on the pre-References pages may the run be declared clean. If any probe fires, treat it as a real leak: extend that paper's `patterns` list (or record a justified exemption for a genuine false-positive, e.g., a body-text mention that is not authorship-revealing) and re-run redaction before finishing.\n\n```python\nimport fitz\nimport re\n\n# One lightweight probe per skill-listed category. These probes should be\n# broad enough to catch a category that was entirely forgotten during planning,\n# but they are NOT the redaction patterns themselves.\nCATEGORY_PROBES = {\n    \"emails (#3)\":                 re.compile(r\"[\\w.+-]+@[\\w-]+\\.[\\w.-]+\"),\n    \"arxiv ids (#5)\":              re.compile(r\"arXiv:\\s*\\d{4}\\.\\d{4,5}\", re.IGNORECASE),\n    \"dois (#6)\":                   re.compile(r\"\\b10\\.\\d{4,9}\u002F\\S+\"),\n    \"equal-contribution (#8)\":     re.compile(r\"equal\\s+contribution\", re.IGNORECASE),\n}\n\ndef _pre_refs_text(doc):\n    refs = None\n    for i, page in enumerate(doc):\n        if \"references\" in page.get_text().lower():\n            refs = i\n            break\n    end = refs if refs is not None else len(doc)\n    return \"\".join(doc[i].get_text() for i in range(end))\n\ndef verify_redaction(original_path, output_path):\n    \"\"\"Verify redaction didn't corrupt the PDF and that no skill-listed\n    category was silently skipped.\"\"\"\n    orig = fitz.open(original_path)\n    redc = fitz.open(output_path)\n\n    orig_len = sum(len(p.get_text()) for p in orig)\n    redc_len = sum(len(p.get_text()) for p in redc)\n\n    print(f\"Original: {len(orig)} pages, {orig_len} chars\")\n    print(f\"Redacted: {len(redc)} pages, {redc_len} chars\")\n    print(f\"Retained: {redc_len\u002Forig_len:.1%}\")\n\n    # DEFENSIVE CHECKS - fail fast if something went wrong\n    if len(redc) != len(orig):\n        raise ValueError(f\"Page count changed: {len(orig)} -> {len(redc)}\")\n    if redc_len \u003C 1000:\n        raise ValueError(f\"PDF corrupted: only {redc_len} chars remain!\")\n    if redc_len \u003C orig_len * 0.7:\n        raise ValueError(f\"Too much removed: kept only {redc_len\u002Forig_len:.0%}\")\n\n    # CATEGORY PROBES - one per skill-listed leakage category.\n    # These run on the redacted, pre-References text only.\n    pre_refs = _pre_refs_text(redc)\n    leaks = {}\n    for label, probe in CATEGORY_PROBES.items():\n        hits = probe.findall(pre_refs)\n        if hits:\n            leaks[label] = hits[:5]\n    if leaks:\n        raise ValueError(\n            \"Category-probe leaks in pre-References pages; extend patterns \"\n            f\"and re-redact: {leaks}\"\n        )\n\n    orig.close()\n    redc.close()\n    print(\"✓ Verification passed\")\n```\n\nNote: the probes above cover the categories that have canonical fixed-string \u002F regex forms (#3, #5, #6, #8). Categories #1, #2, #4, #7 are inherently document-specific (proper nouns), so their coverage is enforced by the per-document checklist under \"Patterns to Redact\" rather than by a generic probe here.\n","# PDF Redaction for Blind Review\n\nRedact identifying information from academic papers for blind review.\n\n## CRITICAL RULES\n\n1. **PRESERVE References section** - Self-citations MUST remain intact\n2. **ONLY redact specific text matches** - Never redact entire pages\u002Fregions\n3. **VERIFY output** - Check that 80%+ of original text remains\n\n## Common Pitfalls to AVOID\n\n```python\n# ❌ WRONG - This removes ALL text from the page:\nfor block in page.get_text(\"blocks\"):\n    page.add_redact_annot(fitz.Rect(block[:4]))\n\n# ❌ WRONG - Drawing rectangles over text:\npage.draw_rect(fitz.Rect(0, 0, 600, 100), fill=(0,0,0))\n\n# ✅ CORRECT - Only redact specific search matches:\nfor rect in page.search_for(\"John Smith\"):\n    page.add_redact_annot(rect)\n```\n\n## Patterns to Redact (Before References Only)\n\n**IMPORTANT: Use FULL names\u002Fphrases, not partial matches!**\n- ✅ \"John Smith\" (full name)\n- ❌ \"Smith\" (partial - would incorrectly match \"Smith et al.\" citations in References)\n\n1. **Author names** - FULL names only (e.g., \"John Smith\", not just \"Smith\")\n2. **Affiliations** - Universities, companies (e.g., \"Duke University\")\n3. **Email addresses** - Pattern: `*@*.edu`, `*@*.com`\n4. **Venue names** - Conference\u002Fworkshop names (e.g., \"ICML 2024\", \"ICML Workshop\")\n5. **arXiv identifiers** - Pattern: `arXiv:XXXX.XXXXX`\n6. **DOIs** - Pattern: `10.XXXX\u002F...`\n7. **Acknowledgement names** - Names in \"Acknowledgements\" section (also grant numbers \u002F funder codes that appear alongside them and can identify authors, e.g., \"W911NF-...\", \"NSF 1234567\")\n8. **Equal contribution footnotes** - canonical fixed-string variants that MUST be probed for every paper, regardless of whether they were noticed on first scan:\n   - `Equal contribution`\n   - `* Equal contribution`\n   - `*Equal contribution`\n   - `Equal Contribution`\n   - `Equal contributions`\n   - `†Equal contribution` \u002F `‡Equal contribution` (and other footnote-marker variants)\n\n### MANDATORY: per-document category coverage checklist\n\nBefore invoking the redactor for a paper, walk the 8 categories above in order and, for each one, EITHER:\n\n- add at least one concrete search string to that paper's `patterns` list, OR\n- record a short justification that the category is genuinely not present in the pre-References pages of this document (e.g., \"no DOI printed on first page\").\n\nCategories that are usually fixed phrases rather than document-specific tokens (notably #8 Equal contribution footnotes, and generic markers like `arXiv:`, `@`, `doi.org\u002F`) SHOULD be added by default to every paper's `patterns` list, because they are cheap to search and easy to overlook during ad-hoc reading of page 1. A category must never be silently skipped just because no instance was noticed while browsing.\n\n## PyMuPDF (fitz) - Recommended Approach\n\n```python\nimport fitz\nimport os\n\ndef redact_with_pymupdf(input_path: str, output_path: str, patterns: list[str]):\n    \"\"\"Redact specific patterns from PDF using PyMuPDF.\"\"\"\n    doc = fitz.open(input_path)\n    original_len = sum(len(p.get_text()) for p in doc)\n\n    # Find References page - stop redacting there\n    references_page = None\n    for i, page in enumerate(doc):\n        if \"references\" in page.get_text().lower():\n            references_page = i\n            break\n\n    for page_num, page in enumerate(doc):\n        if references_page is not None and page_num >= references_page:\n            continue  # Skip References section\n\n        for pattern in patterns:\n            # ONLY redact exact search matches\n            for rect in page.search_for(pattern):\n                page.add_redact_annot(rect, fill=(0, 0, 0))\n        page.apply_redactions()\n\n    os.makedirs(os.path.dirname(output_path), exist_ok=True)\n    doc.save(output_path)\n    doc.close()\n\n    # MUST verify after saving\n    verify_redaction(input_path, output_path)\n```\n\n## REQUIRED: Verification Function\n\n**Always run this after ANY redaction to catch errors early.**\n\nVerification MUST do two independent things:\n\n1. Structural \u002F retention checks (page count, character retention thresholds).\n2. **Category-taxonomy probes** derived directly from the 8 categories in \"Patterns to Redact\", executed on the pre-References pages of the redacted PDF. These probes are intentionally independent of whatever `patterns` list the planner happened to build for this paper, so that a category dropped during planning surfaces here as a leak instead of a false \"clean\" pass.\n\nOnly when both the structural checks pass AND every category probe returns no hit on the pre-References pages may the run be declared clean. If any probe fires, treat it as a real leak: extend that paper's `patterns` list (or record a justified exemption for a genuine false-positive, e.g., a body-text mention that is not authorship-revealing) and re-run redaction before finishing.\n\n```python\nimport fitz\nimport re\n\n# One lightweight probe per skill-listed category. These probes should be\n# broad enough to catch a category that was entirely forgotten during planning,\n# but they are NOT the redaction patterns themselves.\nCATEGORY_PROBES = {\n    \"emails (#3)\":                 re.compile(r\"[\\w.+-]+@[\\w-]+\\.[\\w.-]+\"),\n    \"arxiv ids (#5)\":              re.compile(r\"arXiv:\\s*\\d{4}\\.\\d{4,5}\", re.IGNORECASE),\n    \"dois (#6)\":                   re.compile(r\"\\b10\\.\\d{4,9}\u002F\\S+\"),\n    \"equal-contribution (#8)\":     re.compile(r\"equal\\s+contribution\", re.IGNORECASE),\n}\n\ndef _pre_refs_text(doc):\n    refs = None\n    for i, page in enumerate(doc):\n        if \"references\" in page.get_text().lower():\n            refs = i\n            break\n    end = refs if refs is not None else len(doc)\n    return \"\".join(doc[i].get_text() for i in range(end))\n\ndef verify_redaction(original_path, output_path):\n    \"\"\"Verify redaction didn't corrupt the PDF and that no skill-listed\n    category was silently skipped.\"\"\"\n    orig = fitz.open(original_path)\n    redc = fitz.open(output_path)\n\n    orig_len = sum(len(p.get_text()) for p in orig)\n    redc_len = sum(len(p.get_text()) for p in redc)\n\n    print(f\"Original: {len(orig)} pages, {orig_len} chars\")\n    print(f\"Redacted: {len(redc)} pages, {redc_len} chars\")\n    print(f\"Retained: {redc_len\u002Forig_len:.1%}\")\n\n    # DEFENSIVE CHECKS - fail fast if something went wrong\n    if len(redc) != len(orig):\n        raise ValueError(f\"Page count changed: {len(orig)} -> {len(redc)}\")\n    if redc_len \u003C 1000:\n        raise ValueError(f\"PDF corrupted: only {redc_len} chars remain!\")\n    if redc_len \u003C orig_len * 0.7:\n        raise ValueError(f\"Too much removed: kept only {redc_len\u002Forig_len:.0%}\")\n\n    # CATEGORY PROBES - one per skill-listed leakage category.\n    # These run on the redacted, pre-References text only.\n    pre_refs = _pre_refs_text(redc)\n    leaks = {}\n    for label, probe in CATEGORY_PROBES.items():\n        hits = probe.findall(pre_refs)\n        if hits:\n            leaks[label] = hits[:5]\n    if leaks:\n        raise ValueError(\n            \"Category-probe leaks in pre-References pages; extend patterns \"\n            f\"and re-redact: {leaks}\"\n        )\n\n    orig.close()\n    redc.close()\n    print(\"✓ Verification passed\")\n```\n\nNote: the probes above cover the categories that have canonical fixed-string \u002F regex forms (#3, #5, #6, #8). Categories #1, #2, #4, #7 are inherently document-specific (proper nouns), so their coverage is enforced by the per-document checklist under \"Patterns to Redact\" rather than by a generic probe here.\n",{"name":5,"description":7},"main","cdcb6d230f3b15c3c33f8a2bdd16439f43602cbe","evaluation\u002Fresults\u002FWML-Opus47-15Tasks-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Facademic-pdf-redaction",1,"https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark","https:\u002F\u002Fgithub.com\u002Flinyouxue","2026-11-14 20:28:35",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add linyouxue\u002Fskill-repair-benchmark --skill academic-pdf-redaction",{"tool":37,"command":38},"bunx","bunx skills add linyouxue\u002Fskill-repair-benchmark --skill academic-pdf-redaction",{"tool":40,"command":41},"pnpm","pnpm dlx skills add linyouxue\u002Fskill-repair-benchmark --skill academic-pdf-redaction","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002FWML-Opus47-15Tasks-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Facademic-pdf-redaction\u002FSKILL.md",true,[],[46,53,60,67,72,77],{"skillNo":47,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":50,"categories":51,"keywords":52,"publishedAt":19},"01M7E1HGVRY73NBJ022CJG13X6","video-frame-extraction","Extract frames from video files and save them as images using OpenCV","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fvideo-frame-extraction\u002FSKILL.md",[],[],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M7E1HAG5R7CJ9C8NYJS771MB","openai-vision","Analyze images and multi-frame sequences using OpenAI GPT vision models","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fopenai-vision\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M7E1H7DADKB6QBWKC2KQ63SZ","image-ocr","Extract text content from images using Tesseract OCR via Python, including structured field extraction from receipts and forms","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_03\u002Finputs\u002Fskills\u002Fimage-ocr\u002FSKILL.md",[],[],{"skillNo":68,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7E1H0WBNT43TENGQY7AVZHD","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fvideo-frame-extraction\u002FSKILL.md",[],[],{"skillNo":73,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7E1GTC7D4N4Q2EF03875J9E","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fopenai-vision\u002FSKILL.md",[],[],{"skillNo":78,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7E1GQ92CWFMGH53KSSE4WER","https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fjpg-ocr-stat\u002Fresult\u002Fskilltta-claude4.7\u002Fgroup_03\u002Frun_02\u002Finputs\u002Fskills\u002Fimage-ocr\u002FSKILL.md",[],[]]