[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C3MTY51Q3KS3G88ZN76XNP":3,"skill-same:01M7C3MTY51Q3KS3G88ZN76XNP":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},"01M7C3MTY51Q3KS3G88ZN76XNP","skilltta-task-guidance",null,"Task-specific operational guidance synthesized at test time from retrieved training examples.","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\u002Fskilltta-synthesis-opus47-tatumass-failed15-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Fskilltta-task-guidance\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: skilltta-task-guidance\ndescription: Task-specific operational guidance synthesized at test time from retrieved training examples.\n---\n\n# SKILL.md\n\n## When to use\nUse this skill when the task requires producing anonymized\u002Fredacted versions of PDF documents so that authorship-revealing content is removed while the scholarly content remains readable. The binding contract for this class of task typically specifies:\n- Input PDFs at known paths and output PDFs at a specified directory\u002Ffilename pattern.\n- Categories of information to remove: author names, affiliations, contact info, and any indirect leaks (arXiv IDs, DOIs\u002FURLs pointing to author pages, venue\u002Facceptance notices, funding\u002Facknowledgement blocks, ORCID, footers\u002Fheaders, watermarks, and metadata).\n- A tolerance clause (e.g., self-citations may be acceptable *only if* other identifying info is already removed).\n\nBefore acting, gather:\n- The exact input\u002Foutput paths and filename mapping.\n- The list of leakage categories the prompt enumerates (treat it as inclusive, not exhaustive).\n- The tools available (e.g., `pypdf`, `pdfplumber`, `pymupdf`\u002F`fitz`, `pikepdf`, `ocrmypdf`, `pdf2image` + OCR).\n- Whether PDFs are text-based or scanned (affects whether OCR is required).\n- Whether PDF metadata (Author, Title, Subject, Keywords, XMP) also needs scrubbing (assume yes for anonymization).\n\n## Possible Failure Modes\n- **Text-layer only redaction that leaves visible glyphs**: replacing text in the content stream without covering the rendered characters, so names still appear when opened in a viewer.\n- **Using `pypdf` \"remove text\" tricks that don't actually redact**: many pypdf operations only remove text objects but leave images\u002Fvector artifacts; also they may corrupt layout.\n- **Forgetting PDF metadata**: `\u002FAuthor`, `\u002FTitle`, XMP metadata, and embedded document info often still contain author names after visual redaction.\n- **Missing headers\u002Ffooters on every page**: running-headers with author names or venue names (\"Accepted at NeurIPS 2024\") repeat on many pages and are easy to miss.\n- **Missing the first-page author block**: title-page author list, affiliations, and email addresses are the primary leak surface.\n- **Missing arXiv identifier watermark**: arXiv adds a vertical sidebar like \"arXiv:2xxx.xxxxx [cs.CL]\" — must be redacted.\n- **Missing DOIs, URLs, ORCID iDs, GitHub\u002Fpersonal links** that indirectly reveal identity.\n- **Missing acknowledgements\u002Ffunding sections** that name authors, grants tied to labs, or advisors.\n- **Not covering images\u002Flogos**: institutional logos on title page or figures.\n- **Applying redaction but not \"applying\" it**: with PyMuPDF, `page.add_redact_annot(...)` must be followed by `page.apply_redactions()`; otherwise the annotation is decorative and text is still extractable.\n- **Overzealous redaction** that removes body content, section headers, or citations unnecessarily, harming readability.\n- **Wrong output path or overwriting inputs**: not creating the output directory, or writing to the input location.\n- **Silent failures on one of several inputs**: processing paper1 succeeds but paper2\u002F3 raise and are skipped without notice.\n- **Copying from retrieved examples**: those examples are about encryption\u002Fdata generation and are not directly applicable — do not adapt their code paths as if they were the solution.\n\n## Possible procedures\n1. **Set up output**: ensure the target output directory exists; iterate over each input PDF; map to its corresponding output filename per the contract's pattern.\n2. **Detect PDF type**: try text extraction first (e.g., PyMuPDF `page.get_text(\"text\")` \u002F \"words\" \u002F \"dict\"). If pages return empty text, treat as scanned and either OCR (`ocrmypdf`) first or draw opaque rectangles over identified regions using image-based detection.\n3. **Identify leakage regions per page**:\n   - Title page: author names line(s), affiliations, emails, ORCID, footnote symbols linking to affiliations, corresponding-author notes.\n   - Sidebar\u002Fmargins: arXiv identifier stamp (usually rotated text on left edge of page 1).\n   - Header\u002Ffooter: running titles including author names, venue notices (\"To appear in ...\", \"Accepted at ...\"), page number blocks that include author name, watermarks.\n   - End-of-paper: Acknowledgements section, Author Contributions, Funding, Author Biographies, Correspondence.\n   - Anywhere: email addresses (regex `\\S+@\\S+\\.\\S+`), URLs pointing to personal\u002Flab pages, GitHub repos containing author handles, ORCID (`\\d{4}-\\d{4}-\\d{4}-\\d{3}[\\dX]`), arXiv IDs (`arXiv:\\d{4}\\.\\d{4,5}`), DOIs when tied to venue.\n4. **Preferred implementation** (PyMuPDF pattern):\n   - For each keyword\u002Fregex hit, use `page.search_for(text)` or iterate `page.get_text(\"words\")` and match; collect bounding boxes.\n   - Call `page.add_redact_annot(bbox, fill=(0,0,0))` (or white) for each box.\n   - Call `page.apply_redactions()` — this actually removes the underlying text and draws the fill.\n5. **Scrub metadata**: set document metadata fields (Author, Title if it contains names, Subject, Keywords, Creator, Producer as appropriate) to empty strings; also strip XMP metadata (`doc.set_xml_metadata(\"\")` in PyMuPDF, or use `pikepdf` to clear `docinfo` and XMP).\n6. **Self-citation policy**: leave in-text citations and reference list entries alone by default; only redact reference entries if they still expose the author's name and no other identifier has been left. The contract typically allows self-citations if other leaks are handled.\n7. **Save**: write to the specified output path with `garbage=4, deflate=True` (PyMuPDF) or equivalent to ensure removed objects are not recoverable.\n8. **Recovery checks**: after saving, re-open the redacted PDF, extract all text, and grep for the identifying tokens you found earlier. If any still appear, add them to the redaction pass and re-run.\n\nDecision points:\n- If text extraction is empty → OCR path or image-region redaction.\n- If a name appears in ligatures or split across spans, `search_for` may miss it — fall back to word-by-word matching or use `pdfplumber` for token positions.\n- If metadata cannot be edited with one library, fall back to `pikepdf`\u002F`exiftool`.\n\n## Verification Checklist\n- [ ] Every required input file has produced exactly one output file at the exact path\u002Fpattern the contract specifies.\n- [ ] Output directory was created; input files are unmodified.\n- [ ] Extracting text from each output PDF yields no author names, no emails, no ORCID, no arXiv identifier, no personal\u002Flab URLs, and no venue\u002Facceptance strings.\n- [ ] Title-page author\u002Faffiliation block is visually covered (open the PDF or render page 1 to image to confirm).\n- [ ] Headers\u002Ffooters and margin watermarks (including rotated arXiv stamp) are covered on all pages, not just page 1.\n- [ ] Acknowledgements \u002F funding \u002F author-contributions sections are redacted or verified to be non-identifying.\n- [ ] PDF document metadata (Info dict and XMP) contains no author-identifying values.\n- [ ] Body text, figures, tables, and citations unrelated to identity are preserved and readable.\n- [ ] `page.apply_redactions()` (or equivalent) was actually called; a text-extraction spot check confirms redacted strings are gone, not merely hidden.\n- [ ] Self-citations, if left in, are only left in because all other identifying info has been removed — re-verify by searching for author surnames across the whole document.\n- [ ] Errors on any single input do not silently skip that file; each output exists and is a valid PDF.\n- [ ] No content, code, or specific values were copied from unrelated retrieved examples (those examples concern encryption\u002Fdata generation and are not applicable here).\n","# SKILL.md\n\n## When to use\nUse this skill when the task requires producing anonymized\u002Fredacted versions of PDF documents so that authorship-revealing content is removed while the scholarly content remains readable. The binding contract for this class of task typically specifies:\n- Input PDFs at known paths and output PDFs at a specified directory\u002Ffilename pattern.\n- Categories of information to remove: author names, affiliations, contact info, and any indirect leaks (arXiv IDs, DOIs\u002FURLs pointing to author pages, venue\u002Facceptance notices, funding\u002Facknowledgement blocks, ORCID, footers\u002Fheaders, watermarks, and metadata).\n- A tolerance clause (e.g., self-citations may be acceptable *only if* other identifying info is already removed).\n\nBefore acting, gather:\n- The exact input\u002Foutput paths and filename mapping.\n- The list of leakage categories the prompt enumerates (treat it as inclusive, not exhaustive).\n- The tools available (e.g., `pypdf`, `pdfplumber`, `pymupdf`\u002F`fitz`, `pikepdf`, `ocrmypdf`, `pdf2image` + OCR).\n- Whether PDFs are text-based or scanned (affects whether OCR is required).\n- Whether PDF metadata (Author, Title, Subject, Keywords, XMP) also needs scrubbing (assume yes for anonymization).\n\n## Possible Failure Modes\n- **Text-layer only redaction that leaves visible glyphs**: replacing text in the content stream without covering the rendered characters, so names still appear when opened in a viewer.\n- **Using `pypdf` \"remove text\" tricks that don't actually redact**: many pypdf operations only remove text objects but leave images\u002Fvector artifacts; also they may corrupt layout.\n- **Forgetting PDF metadata**: `\u002FAuthor`, `\u002FTitle`, XMP metadata, and embedded document info often still contain author names after visual redaction.\n- **Missing headers\u002Ffooters on every page**: running-headers with author names or venue names (\"Accepted at NeurIPS 2024\") repeat on many pages and are easy to miss.\n- **Missing the first-page author block**: title-page author list, affiliations, and email addresses are the primary leak surface.\n- **Missing arXiv identifier watermark**: arXiv adds a vertical sidebar like \"arXiv:2xxx.xxxxx [cs.CL]\" — must be redacted.\n- **Missing DOIs, URLs, ORCID iDs, GitHub\u002Fpersonal links** that indirectly reveal identity.\n- **Missing acknowledgements\u002Ffunding sections** that name authors, grants tied to labs, or advisors.\n- **Not covering images\u002Flogos**: institutional logos on title page or figures.\n- **Applying redaction but not \"applying\" it**: with PyMuPDF, `page.add_redact_annot(...)` must be followed by `page.apply_redactions()`; otherwise the annotation is decorative and text is still extractable.\n- **Overzealous redaction** that removes body content, section headers, or citations unnecessarily, harming readability.\n- **Wrong output path or overwriting inputs**: not creating the output directory, or writing to the input location.\n- **Silent failures on one of several inputs**: processing paper1 succeeds but paper2\u002F3 raise and are skipped without notice.\n- **Copying from retrieved examples**: those examples are about encryption\u002Fdata generation and are not directly applicable — do not adapt their code paths as if they were the solution.\n\n## Possible procedures\n1. **Set up output**: ensure the target output directory exists; iterate over each input PDF; map to its corresponding output filename per the contract's pattern.\n2. **Detect PDF type**: try text extraction first (e.g., PyMuPDF `page.get_text(\"text\")` \u002F \"words\" \u002F \"dict\"). If pages return empty text, treat as scanned and either OCR (`ocrmypdf`) first or draw opaque rectangles over identified regions using image-based detection.\n3. **Identify leakage regions per page**:\n   - Title page: author names line(s), affiliations, emails, ORCID, footnote symbols linking to affiliations, corresponding-author notes.\n   - Sidebar\u002Fmargins: arXiv identifier stamp (usually rotated text on left edge of page 1).\n   - Header\u002Ffooter: running titles including author names, venue notices (\"To appear in ...\", \"Accepted at ...\"), page number blocks that include author name, watermarks.\n   - End-of-paper: Acknowledgements section, Author Contributions, Funding, Author Biographies, Correspondence.\n   - Anywhere: email addresses (regex `\\S+@\\S+\\.\\S+`), URLs pointing to personal\u002Flab pages, GitHub repos containing author handles, ORCID (`\\d{4}-\\d{4}-\\d{4}-\\d{3}[\\dX]`), arXiv IDs (`arXiv:\\d{4}\\.\\d{4,5}`), DOIs when tied to venue.\n4. **Preferred implementation** (PyMuPDF pattern):\n   - For each keyword\u002Fregex hit, use `page.search_for(text)` or iterate `page.get_text(\"words\")` and match; collect bounding boxes.\n   - Call `page.add_redact_annot(bbox, fill=(0,0,0))` (or white) for each box.\n   - Call `page.apply_redactions()` — this actually removes the underlying text and draws the fill.\n5. **Scrub metadata**: set document metadata fields (Author, Title if it contains names, Subject, Keywords, Creator, Producer as appropriate) to empty strings; also strip XMP metadata (`doc.set_xml_metadata(\"\")` in PyMuPDF, or use `pikepdf` to clear `docinfo` and XMP).\n6. **Self-citation policy**: leave in-text citations and reference list entries alone by default; only redact reference entries if they still expose the author's name and no other identifier has been left. The contract typically allows self-citations if other leaks are handled.\n7. **Save**: write to the specified output path with `garbage=4, deflate=True` (PyMuPDF) or equivalent to ensure removed objects are not recoverable.\n8. **Recovery checks**: after saving, re-open the redacted PDF, extract all text, and grep for the identifying tokens you found earlier. If any still appear, add them to the redaction pass and re-run.\n\nDecision points:\n- If text extraction is empty → OCR path or image-region redaction.\n- If a name appears in ligatures or split across spans, `search_for` may miss it — fall back to word-by-word matching or use `pdfplumber` for token positions.\n- If metadata cannot be edited with one library, fall back to `pikepdf`\u002F`exiftool`.\n\n## Verification Checklist\n- [ ] Every required input file has produced exactly one output file at the exact path\u002Fpattern the contract specifies.\n- [ ] Output directory was created; input files are unmodified.\n- [ ] Extracting text from each output PDF yields no author names, no emails, no ORCID, no arXiv identifier, no personal\u002Flab URLs, and no venue\u002Facceptance strings.\n- [ ] Title-page author\u002Faffiliation block is visually covered (open the PDF or render page 1 to image to confirm).\n- [ ] Headers\u002Ffooters and margin watermarks (including rotated arXiv stamp) are covered on all pages, not just page 1.\n- [ ] Acknowledgements \u002F funding \u002F author-contributions sections are redacted or verified to be non-identifying.\n- [ ] PDF document metadata (Info dict and XMP) contains no author-identifying values.\n- [ ] Body text, figures, tables, and citations unrelated to identity are preserved and readable.\n- [ ] `page.apply_redactions()` (or equivalent) was actually called; a text-extraction spot check confirms redacted strings are gone, not merely hidden.\n- [ ] Self-citations, if left in, are only left in because all other identifying info has been removed — re-verify by searching for author surnames across the whole document.\n- [ ] Errors on any single input do not silently skip that file; each output exists and is a valid PDF.\n- [ ] No content, code, or specific values were copied from unrelated retrieved examples (those examples concern encryption\u002Fdata generation and are not applicable here).\n",{"name":5,"description":7},"main","cdcb6d230f3b15c3c33f8a2bdd16439f43602cbe","evaluation\u002Fresults\u002Fskilltta-synthesis-opus47-tatumass-failed15-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Fskilltta-task-guidance",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 skilltta-task-guidance",{"tool":37,"command":38},"bunx","bunx skills add linyouxue\u002Fskill-repair-benchmark --skill skilltta-task-guidance",{"tool":40,"command":41},"pnpm","pnpm dlx skills add linyouxue\u002Fskill-repair-benchmark --skill skilltta-task-guidance","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002Fskilltta-synthesis-opus47-tatumass-failed15-20260923\u002Ftasks\u002Fpaper-anonymizer\u002Frepaired_run\u002Finputs\u002Fskills\u002Fskilltta-task-guidance\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",[],[]]