[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C2RV7X81H9CKDB32P47RKM":3,"skill-same:01M7C2RV7X81H9CKDB32P47RKM":51},{"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},"01M7C2RV7X81H9CKDB32P47RKM","pause-detector",null,"Detect pauses and silence in audio using local dynamic thresholds. Use when you need to find natural pauses in lectures, board-writing silences, or breaks between sections. Uses local context comparison to avoid false positives from volume variation.","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\u002Fvideo-silence-remover\u002Frepaired_run\u002Finputs\u002Fskills\u002Fpause-detector\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: pause-detector\ndescription: Detect pauses and silence in audio using local dynamic thresholds. Use when you need to find natural pauses in lectures, board-writing silences, or breaks between sections. Uses local context comparison to avoid false positives from volume variation.\n---\n\n# Pause Detector\n\nDetects pauses and low-energy segments using local dynamic thresholds on pre-computed energy data. Unlike global threshold methods, this compares energy to surrounding context, avoiding false positives when speaker volume varies.\n\n## Use Cases\n\n- Detecting natural pauses in lectures\n- Finding board-writing silences\n- Identifying breaks between sections\n\n## Usage\n\n```bash\npython3 \u002Froot\u002F.claude\u002Fskills\u002Fpause-detector\u002Fscripts\u002Fdetect_pauses.py \\\n    --energies \u002Fpath\u002Fto\u002Fenergies.json \\\n    --output \u002Fpath\u002Fto\u002Fpauses.json\n```\n\n### Parameters\n\n- `--energies`: Path to energy JSON file (from energy-calculator)\n- `--output`: Path to output JSON file\n- `--start-time`: Start analyzing from this second (default: 0)\n- `--threshold-ratio`: Ratio of local average for low energy (default: 0.5)\n- `--min-duration`: Minimum pause duration in seconds (default: 2)\n- `--window-size`: Local average window size (default: 30)\n\n### Output Format\n\n```json\n{\n  \"method\": \"local_dynamic_threshold\",\n  \"segments\": [\n    {\"start\": 610, \"end\": 613, \"duration\": 3},\n    {\"start\": 720, \"end\": 724, \"duration\": 4}\n  ],\n  \"total_segments\": 11,\n  \"total_duration_seconds\": 28,\n  \"parameters\": {\n    \"threshold_ratio\": 0.5,\n    \"window_size\": 30,\n    \"min_duration\": 2\n  }\n}\n```\n\n## How It Works\n\n1. Load pre-computed energy data\n2. Calculate local average using sliding window\n3. Mark seconds where energy \u003C local_avg × threshold_ratio\n4. Group consecutive low-energy seconds into segments\n5. Filter segments by minimum duration\n\n## Dependencies\n\n- Python 3.11+\n- numpy\n- scipy\n\n## Example\n\n```bash\n# Detect pauses after opening (start at 221s)\npython3 \u002Froot\u002F.claude\u002Fskills\u002Fpause-detector\u002Fscripts\u002Fdetect_pauses.py \\\n    --energies energies.json \\\n    --start-time 221 \\\n    --output pauses.json\n\n# Result: Found 11 pauses totaling 28 seconds\n```\n\n## Parameters Tuning\n\n| Parameter | Lower Value | Higher Value |\n|-----------|-------------|--------------|\n| `threshold_ratio` | More aggressive | More conservative |\n| `min_duration` | Shorter pauses | Longer pauses only |\n| `window_size` | Local context | Broader context |\n\n### Post-processing for precision\n\nThe detector's defaults favor recall: with a 1-second energy grid and `min_duration=2`, the raw output frequently contains (a) multiple short segments separated by 1–2 s gaps that actually belong to the same underlying pause, and (b) borderline 2 s hits sitting exactly at the minimum-duration cutoff. Forwarding this raw list into `segment-combiner` \u002F `video-processor` inflates the segment count and reduces IoU precision against ground-truth pauses, even when total removed duration and recall are correct.\n\nWhen the downstream evaluation weighs per-segment precision or IoU (e.g., \"remove long pauses > 2 s\", or scoring individual removed intervals), post-process the candidate list before combining:\n\n1. **Gap-merge near-adjacent segments.** Iterate segments in start order and merge any pair whose inter-segment gap is ≤ a small tolerance (1–2 s is a reasonable default for 1 s energy windows). This collapses fragmented pieces of the same pause into one interval and typically eliminates the largest source of false positives.\n2. **Reconsider borderline hits.** If, after merging, many surviving segments are still exactly at `min_duration` (e.g., a cluster of 2 s hits), prefer either raising `--min-duration` (e.g., to 3) or lowering `--threshold-ratio` (e.g., 0.3–0.4) and re-running detection. Either change trades a small amount of recall for higher precision.\n3. **Sanity-check candidate count vs. task guidance.** If the task describes pauses qualitatively (e.g., \"long pauses, usually > 2 s\") and the raw output returns many more candidates than that description implies, treat it as a signal to apply steps 1–2 rather than pass the raw list through.\n\nA minimal gap-merge is straightforward to apply to the JSON output without changing the detector:\n\n```python\nimport json\n\nGAP_TOL = 1  # seconds; merge segments separated by \u003C= GAP_TOL\n\ndata = json.load(open(\"pauses.json\"))\nsegs = sorted(data[\"segments\"], key=lambda s: s[\"start\"])\nmerged = []\nfor s in segs:\n    if merged and s[\"start\"] - merged[-1][\"end\"] \u003C= GAP_TOL:\n        merged[-1][\"end\"] = max(merged[-1][\"end\"], s[\"end\"])\n        merged[-1][\"duration\"] = merged[-1][\"end\"] - merged[-1][\"start\"]\n    else:\n        merged.append(dict(s))\n\ndata[\"segments\"] = merged\ndata[\"total_segments\"] = len(merged)\ndata[\"total_duration_seconds\"] = sum(s[\"duration\"] for s in merged)\njson.dump(data, open(\"pauses.json\", \"w\"), indent=2)\n```\n\nThe merged file remains compatible with `segment-combiner` and `video-processor`; total removed duration and math consistency are preserved (merging only removes intra-pause gaps of ≤ `GAP_TOL` seconds, which are themselves low-energy).\n\n**Verification \u002F fallback.** After post-processing, re-check: (i) the number of surviving segments is closer to the qualitative description in the task, (ii) no two segments are within `GAP_TOL` of each other, and (iii) the sum of `duration` still matches what `segment-combiner` and the final report show. If gap-merging alone leaves many borderline 2 s hits, re-run detection with a tighter `--threshold-ratio` or a larger `--min-duration` and repeat the check.\n\n## Notes\n\n- Requires energy data from energy-calculator skill\n- Use --start-time to skip detected opening\n- Local threshold adapts to varying speaker volume\n\n### Pipeline usage\n\nInside a detector → combiner → processor pipeline, treat this skill's output as **candidates**, not as the final removal list. Before invoking `segment-combiner`, briefly inspect the candidate segments and decide based on the task's evaluation focus:\n\n- If the evaluator scores per-segment precision or IoU on removed intervals, apply the *Post-processing for precision* steps above (gap-merge, and if needed re-tune `threshold_ratio` \u002F `min_duration`) before combining.\n- If the evaluator only checks total removed duration, compression ratio, or math consistency, the default recall-friendly output is usually sufficient and no post-processing is required.\n\nThis lightweight review gate between detection and combination keeps the recall-oriented defaults intact while avoiding forwarding over-segmented candidates verbatim to video processing.\n","# Pause Detector\n\nDetects pauses and low-energy segments using local dynamic thresholds on pre-computed energy data. Unlike global threshold methods, this compares energy to surrounding context, avoiding false positives when speaker volume varies.\n\n## Use Cases\n\n- Detecting natural pauses in lectures\n- Finding board-writing silences\n- Identifying breaks between sections\n\n## Usage\n\n```bash\npython3 \u002Froot\u002F.claude\u002Fskills\u002Fpause-detector\u002Fscripts\u002Fdetect_pauses.py \\\n    --energies \u002Fpath\u002Fto\u002Fenergies.json \\\n    --output \u002Fpath\u002Fto\u002Fpauses.json\n```\n\n### Parameters\n\n- `--energies`: Path to energy JSON file (from energy-calculator)\n- `--output`: Path to output JSON file\n- `--start-time`: Start analyzing from this second (default: 0)\n- `--threshold-ratio`: Ratio of local average for low energy (default: 0.5)\n- `--min-duration`: Minimum pause duration in seconds (default: 2)\n- `--window-size`: Local average window size (default: 30)\n\n### Output Format\n\n```json\n{\n  \"method\": \"local_dynamic_threshold\",\n  \"segments\": [\n    {\"start\": 610, \"end\": 613, \"duration\": 3},\n    {\"start\": 720, \"end\": 724, \"duration\": 4}\n  ],\n  \"total_segments\": 11,\n  \"total_duration_seconds\": 28,\n  \"parameters\": {\n    \"threshold_ratio\": 0.5,\n    \"window_size\": 30,\n    \"min_duration\": 2\n  }\n}\n```\n\n## How It Works\n\n1. Load pre-computed energy data\n2. Calculate local average using sliding window\n3. Mark seconds where energy \u003C local_avg × threshold_ratio\n4. Group consecutive low-energy seconds into segments\n5. Filter segments by minimum duration\n\n## Dependencies\n\n- Python 3.11+\n- numpy\n- scipy\n\n## Example\n\n```bash\n# Detect pauses after opening (start at 221s)\npython3 \u002Froot\u002F.claude\u002Fskills\u002Fpause-detector\u002Fscripts\u002Fdetect_pauses.py \\\n    --energies energies.json \\\n    --start-time 221 \\\n    --output pauses.json\n\n# Result: Found 11 pauses totaling 28 seconds\n```\n\n## Parameters Tuning\n\n| Parameter | Lower Value | Higher Value |\n|-----------|-------------|--------------|\n| `threshold_ratio` | More aggressive | More conservative |\n| `min_duration` | Shorter pauses | Longer pauses only |\n| `window_size` | Local context | Broader context |\n\n### Post-processing for precision\n\nThe detector's defaults favor recall: with a 1-second energy grid and `min_duration=2`, the raw output frequently contains (a) multiple short segments separated by 1–2 s gaps that actually belong to the same underlying pause, and (b) borderline 2 s hits sitting exactly at the minimum-duration cutoff. Forwarding this raw list into `segment-combiner` \u002F `video-processor` inflates the segment count and reduces IoU precision against ground-truth pauses, even when total removed duration and recall are correct.\n\nWhen the downstream evaluation weighs per-segment precision or IoU (e.g., \"remove long pauses > 2 s\", or scoring individual removed intervals), post-process the candidate list before combining:\n\n1. **Gap-merge near-adjacent segments.** Iterate segments in start order and merge any pair whose inter-segment gap is ≤ a small tolerance (1–2 s is a reasonable default for 1 s energy windows). This collapses fragmented pieces of the same pause into one interval and typically eliminates the largest source of false positives.\n2. **Reconsider borderline hits.** If, after merging, many surviving segments are still exactly at `min_duration` (e.g., a cluster of 2 s hits), prefer either raising `--min-duration` (e.g., to 3) or lowering `--threshold-ratio` (e.g., 0.3–0.4) and re-running detection. Either change trades a small amount of recall for higher precision.\n3. **Sanity-check candidate count vs. task guidance.** If the task describes pauses qualitatively (e.g., \"long pauses, usually > 2 s\") and the raw output returns many more candidates than that description implies, treat it as a signal to apply steps 1–2 rather than pass the raw list through.\n\nA minimal gap-merge is straightforward to apply to the JSON output without changing the detector:\n\n```python\nimport json\n\nGAP_TOL = 1  # seconds; merge segments separated by \u003C= GAP_TOL\n\ndata = json.load(open(\"pauses.json\"))\nsegs = sorted(data[\"segments\"], key=lambda s: s[\"start\"])\nmerged = []\nfor s in segs:\n    if merged and s[\"start\"] - merged[-1][\"end\"] \u003C= GAP_TOL:\n        merged[-1][\"end\"] = max(merged[-1][\"end\"], s[\"end\"])\n        merged[-1][\"duration\"] = merged[-1][\"end\"] - merged[-1][\"start\"]\n    else:\n        merged.append(dict(s))\n\ndata[\"segments\"] = merged\ndata[\"total_segments\"] = len(merged)\ndata[\"total_duration_seconds\"] = sum(s[\"duration\"] for s in merged)\njson.dump(data, open(\"pauses.json\", \"w\"), indent=2)\n```\n\nThe merged file remains compatible with `segment-combiner` and `video-processor`; total removed duration and math consistency are preserved (merging only removes intra-pause gaps of ≤ `GAP_TOL` seconds, which are themselves low-energy).\n\n**Verification \u002F fallback.** After post-processing, re-check: (i) the number of surviving segments is closer to the qualitative description in the task, (ii) no two segments are within `GAP_TOL` of each other, and (iii) the sum of `duration` still matches what `segment-combiner` and the final report show. If gap-merging alone leaves many borderline 2 s hits, re-run detection with a tighter `--threshold-ratio` or a larger `--min-duration` and repeat the check.\n\n## Notes\n\n- Requires energy data from energy-calculator skill\n- Use --start-time to skip detected opening\n- Local threshold adapts to varying speaker volume\n\n### Pipeline usage\n\nInside a detector → combiner → processor pipeline, treat this skill's output as **candidates**, not as the final removal list. Before invoking `segment-combiner`, briefly inspect the candidate segments and decide based on the task's evaluation focus:\n\n- If the evaluator scores per-segment precision or IoU on removed intervals, apply the *Post-processing for precision* steps above (gap-merge, and if needed re-tune `threshold_ratio` \u002F `min_duration`) before combining.\n- If the evaluator only checks total removed duration, compression ratio, or math consistency, the default recall-friendly output is usually sufficient and no post-processing is required.\n\nThis lightweight review gate between detection and combination keeps the recall-oriented defaults intact while avoiding forwarding over-segmented candidates verbatim to video processing.\n",{"name":5,"description":7},"main","cdcb6d230f3b15c3c33f8a2bdd16439f43602cbe","evaluation\u002Fresults\u002FWML-Opus47-15Tasks-20260923\u002Ftasks\u002Fvideo-silence-remover\u002Frepaired_run\u002Finputs\u002Fskills\u002Fpause-detector",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 pause-detector",{"tool":37,"command":38},"bunx","bunx skills add linyouxue\u002Fskill-repair-benchmark --skill pause-detector",{"tool":40,"command":41},"pnpm","pnpm dlx skills add linyouxue\u002Fskill-repair-benchmark --skill pause-detector","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\u002Fvideo-silence-remover\u002Frepaired_run\u002Finputs\u002Fskills\u002Fpause-detector\u002FSKILL.md",true,[45],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"scripts\u002Fdetect_pauses.py","SCRIPT","脚本",3372,"https:\u002F\u002Fgithub.com\u002Flinyouxue\u002Fskill-repair-benchmark\u002Fblob\u002Fmain\u002Fevaluation\u002Fresults\u002FWML-Opus47-15Tasks-20260923\u002Ftasks\u002Fvideo-silence-remover\u002Frepaired_run\u002Finputs\u002Fskills\u002Fpause-detector\u002Fscripts\u002Fdetect_pauses.py",[52,59,66,73,78,83],{"skillNo":53,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"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":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"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":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"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":74,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"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":79,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"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":84,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"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",[],[]]