[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7878GPTS1F5MCAQC5BPWN1Z":3,"skill-same:01M7878GPTS1F5MCAQC5BPWN1Z":80},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M7878GPTS1F5MCAQC5BPWN1Z","url-reader",null,"Enriches URL-only notes, Notion\u002Fbookmark inbox items, and captured social post links by using domain-aware reader backends, an automatic Browser4 fallback, and an agent-executed in-app Browser fallback when required to produce stable Markdown, metadata, image links, and structured failure states. Use when Codex needs to normalize twitter.com links to x.com, read X\u002FTwitter posts or X Articles, inspect whether Instagram posts can be read, download extracted public image assets, or classify login walls, blocked domains, partial extraction, and unsupported private URLs. Do not use for general web browsing, latest-news research, or official API documentation lookup.","yoshiysh\u002Fclaude-plugins","yoshiysh","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F22577999?v=4","01M7875QRN03M4YGSDHPKZRVGP","01M7875QSCPC7R4491P3RV7RFR",1,"MIT","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: url-reader\ndescription: >\n  Enriches URL-only notes, Notion\u002Fbookmark inbox items, and captured social post links by using domain-aware reader backends, an automatic Browser4 fallback, and an agent-executed in-app Browser fallback when required to produce stable Markdown, metadata, image links, and structured failure states.\n  Use when Codex needs to normalize twitter.com links to x.com, read X\u002FTwitter posts or X Articles, inspect whether Instagram posts can be read, download extracted public image assets, or classify login walls, blocked domains, partial extraction, and unsupported private URLs. Do not use for general web browsing, latest-news research, or official API documentation lookup.\n---\n\n# URL Reader\n\nUse a domain-aware read-only enrichment layer before summarizing, classifying, or moving URL-only notes. Prefer this skill when a page has too little local context, when Notion only exposes an embed block, or when the user asks whether a social URL can be read.\n\n## Quick Start\n\nRun the bundled script with the original target URL. The script owns backend and Browser4 attempts, but the calling agent owns the in-app Browser handoff:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...'\n```\n\nWhen the normal reader returns anything other than `Extracted`, `read_url.py` automatically tries Browser4 and keeps whichever result contains better evidence. Browser4 is optional: if `browser4-cli` is unavailable, the original reader result is preserved and the failed Browser4 attempt is recorded. If the public URL still remains incomplete, the JSON contains `browser_fallback.required: true`; the calling agent must then execute the in-app Browser final-fallback protocol in [references\u002Fin-app-browser-fallback.md](references\u002Fin-app-browser-fallback.md) without asking the user for permission. This is a fixed retrieval stage, not a decision to ask the user about.\n\nTo enable the fallback, install Browser4 and its self-contained runtime, then make `browser4-cli` available on `PATH`. A non-standard CLI location can be supplied with `BROWSER4_CLI=\u002Fabsolute\u002Fpath\u002Fto\u002Fbrowser4-cli`. The fallback uses a temporary headless profile and deterministic DOM\u002Ftext commands; it does not run Browser4's autonomous agent, enter credentials, or mutate the target site.\n\nFor machine-readable output:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...' --json\n```\n\nTo save extracted image assets:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...' --json --download-images \u002Ftmp\u002Furl-reader-images\n```\n\n## Rules\n\n- Pass the original URL to the script. Do not manually add reader-service prefixes.\n- Let the script pick the retrieval path from the URL domain. Keep the skill instructions generic so future domain-specific fetchers can be added without changing user prompts.\n- If a URL already contains one or more reader-service prefixes, let the script unwrap them instead of manually editing the URL.\n- Normalize `twitter.com`, `www.twitter.com`, and `mobile.twitter.com` URLs to `x.com` before extraction.\n- Refuse localhost, private IPs, and internal hostnames. Public URL readers should not receive private or machine-local URLs.\n- Treat reader output as extracted public web content, not as an authoritative API response. Keep the original `URL Source` with any summary.\n- Keep failures explicit. Do not infer hidden post text when the reader returns a login page, 403, domain block, empty content, or unrelated navigation.\n\n## Domain Routing\n\nUse these reader paths:\n\n- `x.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>`: fetch `publish.x.com\u002Foembed` first. Use the returned blockquote paragraph as `markdown`, keep `author_name`, `author_url`, and the date-like link text when present, and report `reader_backend: x_oembed`. If that paragraph is only one `t.co` link, derive `x.com\u002F\u003Cuser>\u002Farticle\u002F\u003Cid>` from the original status URL and try the X Article route before treating the post as link-only.\n- `twitter.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>` and mobile\u002Fwww variants: normalize to the equivalent `x.com` URL, then use the X oEmbed path.\n- `twitter.com\u002Fi\u002Fweb\u002Fstatus\u002F\u003Cid>`: normalize to `x.com\u002Fi\u002Fstatus\u002F\u003Cid>` before using X oEmbed.\n- `x.com\u002F\u003Cuser>\u002Farticle\u002F\u003Cid>`: preserve the canonical Article URL as `normalized_url` and `source_url`, and use the generic reader for the Article page. If the result is incomplete, `read_url.py` automatically tries Browser4 against that same canonical Article URL; if it remains incomplete, it sets `browser_fallback.required: true` and the calling agent must use in-app Browser against that exact URL. Do not replace the Article URL with a derived status URL. An opaque `x.com\u002Fi\u002Farticle\u002F\u003Cid>` URL must remain opaque; do not infer an author or a same-ID status URL.\n- Instagram post URLs: use the generic reader path. It can often return the caption, author link, location, hashtags, and signed image URLs.\n- Instagram Reel URLs (`instagram.com\u002Freel\u002F...`): if the reader returns an Instagram login page or generic shell, report `Blocked` and do not treat extracted login assets as usable images.\n- Other public URLs: use the generic reader path first. If the result is `Partial`, `ImagesOnly`, `Blocked`, or `Failed`, `read_url.py` automatically tries Browser4 and then emits the mandatory in-app Browser handoff if the result is still incomplete. Do not omit the handoff because the domain is ordinary.\n- GitHub URLs: run the generic reader first. If it returns `403`, an anonymous-access block, rate limiting, or missing repository content, run `[SKILL_DIR]\u002Fscripts\u002Fread_github_cli.py` when an authenticated `gh` CLI is available before the in-app Browser final fallback. The script uses the CLI's keyring-backed login and never accepts or prints a PAT. If CLI evidence is still incomplete or the CLI is unavailable, execute the required in-app Browser fallback; do not ask the user to choose between them.\n\nX oEmbed is sufficient for post text and author metadata, but it does not reliably expose attached media URLs. If the user asks for X images or video, mark the text extraction separately from media extraction and use another backend or `Needs Review` for media.\n\n## Interpreting Results\n\nUse these extraction statuses:\n\n- `Extracted`: title and meaningful body text are present.\n- `Partial`: title, metadata, image links, or a short caption are present, but the page body is incomplete.\n- `ImagesOnly`: image URLs are present but the textual content is missing or mostly boilerplate.\n- `Blocked`: the selected reader returns an error such as `403`, login wall, domain block, or access denial.\n- `Failed`: network, timeout, invalid URL, or parser failure.\n\nFor Instagram, image URLs may expire, so download them immediately when the user asks for images. Do not download or preserve login-page assets from Reel URLs.\n\nFor X status posts, if oEmbed fails, the script falls back to the generic reader path and then automatically tries Browser4 when the result is incomplete. X Articles use the generic reader path directly, then Browser4, then the mandatory in-app Browser handoff when `browser_fallback.required` is true. The same handoff is mandatory for every other public URL that remains incomplete after its configured reader backends. A result with that flag is not terminal: do not classify, move, register, or mark the item resolved until the in-app Browser attempt is recorded. Browser4 or in-app Browser may still return `Blocked` when a page requires login; never enter credentials or infer hidden text. Do not use `Needs Review` as a default failure state.\n\nBrowser4 extraction is intentionally read-only and deterministic: open the public URL in a temporary headless session, extract `document.title`, `document.body.innerText`, and the live body HTML, then close the session. Do not use Browser4 `agent`, `extract`, `summarize`, `swarm`, or login interactions for this fallback.\n\n## Verifying The Claimed Status\n\n`read_url.py` reports `reader_status` from its own extraction. Before handing the result to\nanything that acts on it — registering a Notion page, classifying an inbox item, telling the\nuser a post could be read — read [agents\u002Fextraction-verifier.md](agents\u002Fextraction-verifier.md)\nand call it with the full JSON as `[READER_JSON]`.\n\nWhy a separate agent: the status is self-reported by the same run that produced the payload, so\nnothing in the pipeline currently contradicts it. A verifier reading only the payload catches\n`Extracted` with an empty `markdown`, `ImagesOnly` counting login-page assets, and results still\ncarrying `browser_fallback.required` that are about to be treated as terminal. Downstream\n(`notion-organize-knowledge`) decides registration from this field, and an overstated status\nbecomes an empty page in the knowledge base that nobody notices later.\n\n- `verdict: consistent` → use the result as reported.\n- `verdict: overstated` → treat `actual_status` as the real one. If it drops to `Blocked` or\n  `Failed`, do not register or move the item; leave it for review with the verifier's `evidence`.\n- `fallback_pending: true` → run the in-app Browser protocol before treating the item as\n  terminal, regardless of status.\n\nSkip the verifier only when the result is not being acted on (e.g. the user asked to inspect one\nURL interactively and is reading the output themselves).\n\n## Output To Use Downstream\n\nFor the full JSON contract, read [references\u002Foutput-contract.md](references\u002Foutput-contract.md).\n\nWhen feeding another workflow, keep these fields:\n\n- `input_url`\n- `normalized_url`\n- `reader_backend` (`browser4` is used when the automatic browser fallback supplies the best evidence)\n- `reader_status`\n- `status_reason`\n- `title`\n- `source_url`\n- `author_name`\n- `author_url`\n- `published_at_text`\n- `markdown`\n- `links`\n- `image_links`\n- `downloaded_images`\n- `attempts`\n- `warnings`\n- `browser_fallback` (`required`, `surface`, `canonical_url`, and `reason`, or `null`)\n- `error`\n\nFor Notion knowledge organization, only register the page when `reader_status` is `Extracted`, `Partial`, or `ImagesOnly` with enough local title context. Do not register or move Instagram Reel URLs that resolve to a login wall; leave them in Inbox as `Needs Review`.\n\n## Validation\n\nUse [evals\u002Fevals.json](evals\u002Fevals.json) as the minimum regression set after changing routing, safety checks, or the JSON contract. Always run the script on at least one X post, one X Article, one generic public URL, and one refused private\u002Flocal URL. When Browser4 is installed, also verify a public page whose normal reader is incomplete and confirm that `attempts` contains `browser4`; when it is unavailable, verify that the original result remains usable and the missing dependency is explicit. For any incomplete public URL, verify that `browser_fallback.required` is true and that the calling agent performs the in-app Browser protocol before treating the item as terminal.\n","# URL Reader\n\nUse a domain-aware read-only enrichment layer before summarizing, classifying, or moving URL-only notes. Prefer this skill when a page has too little local context, when Notion only exposes an embed block, or when the user asks whether a social URL can be read.\n\n## Quick Start\n\nRun the bundled script with the original target URL. The script owns backend and Browser4 attempts, but the calling agent owns the in-app Browser handoff:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...'\n```\n\nWhen the normal reader returns anything other than `Extracted`, `read_url.py` automatically tries Browser4 and keeps whichever result contains better evidence. Browser4 is optional: if `browser4-cli` is unavailable, the original reader result is preserved and the failed Browser4 attempt is recorded. If the public URL still remains incomplete, the JSON contains `browser_fallback.required: true`; the calling agent must then execute the in-app Browser final-fallback protocol in [references\u002Fin-app-browser-fallback.md](references\u002Fin-app-browser-fallback.md) without asking the user for permission. This is a fixed retrieval stage, not a decision to ask the user about.\n\nTo enable the fallback, install Browser4 and its self-contained runtime, then make `browser4-cli` available on `PATH`. A non-standard CLI location can be supplied with `BROWSER4_CLI=\u002Fabsolute\u002Fpath\u002Fto\u002Fbrowser4-cli`. The fallback uses a temporary headless profile and deterministic DOM\u002Ftext commands; it does not run Browser4's autonomous agent, enter credentials, or mutate the target site.\n\nFor machine-readable output:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...' --json\n```\n\nTo save extracted image assets:\n\n```bash\npython3 [SKILL_DIR]\u002Fscripts\u002Fread_url.py 'https:\u002F\u002Fwww.instagram.com\u002Fp\u002F...' --json --download-images \u002Ftmp\u002Furl-reader-images\n```\n\n## Rules\n\n- Pass the original URL to the script. Do not manually add reader-service prefixes.\n- Let the script pick the retrieval path from the URL domain. Keep the skill instructions generic so future domain-specific fetchers can be added without changing user prompts.\n- If a URL already contains one or more reader-service prefixes, let the script unwrap them instead of manually editing the URL.\n- Normalize `twitter.com`, `www.twitter.com`, and `mobile.twitter.com` URLs to `x.com` before extraction.\n- Refuse localhost, private IPs, and internal hostnames. Public URL readers should not receive private or machine-local URLs.\n- Treat reader output as extracted public web content, not as an authoritative API response. Keep the original `URL Source` with any summary.\n- Keep failures explicit. Do not infer hidden post text when the reader returns a login page, 403, domain block, empty content, or unrelated navigation.\n\n## Domain Routing\n\nUse these reader paths:\n\n- `x.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>`: fetch `publish.x.com\u002Foembed` first. Use the returned blockquote paragraph as `markdown`, keep `author_name`, `author_url`, and the date-like link text when present, and report `reader_backend: x_oembed`. If that paragraph is only one `t.co` link, derive `x.com\u002F\u003Cuser>\u002Farticle\u002F\u003Cid>` from the original status URL and try the X Article route before treating the post as link-only.\n- `twitter.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>` and mobile\u002Fwww variants: normalize to the equivalent `x.com` URL, then use the X oEmbed path.\n- `twitter.com\u002Fi\u002Fweb\u002Fstatus\u002F\u003Cid>`: normalize to `x.com\u002Fi\u002Fstatus\u002F\u003Cid>` before using X oEmbed.\n- `x.com\u002F\u003Cuser>\u002Farticle\u002F\u003Cid>`: preserve the canonical Article URL as `normalized_url` and `source_url`, and use the generic reader for the Article page. If the result is incomplete, `read_url.py` automatically tries Browser4 against that same canonical Article URL; if it remains incomplete, it sets `browser_fallback.required: true` and the calling agent must use in-app Browser against that exact URL. Do not replace the Article URL with a derived status URL. An opaque `x.com\u002Fi\u002Farticle\u002F\u003Cid>` URL must remain opaque; do not infer an author or a same-ID status URL.\n- Instagram post URLs: use the generic reader path. It can often return the caption, author link, location, hashtags, and signed image URLs.\n- Instagram Reel URLs (`instagram.com\u002Freel\u002F...`): if the reader returns an Instagram login page or generic shell, report `Blocked` and do not treat extracted login assets as usable images.\n- Other public URLs: use the generic reader path first. If the result is `Partial`, `ImagesOnly`, `Blocked`, or `Failed`, `read_url.py` automatically tries Browser4 and then emits the mandatory in-app Browser handoff if the result is still incomplete. Do not omit the handoff because the domain is ordinary.\n- GitHub URLs: run the generic reader first. If it returns `403`, an anonymous-access block, rate limiting, or missing repository content, run `[SKILL_DIR]\u002Fscripts\u002Fread_github_cli.py` when an authenticated `gh` CLI is available before the in-app Browser final fallback. The script uses the CLI's keyring-backed login and never accepts or prints a PAT. If CLI evidence is still incomplete or the CLI is unavailable, execute the required in-app Browser fallback; do not ask the user to choose between them.\n\nX oEmbed is sufficient for post text and author metadata, but it does not reliably expose attached media URLs. If the user asks for X images or video, mark the text extraction separately from media extraction and use another backend or `Needs Review` for media.\n\n## Interpreting Results\n\nUse these extraction statuses:\n\n- `Extracted`: title and meaningful body text are present.\n- `Partial`: title, metadata, image links, or a short caption are present, but the page body is incomplete.\n- `ImagesOnly`: image URLs are present but the textual content is missing or mostly boilerplate.\n- `Blocked`: the selected reader returns an error such as `403`, login wall, domain block, or access denial.\n- `Failed`: network, timeout, invalid URL, or parser failure.\n\nFor Instagram, image URLs may expire, so download them immediately when the user asks for images. Do not download or preserve login-page assets from Reel URLs.\n\nFor X status posts, if oEmbed fails, the script falls back to the generic reader path and then automatically tries Browser4 when the result is incomplete. X Articles use the generic reader path directly, then Browser4, then the mandatory in-app Browser handoff when `browser_fallback.required` is true. The same handoff is mandatory for every other public URL that remains incomplete after its configured reader backends. A result with that flag is not terminal: do not classify, move, register, or mark the item resolved until the in-app Browser attempt is recorded. Browser4 or in-app Browser may still return `Blocked` when a page requires login; never enter credentials or infer hidden text. Do not use `Needs Review` as a default failure state.\n\nBrowser4 extraction is intentionally read-only and deterministic: open the public URL in a temporary headless session, extract `document.title`, `document.body.innerText`, and the live body HTML, then close the session. Do not use Browser4 `agent`, `extract`, `summarize`, `swarm`, or login interactions for this fallback.\n\n## Verifying The Claimed Status\n\n`read_url.py` reports `reader_status` from its own extraction. Before handing the result to\nanything that acts on it — registering a Notion page, classifying an inbox item, telling the\nuser a post could be read — read [agents\u002Fextraction-verifier.md](agents\u002Fextraction-verifier.md)\nand call it with the full JSON as `[READER_JSON]`.\n\nWhy a separate agent: the status is self-reported by the same run that produced the payload, so\nnothing in the pipeline currently contradicts it. A verifier reading only the payload catches\n`Extracted` with an empty `markdown`, `ImagesOnly` counting login-page assets, and results still\ncarrying `browser_fallback.required` that are about to be treated as terminal. Downstream\n(`notion-organize-knowledge`) decides registration from this field, and an overstated status\nbecomes an empty page in the knowledge base that nobody notices later.\n\n- `verdict: consistent` → use the result as reported.\n- `verdict: overstated` → treat `actual_status` as the real one. If it drops to `Blocked` or\n  `Failed`, do not register or move the item; leave it for review with the verifier's `evidence`.\n- `fallback_pending: true` → run the in-app Browser protocol before treating the item as\n  terminal, regardless of status.\n\nSkip the verifier only when the result is not being acted on (e.g. the user asked to inspect one\nURL interactively and is reading the output themselves).\n\n## Output To Use Downstream\n\nFor the full JSON contract, read [references\u002Foutput-contract.md](references\u002Foutput-contract.md).\n\nWhen feeding another workflow, keep these fields:\n\n- `input_url`\n- `normalized_url`\n- `reader_backend` (`browser4` is used when the automatic browser fallback supplies the best evidence)\n- `reader_status`\n- `status_reason`\n- `title`\n- `source_url`\n- `author_name`\n- `author_url`\n- `published_at_text`\n- `markdown`\n- `links`\n- `image_links`\n- `downloaded_images`\n- `attempts`\n- `warnings`\n- `browser_fallback` (`required`, `surface`, `canonical_url`, and `reason`, or `null`)\n- `error`\n\nFor Notion knowledge organization, only register the page when `reader_status` is `Extracted`, `Partial`, or `ImagesOnly` with enough local title context. Do not register or move Instagram Reel URLs that resolve to a login wall; leave them in Inbox as `Needs Review`.\n\n## Validation\n\nUse [evals\u002Fevals.json](evals\u002Fevals.json) as the minimum regression set after changing routing, safety checks, or the JSON contract. Always run the script on at least one X post, one X Article, one generic public URL, and one refused private\u002Flocal URL. When Browser4 is installed, also verify a public page whose normal reader is incomplete and confirm that `attempts` contains `browser4`; when it is unavailable, verify that the original result remains usable and the missing dependency is explicit. For any incomplete public URL, verify that `browser_fallback.required` is true and that the calling agent performs the in-app Browser protocol before treating the item as terminal.\n",{"name":5,"description":7},"main","75d5565f1421b36c81aab66ed19bf2ccf167a777","plugins\u002Fresearch\u002Fskills\u002Furl-reader",0,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins","https:\u002F\u002Fgithub.com\u002Fyoshiysh","2026-11-14 23:07:31",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add yoshiysh\u002Fclaude-plugins --skill url-reader",{"tool":36,"command":37},"bunx","bunx skills add yoshiysh\u002Fclaude-plugins --skill url-reader",{"tool":39,"command":40},"pnpm","pnpm dlx skills add yoshiysh\u002Fclaude-plugins --skill url-reader","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002FSKILL.md",true,[44,50,54,58,62,66,72,76],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"agents\u002Fextraction-verifier.md","REFERENCE","参考文档",4024,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Fagents\u002Fextraction-verifier.md",{"path":51,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":52,"sourceHtmlUrl":53},"agents\u002Fopenai.yaml",313,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Fagents\u002Fopenai.yaml",{"path":55,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":56,"sourceHtmlUrl":57},"evals\u002Fevals.json",4313,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Fevals\u002Fevals.json",{"path":59,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":60,"sourceHtmlUrl":61},"references\u002Fin-app-browser-fallback.md",4109,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Freferences\u002Fin-app-browser-fallback.md",{"path":63,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":64,"sourceHtmlUrl":65},"references\u002Foutput-contract.md",3106,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Freferences\u002Foutput-contract.md",{"path":67,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":70,"sourceHtmlUrl":71},"scripts\u002Fread_github_cli.py","SCRIPT","脚本",2381,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Fscripts\u002Fread_github_cli.py",{"path":73,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":74,"sourceHtmlUrl":75},"scripts\u002Fread_url.py",46845,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Fscripts\u002Fread_url.py",{"path":77,"resourceType":68,"resourceTypeDesc":69,"sizeBytes":78,"sourceHtmlUrl":79},"tests\u002Ftest_read_url.py",20888,"https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fresearch\u002Fskills\u002Furl-reader\u002Ftests\u002Ftest_read_url.py",[81,88,95,102,109,116],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":18},"01M7DZ6802JNP0SPM5PVQCN9EX","coverage-build","文書でもプログラムでも、作る（create）・既にあるものを目標に合わせて直す（update）・作らずに目標に照らして確かめる（review）作業を、 目標の固定 →（update は現状の検査）→ 設計 → 設計レビュー → 実装 → 実装レビューの段で回し、見落としを構造で減らす汎用の仕組み。 1 体には 1 単位だけを任せ、テストは実装と別の担当が先に書き、レビューは coverage-audit を呼び、悪かった出来事を探索グラフで振り返って 欠陥を原因の層（目標・設計・実装）へ戻す（指摘の箇所だけを直す経路は持たない）。台帳は run をまたいで持ち越し、変わったものだけを見直す。 「この機能を設計からレビューまで通して作って」「このスキルを目標に合わせて直して（文書とコードが混ざってよい）」 「この実装が目標を満たすか確かめて、直すならどの層からかを返して」のような依頼で使う （Use when building or updating documents or code through goal, design, design review, implementation, and implementation review with layer-routed rework）。 対象外: 目標に照らした網羅的な検査だけで振り返りも要らないもの（coverage-audit）、1 体で 1 回で済む小さな変更、このリポジトリの差分・PR のレビュー（repo-review）。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fcoverage-build\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":18},"01M7DZ609XBH3H9Y616XXZC2DA","coverage-audit","与えられた目的と対象（設計・計画・要求・仕様・コード・テストなどのファイル）を、資産 × 観点の枡に分けて網羅的に検査し、 確かめた指摘と、見ていない所（開示した枡）を台帳に残して返す検査専用の仕組み。1 体には 1 枡だけを渡し、 読んだ証拠を本文の区画ごとの逐語の引用で機械が照らし、読む役と別のモデル・反証側の検証役が指摘を確かめ、 悪かった出来事を振り返り、確かめた欠陥を根本原因で束ねて型を全ての資産へ横展開する。台帳は run をまたいで持ち越し、変わった枡だけを読み直す。 「修正の前にこの設計・計画を観点ごとに漏れなく検査して」「この実装が目的と仕様を満たすか網羅的に検証して」 「前回の台帳の続きから、変わった所だけ検査し直して」のような依頼で使う（Use when files must be audited for every part × every lens with machine-checked evidence, without fixing them）。 対象外: 修正・生成の実行（このスキルは指摘と網羅の記録までを返し、直すのは呼ぶ側）、1 体で 1 回で済む小さな確認、このリポジトリの差分・PR のレビュー（repo-review）。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fcoverage-audit\u002FSKILL.md",[],[],{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":18},"01M787A9W12AT2F4FB0DMCP3MT","summarizing-meeting-notes","会議後に書かれた議事録・会議メモ・文字起こしテキストと、会議の録音ファイル（.mp3 \u002F .m4a）を読み、 決定事項・アクションアイテム・未決事項を区分した要約を作る。 「この議事録を要約して」「会議メモから宿題を抜き出して」「定例の決定事項をまとめて」 「先週の打ち合わせの要点を出して」のように、 会議の記録から決定・宿題・未決を知りたい依頼なら使うこと。 記録が箇条書き数行のメモでも使うこと。 対象外: 会議中のリアルタイムな記録取り、議事録以外の文書（企画書・報告書）の要約、 1on1・面談の記録の要約。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-creator-accuracy\u002Ftarget\u002Fsummarizing-meeting-notes\u002FSKILL.md",[],[],{"skillNo":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":18},"01M787A5ERGNSYWJP6EX6CZKZ9","release-notes-editor","リリースノートの下書きを、利用者と 1 項目ずつ確かめながら書く。「次のリリースのリリースノートを書いて」「リリースノートを作って」という依頼で使う。リリースノートの翻訳と、コミットメッセージの書き直しには使わない。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fbp-trial\u002Fupdate-target\u002FSKILL.md",[],[],{"skillNo":110,"name":111,"nameZh":6,"description":112,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":113,"categories":114,"keywords":115,"publishedAt":18},"01M787A2CA9B4SW809PKK202KK","trace-cause","リポジトリで再現する 1 つの失敗（落ちるテスト・誤った出力）の根本原因を、確かめた事実の連なりとして特定する。「このテストが落ちる原因を突き止めて」「なぜこのコマンドがこの出力になるのか調べて」のように、再現手順のある失敗の原因調査を頼まれたときに使う。再現手順の無い相談、修正そのものの依頼、一般的な質問には使わない。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fbp-trial\u002Freview-target\u002FSKILL.md",[],[],{"skillNo":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":120,"categories":121,"keywords":122,"publishedAt":18},"01M7879YT3PR92S3YX137ZVB02","glossary-sync","リポジトリの用語集 docs\u002Fglossary.md と、文書の中の用語の使い方の食い違いを洗い出す。「用語集と文書の食い違いを調べて」という依頼で使う。用語集の新規作成と翻訳には使わない。","https:\u002F\u002Fgithub.com\u002Fyoshiysh\u002Fclaude-plugins\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fbp-trial\u002Faudit-target\u002FSKILL.md",[],[]]