跳转到正文
报告库
用途分类 / 其他用途

Is This Photo Real Skill 安全审计

作者说它能做什么(原文)

>-

第三方安全检查结论

先别安装或运行

已检查文件
3
发现的风险
4
会不会运行危险命令?检查是否下载程序后直接运行、让他人远程控制电脑,或藏起要运行的命令。未发现风险
会不会泄露文件和密钥?检查是否发送含密码或密钥的文件,以及代码里是否直接写了密钥。发现 2 项风险
高风险

原图、裁剪和视频关键帧可能被披露给多个搜索或取证服务

原文依据:4 处
发现了什么

流程要求在至少三个反向图片搜索引擎中提交完整画面,并继续提交裁剪、镜像和视频关键帧。参考资料也说明某些在线取证服务会保留或公开上传内容。虽然 Skill 提醒用户检查发布政策,但没有在敏感材料流程中强制采用本地处理或取得当事人同意。

为什么需要注意

KYC 证件、人脸、事故、犯罪、受害者或私密影像可能被第三方记录、用于服务处理,甚至出现在公开图库;一旦上传通常无法可靠撤回。

该流程要求把整图、多个视频关键帧、裁剪和镜像版本提交给多个反向搜索引擎;在线提交会把敏感影像披露给相应服务。资料明确承认部分取证服务可能公开上传内容,并建议敏感文件优先使用本地工具。风险只在实际使用这些在线服务时发生,且 Skill 已给出警告,但没有把本地处理设为强制条件。用户可限制为本地提帧与分析,并在上传前确认保留、公开和删除政策。

reference/verification-checklist.md:26来自说明文档打开原文件
- [ ] Reverse image search: full frame across at least three engines. For video, extract      keyframes and search several.- [ ] Crop the distinctive elements and re-search each. Mirror and re-search.- [ ] Sort for the **oldest** copy where the engine allows.- [ ] Open every match page. Harvest: caption, date, photographer credit, agency, named      people, image filename.- [ ] Corroborate the earliest page's date against an independent archive capture.- [ ] Search the caption text and any distinctive on-image text as a string, in the      original language.
查看另外 3 个位置
reference/tool-catalogue.md:7来自说明文档打开原文件
Before anything: check whether a tool **publishes** your submission. Some forensicservices maintain public galleries of uploads. For anything sensitive, prefer localtools.
reference/tool-catalogue.md:89来自说明文档打开原文件
### FotoForensicsHosted suite built around ELA, plus JPEG quality estimation, last-save quality,metadata, embedded thumbnail extraction, and hidden-pixel and string checks. Thenon-ELA parts are the useful parts.**Failure mode:** ELA is the headline feature and the most misread analysis in thefield — see the parent skill. The service also retains and may display submissions;do not upload sensitive material.
SKILL.md:252来自说明文档打开原文件
Publishing an accusation of fabrication against a named person carries defamationrisk in most jurisdictions, and "our tool said so" is not a defence. Uploadingmaterial to online forensic services discloses it to those services and, for some, topublic galleries — check whether a tool publishes submissions before submittinganything sensitive. If the media depicts a crime, a victim, or intimate content:minimise copies, do not redistribute, and in the case of child sexual abuse materialstop immediately and report to the appropriate authority rather than analysing it. See[../../ETHICS.md](../../ETHICS.md).
中风险

照片核验可被扩展为对个人、账号和社交活动的广泛调查

原文依据:2 处
发现了什么

“Pivots”部分把账号传播分析、社交活动模式、泄漏搜索以及查找照片中的具名个人列为后续路径。这些任务超出了判断媒体真伪所必需的范围,且提供的材料没有要求取得同意、证明必要性或限制收集范围。

为什么需要注意

执行这些扩展步骤可能汇总个人身份、账号关系、活动模式或泄漏数据,对普通拍摄对象、记者、受害者或其他无关人员造成隐私、骚扰或错误关联风险。

“Pivots”明确把媒体核验引向账号追踪、社交活动模式、泄漏搜索、网络关系图和查找具名个人。这些是可选的后续路径,并非自动执行;在虚假信息传播调查中也可能有正当用途。但若用户或代理继续采用这些路径,会扩大个人及关联账号的数据收集范围,而该表本身没有限定必要性、同意或最小化范围。用户可将授权限定为媒体来源核验,并要求任何个人或社交网络调查另行说明目的、范围和合法依据。

SKILL.md:245来自说明文档打开原文件
| Deleted or altered source pages | `read-deleted-pages` || Publishing or seeding domain | `who-owns-this-domain`, `recon-a-domain-passively` || Accounts amplifying the media | `hunt-a-handle`, `pattern-of-life-from-socials` || Coordinated network behind the spread | `graph-the-network`, `find-leaks-in-the-wild` || Named individuals in or credited on the media | `find-anyone` |
查看另外 1 个位置
SKILL.md:239来自说明文档打开原文件
| What you got | Send to ||---|---|| Earlier copies, credits, original caption | `find-the-original-image` || Editing chain, device, timestamps | `secrets-in-file-metadata` || Location and date verification | `geolocate-from-pixels`, `where-was-this-taken` || Deleted or altered source pages | `read-deleted-pages` || Publishing or seeding domain | `who-owns-this-domain`, `recon-a-domain-passively` || Accounts amplifying the media | `hunt-a-handle`, `pattern-of-life-from-socials` || Coordinated network behind the spread | `graph-the-network`, `find-leaks-in-the-wild` || Named individuals in or credited on the media | `find-anyone` |
会不会删除文件或一直在后台运行?检查是否大范围删除文件、改写磁盘,或设置自动启动。未发现风险
会不会绕过安全保护?检查是否跳过网站安全验证、开放过多文件权限,或取消操作前的确认。未发现风险
会不会误导 AI 或隐藏内容?检查工作说明是否要求 AI 忽略你的指令、干扰检查结果,或夹带看不见的文字。发现 1 项风险
中风险

要求打开所有匹配网页,但未隔离网页中的恶意指令

原文依据:4 处
发现了什么

核验流程要求代理打开每个搜索结果页面、读取文字并调用网页档案。外部网页由不受信任的发布者控制,可能包含面向 AI 代理的提示注入;提供的说明没有要求忽略网页指令、限制工具调用或只提取指定证据字段。

为什么需要注意

如果具备浏览器、文件或账户工具的代理把网页文字当作操作命令,网页可能影响结论,或诱导代理执行与媒体核验无关的读取、上传或账户操作。

这段证据能说明什么

资料确实要求打开所有匹配页面和查阅网页档案,因此会接触不受信任的网页内容;但可见内容只要求提取标题、日期、署名等证据,没有指示代理服从网页中的命令,也没有提供会执行网页文本的脚本。因此提示注入后果在现有证据中无法确认。用户可要求浏览步骤采用只读模式,仅提取列明字段,并禁止网页内容触发工具调用或权限变更。

这项判断针对展示的代码和适用条件,不表示风险已经实际发生。
reference/verification-checklist.md:26来自说明文档打开原文件
- [ ] Reverse image search: full frame across at least three engines. For video, extract      keyframes and search several.- [ ] Crop the distinctive elements and re-search each. Mirror and re-search.- [ ] Sort for the **oldest** copy where the engine allows.- [ ] Open every match page. Harvest: caption, date, photographer credit, agency, named      people, image filename.- [ ] Corroborate the earliest page's date against an independent archive capture.- [ ] Search the caption text and any distinctive on-image text as a string, in the      original language.
查看另外 3 个位置
SKILL.md:47来自说明文档打开原文件
   downstream conclusion.2. **Provenance.** Run `find-the-original-image`; for video, extract and search   keyframes. You want an earlier appearance, a different caption, a photographer   credit, and an on-page date corroborated through `read-deleted-pages`. An earlier copy   with a different caption ends the case.3. **Metadata.** Run `secrets-in-file-metadata`: editing chain, thumbnail-versus-image
reference/verification-checklist.md:30来自说明文档打开原文件
- [ ] Sort for the **oldest** copy where the engine allows.- [ ] Open every match page. Harvest: caption, date, photographer credit, agency, named      people, image filename.- [ ] Corroborate the earliest page's date against an independent archive capture.- [ ] Search the caption text and any distinctive on-image text as a string, in the      original language.
reference/tool-catalogue.md:22来自说明文档打开原文件
### Web archivesArchived captures date a page independently of what the page claims about itself. See`read-deleted-pages`.**Failure mode:** no capture means no evidence either way; robots-directiveretroactivity and takedowns can remove captures that existed.
会不会偷偷改推广链接或收款方?检查是否强制替换推广链接或收款对象,同时要求隐瞒更改。发现 1 项风险
中风险

带主观判断的核验结论被定位用于 KYC、保险和证据决定

原文依据:7 处
发现了什么

Skill 明确将 KYC、入职审核、保险理赔和证据可采性列为用途,同时允许基于物理异常、生成痕迹和缺少来源给出“可能被操纵”或“可能为合成”的分级。这些步骤包含人工判断和不完整证据,并不等同于身份欺诈、理赔造假或法律可采性的证明。

为什么需要注意

若结论被直接用于自动或单人决策,真实用户可能被拒绝开户、入职或理赔;也可能对被拍摄者形成错误指控和声誉或法律损害。

Skill 明确面向 KYC、入职欺诈、保险理赔和证据可采性等高影响场景,并允许依据可解释但仍含人工判断的迹象给出“可能被操纵”或“可能为合成”的结论。这些结论若被当作拒绝身份、理赔或法律证据的充分依据,可能造成重大误判。资料同时要求逐项分级、陈述限制且避免只说“真假”,这降低但不消除风险。用户可要求这些结果仅作调查线索,并由具备相应权限的人员结合独立证据复核。

SKILL.md:4来自说明文档打开原文件
description: >-  Verify whether an image or video is authentic, original and correctly captioned — provenance  checks, error level analysis, noise and JPEG compression analysis, clone and copy-move  detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and  AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a  photo or video, checking for a deepfake or AI-generated image, spotting manipulation, or  testing whether footage is recycled or miscaptioned. Applies to KYC and onboarding fraud,  insurance claim review, disinformation analysis, and evidence admissibility. Reference at  useosint.com/skills/is-this-photo-real.
查看另外 6 个位置
SKILL.md:182来自说明文档打开原文件
  `investigate-without-getting-made`.- **Identification from resemblance** is the highest-consequence error here — "this is  person Z because they look alike" is not a finding.
SKILL.md:193来自说明文档打开原文件
  details consistent, nothing contradicted in metadata or geometry.- **Confirmed recontextualised** — the same image demonstrably published earlier with  a different, better-sourced caption. The most common positive finding here.- **Probably manipulated** — a specific, describable physical or geometric  inconsistency you can point at, ideally with a source for the inserted element.  Not a heatmap.- **Probably synthetic** — multiple durable generative tells, no provenance history,  no camera-consistent compression or metadata, no earlier copies. Name the tells.- **Unconfirmed** — no earlier copy found, nothing wrong found. Where most cases end.- **Cannot be assessed** — the available copy is too processed for the tests the question
SKILL.md:252来自说明文档打开原文件
Publishing an accusation of fabrication against a named person carries defamationrisk in most jurisdictions, and "our tool said so" is not a defence. Uploadingmaterial to online forensic services discloses it to those services and, for some, to
SKILL.md:7来自说明文档打开原文件
  detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and  AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a  photo or video, checking for a deepfake or AI-generated image, spotting manipulation, or  testing whether footage is recycled or miscaptioned. Applies to KYC and onboarding fraud,  insurance claim review, disinformation analysis, and evidence admissibility. Reference at  useosint.com/skills/is-this-photo-real.
SKILL.md:195来自说明文档打开原文件
  a different, better-sourced caption. The most common positive finding here.- **Probably manipulated** — a specific, describable physical or geometric  inconsistency you can point at, ideally with a source for the inserted element.  Not a heatmap.- **Probably synthetic** — multiple durable generative tells, no provenance history,  no camera-consistent compression or metadata, no earlier copies. Name the tells.- **Unconfirmed** — no earlier copy found, nothing wrong found. Where most cases end.- **Cannot be assessed** — the available copy is too processed for the tests the question  needs. Say which tests were invalid and why.
SKILL.md:228来自说明文档打开原文件
**Reporting standard.** State what you verified, what you could not, and what eachconclusion rests on — never a bare "fake" or "real". A defensible line reads: *theimage was published at least three years before the claimed event, on a news site,credited to a named photographer; the location matches that credit and not the claim; nomanipulation was detected, but the only available copy was a platform re-encode, sosignal-level tests were not meaningful.* "Our analysis shows this is fake" is not.Include the tests you ran and their negative results, the file's processing history, and

Skill 逻辑拆解

8 个说明模块

该 Skill 是一套媒体核验流程,要求先确认原始来源和说明文字,再检查元数据、地点、物理一致性,最后才使用容易误判的像素级取证。

查看原文
SKILL.md:17来自说明文档打开原文件
Verification order is the whole skill: **provenance first, pixels last.** Finding theearliest copy and reading its caption settles more cases than every forensic filtercombined, and it produces evidence you can show someone. Pixel forensics produces acolourful heatmap and an argument.
SKILL.md:43来自说明文档打开原文件
1. **Get the best copy.** Every re-encode, resize and screenshot destroys forensic   signal. Chase the original upload or the agency version, not the platform rendition.   Hash it, work on copies. If all you have is a screenshot, say so and lower every   downstream conclusion.2. **Provenance.** Run `find-the-original-image`; for video, extract and search   keyframes. You want an earlier appearance, a different caption, a photographer   credit, and an on-page date corroborated through `read-deleted-pages`. An earlier copy   with a different caption ends the case.3. **Metadata.** Run `secrets-in-file-metadata`: editing chain, thumbnail-versus-image   comparison, timestamp inconsistencies, whether MakerNotes fit the claimed device.4. **Provenance signing.** Check for C2PA Content Credentials.5. **Internal consistency.** Signage language, plates, currency, uniforms, vehicle   models, season, weather and shadows against the claimed date and place, via   `geolocate-from-pixels`. Ordinary detective work, more productive than forensics.6. **Physical consistency.** Lighting, shadows, reflections, perspective.7. **Signal-level forensics.** Noise residuals, JPEG quantisation and double compression,   clone detection, colour-filter-array traces. Easy to over-read, worthless on a   platform-processed file.8. **Write what you verified**, not a verdict.

提供的内容是 Markdown 操作说明和参考资料;其中列出了本地媒体检查命令,但没有提供可核验的自动执行脚本或安装步骤。

查看原文
reference/tool-catalogue.md:59来自说明文档打开原文件
### ffprobe and MediaInfoContainer and stream inspection for video: encoder strings, frame rate, rotationmatrices, per-track metadata, frame types and timestamps.```bashffprobe -v error -show_format -show_streams video.mp4ffprobe -v error -select_streams v -show_frames -show_entries frame=pict_type,pkt_pts_time video.mp4mediainfo video.mp4```
reference/verification-checklist.md:16来自说明文档打开原文件
      things specifically later.- [ ] Obtain the best available copy. Original file over platform rendition, platform      rendition over screenshot. For platform video use `yt-dlp` rather than a screen      recording.- [ ] Hash it. `sha256sum` the file, record the hash, and work only on copies.- [ ] Record the **processing history you know**: where you got it, whether it was      re-encoded, whether it is a crop or screenshot. This determines which tiers below      are valid at all.

该 Skill 明确认识到检测器、元数据和 C2PA 的证据边界,并要求分别陈述已验证、未验证和无法评估的事项,而不是仅输出“真/假”。

查看原文
SKILL.md:103来自说明文档打开原文件
C2PA binds a cryptographically signed manifest to a file recording capture and edithistory. Where it exists it is the strongest provenance evidence available, because itis verifiable rather than inferential. A **valid** manifest means the signer assertsthis history, the file is unchanged since signing, and you know who to holdresponsible — not that the content is true. A signed photograph of a staged scene is asigned photograph.**Absence** means almost nothing: most cameras do not sign, most editing pipelines donot preserve manifests, and platforms strip them during re-encoding. Missingcredentials are the default state, not a red flag. Same for the IPTCdigital-source-type field used to label synthetic media, and for model-specificinvisible watermarks — a positive is strong where you can check it, a negative onlysays one vendor's mark was not found.
SKILL.md:228来自说明文档打开原文件
**Reporting standard.** State what you verified, what you could not, and what eachconclusion rests on — never a bare "fake" or "real". A defensible line reads: *theimage was published at least three years before the claimed event, on a news site,credited to a named photographer; the location matches that credit and not the claim; nomanipulation was detected, but the only available copy was a platform re-encode, sosignal-level tests were not meaningful.* "Our analysis shows this is fake" is not.Include the tests you ran and their negative results, the file's processing history, andyour assumptions, then hand the package to `write-the-intel-brief`.

核心流程依赖多个未包含在本次材料中的其他 Skill,例如原图搜索、元数据分析、地理定位和已删除页面读取;因此无法从所提供源码核验这些步骤的网络、数据保留或权限行为。

查看原文
SKILL.md:47来自说明文档打开原文件
   downstream conclusion.2. **Provenance.** Run `find-the-original-image`; for video, extract and search   keyframes. You want an earlier appearance, a different caption, a photographer   credit, and an on-page date corroborated through `read-deleted-pages`. An earlier copy   with a different caption ends the case.3. **Metadata.** Run `secrets-in-file-metadata`: editing chain, thumbnail-versus-image   comparison, timestamp inconsistencies, whether MakerNotes fit the claimed device.4. **Provenance signing.** Check for C2PA Content Credentials.5. **Internal consistency.** Signage language, plates, currency, uniforms, vehicle   models, season, weather and shadows against the claimed date and place, via   `geolocate-from-pixels`. Ordinary detective work, more productive than forensics.6. **Physical consistency.** Lighting, shadows, reflections, perspective.
SKILL.md:239来自说明文档打开原文件
| What you got | Send to ||---|---|| Earlier copies, credits, original caption | `find-the-original-image` || Editing chain, device, timestamps | `secrets-in-file-metadata` || Location and date verification | `geolocate-from-pixels`, `where-was-this-taken` || Deleted or altered source pages | `read-deleted-pages` || Publishing or seeding domain | `who-owns-this-domain`, `recon-a-domain-passively` || Accounts amplifying the media | `hunt-a-handle`, `pattern-of-life-from-socials` || Coordinated network behind the spread | `graph-the-network`, `find-leaks-in-the-wild` || Named individuals in or credited on the media | `find-anyone` |
从这里开始 · 工作说明SKILL.md
is-this-photo-real
连线表示工作说明包含的模块,不是实际运行顺序。点击模块可查看原文。 另有 4 个章节,可在原文件中查看。

文件引用关系图

2 处引用
哪些文件发起引用引用了什么
连线表示真实的文件引用,不是运行顺序。点击节点可高亮相关连线,并查看具体文件和原文位置。虚线表示还有文件需要定位。
文件与检查记录3 个文件

检查范围与遗漏

逐文件查看涉及的内容

下方列出本次涉及的原文范围;纳入检查不代表已查清所有问题。

  • SKILL.md已纳入全文
  • reference/tool-catalogue.md已纳入全文
  • reference/verification-checklist.md已纳入全文

这份报告只针对上方版本。我们看了拿到的代码和说明文件,没有实际运行 Skill,也没有检查它另外安装的软件包。因此,这不是“保证安全”的承诺;换了版本或使用环境,结果也可能不同。

  • SKILL.md工作说明
  • reference/tool-catalogue.md配套文件
  • reference/verification-checklist.md配套文件

代码和说明中提到的操作

读取密钥或账号配置
SKILL.md:6来自说明文档打开原文件
  checks, error level analysis, noise and JPEG compression analysis, clone and copy-move  detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and  AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a
SKILL.md:53来自说明文档打开原文件
   comparison, timestamp inconsistencies, whether MakerNotes fit the claimed device.4. **Provenance signing.** Check for C2PA Content Credentials.5. **Internal consistency.** Signage language, plates, currency, uniforms, vehicle
SKILL.md:101来自说明文档打开原文件
## C2PA and Content Credentials
运行命令
reference/tool-catalogue.md:64来自说明文档打开原文件
```bashffprobe -v error -show_format -show_streams video.mp4
读取了多少行
677
文件校验值(用于核对版本)
90f1bb32280b52030d1938de41bda7f15dd15ee2613ce10b404ff4b654f8df3d