Skip to content

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

fix(social-publish): harden media constraint verification

Barack Sokullua4a4cc809429
diff --git a/src/chrome/src/agent/agent.js b/src/chrome/src/agent/agent.js
index aeb060a22..7007eca17 100644
--- a/src/chrome/src/agent/agent.js
+++ b/src/chrome/src/agent/agent.js
@@ -261,7 +261,7 @@ const SOCIAL_POST_DESTINATION_NEGATION = new RegExp(
'iu',
);
const SOCIAL_CONTRASTIVE_EXCLUSION = /(?<![\p{L}\p{N}_])(?:rather\s+than|instead\s+of)(?:\s+(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?))?(?:\s+(?:on|onto|to|via|in|at))?\s*$/iu;
-const SOCIAL_DESTINATION_EXCLUSION = /^\s*(?:exclude|excluding)\s+(?:(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?)\s+)?(?:(?:on|onto|to|via|in|at)\s+)?(?:the\s+)?(?:x|twitter|bluesky|bsky(?:\.app)?)(?![\p{L}\p{N}_.-])\s*$/iu;
+const SOCIAL_DESTINATION_EXCLUSION = /^\s*(?:exclude|excluding|no)\s+(?:(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?)\s+)?(?:(?:on|onto|to|via|in|at)\s+)?(?:the\s+)?(?:x|twitter|bluesky|bsky(?:\.app)?)(?![\p{L}\p{N}_.-])\s*$/iu;
const socialPostNegationGovernsPublish = value => (
SOCIAL_POST_NEGATION.test(String(value || '').trim())
|| SOCIAL_POST_DESTINATION_NEGATION.test(String(value || ''))
@@ -356,6 +356,9 @@ const GENERIC_ATTACHMENT_WORDS = new Set([
const RAW_GIF_NOUN_REGEX = /\b(?:gif|gifs)\b|\.gif(?:[?#]|$)|(?:animated[-_ ]gif|動圖|动图|움짤)/i;
const RAW_VIDEO_NOUN_REGEX = /\b(?:video|videos|mp4|mov|webm|mkv|clip|clips|recording|recordings|vidéo|vidéos)\b|(?:動画|视频|影片|видео|동영상|비디오|영상)/i;
const RAW_IMAGE_NOUN_REGEX = /\b(?:image|images|photo|photos|picture|pictures|pic|pics|png|jpg|jpeg|webp|foto|fotos|bild|bilder|imagen(?:es)?|imágenes|imagem|imagens)\b|(?:画像|写真|图片|照片|圖片|изображение|фото|사진|이미지|포토)/i;
+const IMAGE_ATTACHMENT_FORMAT = '(?:png|jpe?g|webp|avif|heic|bmp|svg)';
+const VIDEO_ATTACHMENT_FORMAT = '(?:mp4|mov|webm|mkv)';
+
// A minimum qualifier binds the count phrase it sits in, not the whole
// requirement: "at least two images and one video" still wants exactly one
// video. These are the boundaries between those phrases.
@@ -508,6 +511,13 @@ const GIF_NEGATION_REGEX = new RegExp(
'iu',
);
+const normalizeAttachmentNegationArticles = value => String(value || '').replace(
+ new RegExp(`(?<![${SOCIAL_WORD_EDGE}])(no|not|without)(\\s+)(a|an|the)(\\s+)`, 'giu'),
+ (_match, negator, beforeArticle, article, afterArticle) => (
+ negator + beforeArticle + ' '.repeat(article.length) + afterArticle
+ ),
+);
+
const VISION_SUB_CALL_TIMEOUT_MS = 90_000;
const CONTENT_ACTION_TIMEOUT_MS = 60_000;
const CONTENT_ACTION_RESPONSE_GRACE_MS = 5_000;
@@ -3039,11 +3049,12 @@ export class Agent extends LoopDetector {
_isNegativeAttachmentRequirement(text) {
const s = String(text || '').trim().toLowerCase();
if (!s) return false;
+ const negationText = normalizeAttachmentNegationArticles(s);
if (/^(?:none|no|false|0|zero|なし|無し|無|없음)$/i.test(s)) return true;
const negPrefix = '(?:no|without|without\\s+any|sin|sans|sem|senza|ohne|без|kein|keine|aucun|aucune|ningun|ningún|ninguna|nenhum|nenhuma|nessun|nessuno|nessuna|nie)';
const nounSuffix = '(?:attachments?|files?|m[eéèê]dias?|m[ií]dia|medien|medios?|uploads?|images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|fichiers?|pièces?|archivos?|adjuntos?|anexos?|allegat[oi]?|anhänge?|anhang|dateien?|вложений|вложения|фото(?:графий)?|изображений|видео|файлов)';
const nounListSeparator = '(?:\\s*[,;]\\s*(?:(?:and|or|nor)\\s+)?|\\s+(?:and|or|nor)\\s+)';
- if (new RegExp(`^${negPrefix}\\s+(?:any\\s+)?${nounSuffix}(?:${nounListSeparator}(?:any\\s+)?${nounSuffix})*$`, 'i').test(s)) return true;
+ if (new RegExp(`^${negPrefix}\\s+(?:any\\s+)?${nounSuffix}(?:${nounListSeparator}(?:any\\s+)?${nounSuffix})*$`, 'i').test(negationText)) return true;
if (new RegExp(`^(?:0|zero)\\s*${nounSuffix}?$`, 'i').test(s)) return true;
if (/^(?:(?:无|没有|不带|零个|0个)\s*(?:附件|图片|照片|视频|媒体|文件)|(?:附件|图片|照片|视频|媒体|文件)\s*(?:无|没有|为0|为零))$/i.test(s)) return true;
if (/^(?:(?:添付|メディア|画像|写真|動画|ファイル)\s*(?:なし|無し|ゼロ|0)|(?:なし|無し)\s*(?:添付|メディア|画像|写真|動画|ファイル))$/i.test(s)) return true;
@@ -3076,7 +3087,8 @@ export class Agent extends LoopDetector {
}
const coordinatedMediaNoun = '(?:images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|animated[-_ ]gifs?)';
- const coordinatedMediaListNegation = text.match(new RegExp(
+ const negationText = normalizeAttachmentNegationArticles(text);
+ const coordinatedMediaListNegation = negationText.match(new RegExp(
`(?<![${SOCIAL_WORD_EDGE}])(?:no|not|without(?:\\s+any)?)\\s+(?:any\\s+)?(${coordinatedMediaNoun}(?:(?:\\s*[,;]\\s*(?:(?:and|or|nor)\\s+)?|\\s+(?:and|or|nor)\\s+)(?:any\\s+)?${coordinatedMediaNoun})+)`,
'iu',
));
@@ -3084,11 +3096,11 @@ export class Agent extends LoopDetector {
const coordinatedMediaNegationSpan = coordinatedMediaListNegation
? [coordinatedMediaListNegation.index || 0, (coordinatedMediaListNegation.index || 0) + coordinatedMediaListNegation[0].length]
: null;
- const isImageNegated = IMAGE_NEGATION_REGEX.test(text)
+ const isImageNegated = IMAGE_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_IMAGE_NOUN_REGEX.test(coordinatedNegatedMedia));
- const isVideoNegated = VIDEO_NEGATION_REGEX.test(text)
+ const isVideoNegated = VIDEO_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_VIDEO_NOUN_REGEX.test(coordinatedNegatedMedia));
- const isGifNegated = GIF_NEGATION_REGEX.test(text)
+ const isGifNegated = GIF_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_GIF_NOUN_REGEX.test(coordinatedNegatedMedia));
if (isImageNegated && isVideoNegated) {
@@ -3117,6 +3129,17 @@ export class Agent extends LoopDetector {
const wantsOrdinaryVideo = hasRawOrdinaryVideo && !isVideoNegated;
const wantsVideo = wantsOrdinaryVideo || wantsGif;
const wantsImage = hasRawImage && !isImageNegated;
+ const normalizeAttachmentFormat = format => (/^jpe?g$/iu.test(format) ? 'jpeg' : String(format || '').toLowerCase());
+ const requestedImageFormats = [...text.matchAll(new RegExp(
+ `(?<![${SOCIAL_WORD_EDGE}])(${IMAGE_ATTACHMENT_FORMAT})(?![${SOCIAL_WORD_EDGE}])\\s+(?:images?|photos?|pictures?|pics?)`,
+ 'giu',
+ ))].map(match => normalizeAttachmentFormat(match[1]))
+ .filter((format, index, formats) => formats.indexOf(format) === index);
+ const requestedVideoFormats = [...text.matchAll(new RegExp(
+ `(?<![${SOCIAL_WORD_EDGE}])(${VIDEO_ATTACHMENT_FORMAT})(?![${SOCIAL_WORD_EDGE}])\\s+(?:videos?|clips?|recordings?)`,
+ 'giu',
+ ))].map(match => normalizeAttachmentFormat(match[1]))
+ .filter((format, index, formats) => formats.indexOf(format) === index);
const parseCountWord = (str) => {
if (!str) return 0;
@@ -3171,7 +3194,7 @@ export class Agent extends LoopDetector {
const rangeCountToken = '(?:\\d+|zero|one|two|three|four|five|six|seven|eight|nine|ten|a|an)';
const boundedCountRangeRegex = new RegExp(
`\\b(?:between\\s+(${rangeCountToken})\\s+and\\s+(${rangeCountToken})|from\\s+(${rangeCountToken})\\s+to\\s+(${rangeCountToken})|(${rangeCountToken})\\s+to\\s+(${rangeCountToken}))`
- + '\\s+(images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|attachments?|files?|media|uploads?|items?|assets?)\\b',
+ + `\\s+((?:${IMAGE_ATTACHMENT_FORMAT}\\s+)?(?:images?|photos?|pictures?|pics?)|(?:${VIDEO_ATTACHMENT_FORMAT}\\s+)?(?:videos?|clips?|recordings?)|gifs?|attachments?|files?|media|uploads?|items?|assets?)\\b`,
'giu',
);
const boundedCountRanges = [...text.matchAll(boundedCountRangeRegex)]
@@ -3197,9 +3220,12 @@ export class Agent extends LoopDetector {
// scan read the requirement with the qualifier removed.
const hasMinCountQualifier = MIN_ATTACHMENT_COUNT_REGEX.test(text) || hasBoundedCountRange;
const hasMaxCountQualifier = MAX_ATTACHMENT_COUNT_REGEX.test(text) || hasBoundedCountRange;
- const countText = (hasMinCountQualifier || hasMaxCountQualifier)
+ const rawCountText = (hasMinCountQualifier || hasMaxCountQualifier)
? text.replace(MIN_ATTACHMENT_COUNT_STRIP_REGEX, ' ').replace(MAX_ATTACHMENT_COUNT_STRIP_REGEX, ' ').replace(/\s+/g, ' ').trim()
: text;
+ const countText = rawCountText
+ .replace(new RegExp(`(?<![${SOCIAL_WORD_EDGE}])${IMAGE_ATTACHMENT_FORMAT}(?![${SOCIAL_WORD_EDGE}])(?=\\s+(?:images?|photos?|pictures?|pics?))`, 'giu'), '')
+ .replace(new RegExp(`(?<![${SOCIAL_WORD_EDGE}])${VIDEO_ATTACHMENT_FORMAT}(?![${SOCIAL_WORD_EDGE}])(?=\\s+(?:videos?|clips?|recordings?))`, 'giu'), '');
let isGeneric = false;
if (CJK_GENERIC_ATTACHMENT_REGEX.test(countText)) {
@@ -3551,6 +3577,7 @@ export class Agent extends LoopDetector {
const mediaAlternativeBranches = mediaAlternativeBranchTexts.length > 1
? mediaAlternativeBranchTexts
.filter(Boolean)
+ .map(branch => branch.replace(/^either\s+/iu, '').trim())
.map(branch => this._parseWorkflowAttachmentRequirement(branch))
.filter(branch => branch.isGeneric)
: [];
@@ -3591,6 +3618,8 @@ export class Agent extends LoopDetector {
wantsVideo,
wantsOrdinaryVideo,
wantsGif,
+ requestedImageFormats,
+ requestedVideoFormats,
specificTargets,
specificTargetAlternatives: namedTargetAlternatives || [],
normalized: text,
@@ -3632,10 +3661,25 @@ export class Agent extends LoopDetector {
return 'image';
};
const assignment = new Array(attachments.length).fill(-1);
+ let firstAssignment = null;
+ let assignmentCount = 0;
const matchNames = (attachmentIndex, usedNames) => {
- if (attachmentIndex >= attachments.length) return true;
+ if (attachmentIndex >= attachments.length) {
+ assignmentCount++;
+ if (!firstAssignment) firstAssignment = assignment.slice();
+ return assignmentCount > 1;
+ }
const observedKind = attachmentKind(attachments[attachmentIndex]);
- for (let nameIndex = 0; nameIndex < effectiveNames.length; nameIndex++) {
+ const observedNames = this._extractAttachmentCandidateNames(attachments[attachmentIndex]);
+ const evidencedNameIndices = effectiveNames.map((name, nameIndex) => (
+ this._targetAttachmentNames(name).some(targetName => observedNames.includes(targetName))
+ ? nameIndex
+ : -1
+ )).filter(nameIndex => nameIndex >= 0);
+ const candidateNameIndices = evidencedNameIndices.length > 0
+ ? evidencedNameIndices
+ : effectiveNames.map((_name, nameIndex) => nameIndex);
+ for (const nameIndex of candidateNameIndices) {
if (usedNames.has(nameIndex)) continue;
const uploadedKind = nameKind(effectiveNames[nameIndex]);
if (uploadedKind && uploadedKind !== observedKind) continue;
@@ -3647,13 +3691,19 @@ export class Agent extends LoopDetector {
}
return false;
};
- if (!matchNames(0, new Set())) return record;
+ matchNames(0, new Set());
+ if (!firstAssignment) return record;
+ // Multiple type-compatible bijections prove only the aggregate filename
+ // set. Keep that evidence for attachment checks, but mark the node mapping
+ // so filename-scoped metadata such as alt text fails closed below.
+ const uploadNameBindingAmbiguous = assignmentCount > 1;
return {
...record,
+ ...(uploadNameBindingAmbiguous ? { uploadNameBindingAmbiguous: true } : {}),
attachments: attachments.map((attachment, index) => (
typeof attachment === 'string'
- ? { src: attachment, name: effectiveNames[assignment[index]] }
- : { ...attachment, name: effectiveNames[assignment[index]] }
+ ? { src: attachment, name: effectiveNames[firstAssignment[index]] }
+ : { ...attachment, name: effectiveNames[firstAssignment[index]] }
)),
};
}
@@ -3731,6 +3781,38 @@ export class Agent extends LoopDetector {
|| /\.gif(?:[?#]|$)/i.test(name);
};
+ const canonicalAttachmentFormat = value => {
+ const raw = String(value || '').toLowerCase();
+ const match = raw.match(/\.(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?=[?#]|$)/i)
+ || raw.match(/[?&](?:format|fm)=(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?:[&#]|$)/i)
+ || raw.match(/^(?:image|video)\/(?:x-)?(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?:\+xml)?(?:;|$)/i);
+ if (match) return /^jpe?g$/i.test(match[1]) ? 'jpeg' : match[1].toLowerCase();
+ if (/^video\/quicktime(?:;|$)/i.test(raw)) return 'mov';
+ if (/^video\/(?:x-)?matroska(?:;|$)/i.test(raw)) return 'mkv';
+ return '';
+ };
+ const attachmentFormat = att => {
+ if (isGifAttachment(att)) return 'gif';
+ if (att && typeof att === 'object') {
+ for (const value of [att.name, att.fileName, att.filename]) {
+ const format = canonicalAttachmentFormat(value);
+ if (format) return format;
+ }
+ for (const value of [att.mimeType, att.mime, att.contentType]) {
+ const format = canonicalAttachmentFormat(value);
+ if (format) return format;
+ }
+ }
+ return canonicalAttachmentFormat(typeof att === 'string' ? att : (att?.src || att?.url));
+ };
+ const attachmentsMatchFormats = (requiredFormats, candidates) => {
+ if (!requiredFormats?.length) return true;
+ const observedFormats = candidates.map(attachmentFormat);
+ return observedFormats.length > 0
+ && observedFormats.every(format => format && requiredFormats.includes(format))
+ && requiredFormats.every(format => observedFormats.includes(format));
+ };
+
const canMatchSpecificTargets = (targets, candidates) => {
const matchFrom = (targetIdx, usedIndices) => {
if (targetIdx >= targets.length) return true;
@@ -3769,6 +3851,9 @@ export class Agent extends LoopDetector {
const videoCount = rawAttachments.filter(isVideoAttachment).length;
const gifCount = rawAttachments.filter(isGifAttachment).length;
const ordinaryVideoCount = rawAttachments.filter(att => isVideoAttachment(att) && !isGifAttachment(att)).length;
+ if (!attachmentsMatchFormats(parsed.requestedImageFormats, rawAttachments.filter(isImageAttachment))) return false;
+ if (!attachmentsMatchFormats(parsed.requestedVideoFormats,
+ rawAttachments.filter(att => isVideoAttachment(att) && !isGifAttachment(att)))) return false;
const imageAlternativeCounts = parsed.imageAlternativeCounts || [];
const videoAlternativeCounts = parsed.videoAlternativeCounts || [];
const gifAlternativeCounts = parsed.gifAlternativeCounts || [];
@@ -4062,6 +4147,7 @@ export class Agent extends LoopDetector {
const target = this._workflowMetadataValue(
requirement?.attachment ?? requirement?.target ?? requirement?.filename,
);
+ if (target && record?.uploadNameBindingAmbiguous === true) return false;
const relevantAttachments = target
? attachments.filter((attachment) => {
if (!attachment || typeof attachment !== 'object') return false;
diff --git a/src/firefox/src/agent/agent.js b/src/firefox/src/agent/agent.js
index 6601728bc..bf917ce04 100644
--- a/src/firefox/src/agent/agent.js
+++ b/src/firefox/src/agent/agent.js
@@ -195,7 +195,7 @@ const SOCIAL_POST_DESTINATION_NEGATION = new RegExp(
'iu',
);
const SOCIAL_CONTRASTIVE_EXCLUSION = /(?<![\p{L}\p{N}_])(?:rather\s+than|instead\s+of)(?:\s+(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?))?(?:\s+(?:on|onto|to|via|in|at))?\s*$/iu;
-const SOCIAL_DESTINATION_EXCLUSION = /^\s*(?:exclude|excluding)\s+(?:(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?)\s+)?(?:(?:on|onto|to|via|in|at)\s+)?(?:the\s+)?(?:x|twitter|bluesky|bsky(?:\.app)?)(?![\p{L}\p{N}_.-])\s*$/iu;
+const SOCIAL_DESTINATION_EXCLUSION = /^\s*(?:exclude|excluding|no)\s+(?:(?:post(?:ing)?|publish(?:ing)?|shar(?:e|ing)|tweet(?:ing)?|send(?:ing)?)\s+)?(?:(?:on|onto|to|via|in|at)\s+)?(?:the\s+)?(?:x|twitter|bluesky|bsky(?:\.app)?)(?![\p{L}\p{N}_.-])\s*$/iu;
const socialPostNegationGovernsPublish = value => (
SOCIAL_POST_NEGATION.test(String(value || '').trim())
|| SOCIAL_POST_DESTINATION_NEGATION.test(String(value || ''))
@@ -290,6 +290,9 @@ const GENERIC_ATTACHMENT_WORDS = new Set([
const RAW_GIF_NOUN_REGEX = /\b(?:gif|gifs)\b|\.gif(?:[?#]|$)|(?:animated[-_ ]gif|動圖|动图|움짤)/i;
const RAW_VIDEO_NOUN_REGEX = /\b(?:video|videos|mp4|mov|webm|mkv|clip|clips|recording|recordings|vidéo|vidéos)\b|(?:動画|视频|影片|видео|동영상|비디오|영상)/i;
const RAW_IMAGE_NOUN_REGEX = /\b(?:image|images|photo|photos|picture|pictures|pic|pics|png|jpg|jpeg|webp|foto|fotos|bild|bilder|imagen(?:es)?|imágenes|imagem|imagens)\b|(?:画像|写真|图片|照片|圖片|изображение|фото|사진|이미지|포토)/i;
+const IMAGE_ATTACHMENT_FORMAT = '(?:png|jpe?g|webp|avif|heic|bmp|svg)';
+const VIDEO_ATTACHMENT_FORMAT = '(?:mp4|mov|webm|mkv)';
+
// A minimum qualifier binds the count phrase it sits in, not the whole
// requirement: "at least two images and one video" still wants exactly one
// video. These are the boundaries between those phrases.
@@ -442,6 +445,13 @@ const GIF_NEGATION_REGEX = new RegExp(
'iu',
);
+const normalizeAttachmentNegationArticles = value => String(value || '').replace(
+ new RegExp(`(?<![${SOCIAL_WORD_EDGE}])(no|not|without)(\\s+)(a|an|the)(\\s+)`, 'giu'),
+ (_match, negator, beforeArticle, article, afterArticle) => (
+ negator + beforeArticle + ' '.repeat(article.length) + afterArticle
+ ),
+);
+
const VISION_SUB_CALL_TIMEOUT_MS = 90_000;
const CONTENT_ACTION_TIMEOUT_MS = 60_000;
const CONTENT_ACTION_RESPONSE_GRACE_MS = 5_000;
@@ -2754,11 +2764,12 @@ export class Agent extends LoopDetector {
_isNegativeAttachmentRequirement(text) {
const s = String(text || '').trim().toLowerCase();
if (!s) return false;
+ const negationText = normalizeAttachmentNegationArticles(s);
if (/^(?:none|no|false|0|zero|なし|無し|無|없음)$/i.test(s)) return true;
const negPrefix = '(?:no|without|without\\s+any|sin|sans|sem|senza|ohne|без|kein|keine|aucun|aucune|ningun|ningún|ninguna|nenhum|nenhuma|nessun|nessuno|nessuna|nie)';
const nounSuffix = '(?:attachments?|files?|m[eéèê]dias?|m[ií]dia|medien|medios?|uploads?|images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|fichiers?|pièces?|archivos?|adjuntos?|anexos?|allegat[oi]?|anhänge?|anhang|dateien?|вложений|вложения|фото(?:графий)?|изображений|видео|файлов)';
const nounListSeparator = '(?:\\s*[,;]\\s*(?:(?:and|or|nor)\\s+)?|\\s+(?:and|or|nor)\\s+)';
- if (new RegExp(`^${negPrefix}\\s+(?:any\\s+)?${nounSuffix}(?:${nounListSeparator}(?:any\\s+)?${nounSuffix})*$`, 'i').test(s)) return true;
+ if (new RegExp(`^${negPrefix}\\s+(?:any\\s+)?${nounSuffix}(?:${nounListSeparator}(?:any\\s+)?${nounSuffix})*$`, 'i').test(negationText)) return true;
if (new RegExp(`^(?:0|zero)\\s*${nounSuffix}?$`, 'i').test(s)) return true;
if (/^(?:(?:无|没有|不带|零个|0个)\s*(?:附件|图片|照片|视频|媒体|文件)|(?:附件|图片|照片|视频|媒体|文件)\s*(?:无|没有|为0|为零))$/i.test(s)) return true;
if (/^(?:(?:添付|メディア|画像|写真|動画|ファイル)\s*(?:なし|無し|ゼロ|0)|(?:なし|無し)\s*(?:添付|メディア|画像|写真|動画|ファイル))$/i.test(s)) return true;
@@ -2791,7 +2802,8 @@ export class Agent extends LoopDetector {
}
const coordinatedMediaNoun = '(?:images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|animated[-_ ]gifs?)';
- const coordinatedMediaListNegation = text.match(new RegExp(
+ const negationText = normalizeAttachmentNegationArticles(text);
+ const coordinatedMediaListNegation = negationText.match(new RegExp(
`(?<![${SOCIAL_WORD_EDGE}])(?:no|not|without(?:\\s+any)?)\\s+(?:any\\s+)?(${coordinatedMediaNoun}(?:(?:\\s*[,;]\\s*(?:(?:and|or|nor)\\s+)?|\\s+(?:and|or|nor)\\s+)(?:any\\s+)?${coordinatedMediaNoun})+)`,
'iu',
));
@@ -2799,11 +2811,11 @@ export class Agent extends LoopDetector {
const coordinatedMediaNegationSpan = coordinatedMediaListNegation
? [coordinatedMediaListNegation.index || 0, (coordinatedMediaListNegation.index || 0) + coordinatedMediaListNegation[0].length]
: null;
- const isImageNegated = IMAGE_NEGATION_REGEX.test(text)
+ const isImageNegated = IMAGE_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_IMAGE_NOUN_REGEX.test(coordinatedNegatedMedia));
- const isVideoNegated = VIDEO_NEGATION_REGEX.test(text)
+ const isVideoNegated = VIDEO_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_VIDEO_NOUN_REGEX.test(coordinatedNegatedMedia));
- const isGifNegated = GIF_NEGATION_REGEX.test(text)
+ const isGifNegated = GIF_NEGATION_REGEX.test(negationText)
|| Boolean(coordinatedNegatedMedia && RAW_GIF_NOUN_REGEX.test(coordinatedNegatedMedia));
if (isImageNegated && isVideoNegated) {
@@ -2832,6 +2844,17 @@ export class Agent extends LoopDetector {
const wantsOrdinaryVideo = hasRawOrdinaryVideo && !isVideoNegated;
const wantsVideo = wantsOrdinaryVideo || wantsGif;
const wantsImage = hasRawImage && !isImageNegated;
+ const normalizeAttachmentFormat = format => (/^jpe?g$/iu.test(format) ? 'jpeg' : String(format || '').toLowerCase());
+ const requestedImageFormats = [...text.matchAll(new RegExp(
+ `(?<![${SOCIAL_WORD_EDGE}])(${IMAGE_ATTACHMENT_FORMAT})(?![${SOCIAL_WORD_EDGE}])\\s+(?:images?|photos?|pictures?|pics?)`,
+ 'giu',
+ ))].map(match => normalizeAttachmentFormat(match[1]))
+ .filter((format, index, formats) => formats.indexOf(format) === index);
+ const requestedVideoFormats = [...text.matchAll(new RegExp(
+ `(?<![${SOCIAL_WORD_EDGE}])(${VIDEO_ATTACHMENT_FORMAT})(?![${SOCIAL_WORD_EDGE}])\\s+(?:videos?|clips?|recordings?)`,
+ 'giu',
+ ))].map(match => normalizeAttachmentFormat(match[1]))
+ .filter((format, index, formats) => formats.indexOf(format) === index);
const parseCountWord = (str) => {
if (!str) return 0;
@@ -2886,7 +2909,7 @@ export class Agent extends LoopDetector {
const rangeCountToken = '(?:\\d+|zero|one|two|three|four|five|six|seven|eight|nine|ten|a|an)';
const boundedCountRangeRegex = new RegExp(
`\\b(?:between\\s+(${rangeCountToken})\\s+and\\s+(${rangeCountToken})|from\\s+(${rangeCountToken})\\s+to\\s+(${rangeCountToken})|(${rangeCountToken})\\s+to\\s+(${rangeCountToken}))`
- + '\\s+(images?|photos?|pictures?|pics?|videos?|clips?|recordings?|gifs?|attachments?|files?|media|uploads?|items?|assets?)\\b',
+ + `\\s+((?:${IMAGE_ATTACHMENT_FORMAT}\\s+)?(?:images?|photos?|pictures?|pics?)|(?:${VIDEO_ATTACHMENT_FORMAT}\\s+)?(?:videos?|clips?|recordings?)|gifs?|attachments?|files?|media|uploads?|items?|assets?)\\b`,
'giu',
);
const boundedCountRanges = [...text.matchAll(boundedCountRangeRegex)]
@@ -2912,9 +2935,12 @@ export class Agent extends LoopDetector {
// scan read the requirement with the qualifier removed.
const hasMinCountQualifier = MIN_ATTACHMENT_COUNT_REGEX.test(text) || hasBoundedCountRange;
const hasMaxCountQualifier = MAX_ATTACHMENT_COUNT_REGEX.test(text) || hasBoundedCountRange;
- const countText = (hasMinCountQualifier || hasMaxCountQualifier)
+ const rawCountText = (hasMinCountQualifier || hasMaxCountQualifier)
? text.replace(MIN_ATTACHMENT_COUNT_STRIP_REGEX, ' ').replace(MAX_ATTACHMENT_COUNT_STRIP_REGEX, ' ').replace(/\s+/g, ' ').trim()
: text;
+ const countText = rawCountText
+ .replace(new RegExp(`(?<![${SOCIAL_WORD_EDGE}])${IMAGE_ATTACHMENT_FORMAT}(?![${SOCIAL_WORD_EDGE}])(?=\\s+(?:images?|photos?|pictures?|pics?))`, 'giu'), '')
+ .replace(new RegExp(`(?<![${SOCIAL_WORD_EDGE}])${VIDEO_ATTACHMENT_FORMAT}(?![${SOCIAL_WORD_EDGE}])(?=\\s+(?:videos?|clips?|recordings?))`, 'giu'), '');
let isGeneric = false;
if (CJK_GENERIC_ATTACHMENT_REGEX.test(countText)) {
@@ -3266,6 +3292,7 @@ export class Agent extends LoopDetector {
const mediaAlternativeBranches = mediaAlternativeBranchTexts.length > 1
? mediaAlternativeBranchTexts
.filter(Boolean)
+ .map(branch => branch.replace(/^either\s+/iu, '').trim())
.map(branch => this._parseWorkflowAttachmentRequirement(branch))
.filter(branch => branch.isGeneric)
: [];
@@ -3306,6 +3333,8 @@ export class Agent extends LoopDetector {
wantsVideo,
wantsOrdinaryVideo,
wantsGif,
+ requestedImageFormats,
+ requestedVideoFormats,
specificTargets,
specificTargetAlternatives: namedTargetAlternatives || [],
normalized: text,
@@ -3347,10 +3376,25 @@ export class Agent extends LoopDetector {
return 'image';
};
const assignment = new Array(attachments.length).fill(-1);
+ let firstAssignment = null;
+ let assignmentCount = 0;
const matchNames = (attachmentIndex, usedNames) => {
- if (attachmentIndex >= attachments.length) return true;
+ if (attachmentIndex >= attachments.length) {
+ assignmentCount++;
+ if (!firstAssignment) firstAssignment = assignment.slice();
+ return assignmentCount > 1;
+ }
const observedKind = attachmentKind(attachments[attachmentIndex]);
- for (let nameIndex = 0; nameIndex < effectiveNames.length; nameIndex++) {
+ const observedNames = this._extractAttachmentCandidateNames(attachments[attachmentIndex]);
+ const evidencedNameIndices = effectiveNames.map((name, nameIndex) => (
+ this._targetAttachmentNames(name).some(targetName => observedNames.includes(targetName))
+ ? nameIndex
+ : -1
+ )).filter(nameIndex => nameIndex >= 0);
+ const candidateNameIndices = evidencedNameIndices.length > 0
+ ? evidencedNameIndices
+ : effectiveNames.map((_name, nameIndex) => nameIndex);
+ for (const nameIndex of candidateNameIndices) {
if (usedNames.has(nameIndex)) continue;
const uploadedKind = nameKind(effectiveNames[nameIndex]);
if (uploadedKind && uploadedKind !== observedKind) continue;
@@ -3362,13 +3406,19 @@ export class Agent extends LoopDetector {
}
return false;
};
- if (!matchNames(0, new Set())) return record;
+ matchNames(0, new Set());
+ if (!firstAssignment) return record;
+ // Multiple type-compatible bijections prove only the aggregate filename
+ // set. Keep that evidence for attachment checks, but mark the node mapping
+ // so filename-scoped metadata such as alt text fails closed below.
+ const uploadNameBindingAmbiguous = assignmentCount > 1;
return {
...record,
+ ...(uploadNameBindingAmbiguous ? { uploadNameBindingAmbiguous: true } : {}),
attachments: attachments.map((attachment, index) => (
typeof attachment === 'string'
- ? { src: attachment, name: effectiveNames[assignment[index]] }
- : { ...attachment, name: effectiveNames[assignment[index]] }
+ ? { src: attachment, name: effectiveNames[firstAssignment[index]] }
+ : { ...attachment, name: effectiveNames[firstAssignment[index]] }
)),
};
}
@@ -3446,6 +3496,38 @@ export class Agent extends LoopDetector {
|| /\.gif(?:[?#]|$)/i.test(name);
};
+ const canonicalAttachmentFormat = value => {
+ const raw = String(value || '').toLowerCase();
+ const match = raw.match(/\.(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?=[?#]|$)/i)
+ || raw.match(/[?&](?:format|fm)=(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?:[&#]|$)/i)
+ || raw.match(/^(?:image|video)\/(?:x-)?(png|jpe?g|webp|avif|heic|bmp|svg|mp4|mov|webm|mkv)(?:\+xml)?(?:;|$)/i);
+ if (match) return /^jpe?g$/i.test(match[1]) ? 'jpeg' : match[1].toLowerCase();
+ if (/^video\/quicktime(?:;|$)/i.test(raw)) return 'mov';
+ if (/^video\/(?:x-)?matroska(?:;|$)/i.test(raw)) return 'mkv';
+ return '';
+ };
+ const attachmentFormat = att => {
+ if (isGifAttachment(att)) return 'gif';
+ if (att && typeof att === 'object') {
+ for (const value of [att.name, att.fileName, att.filename]) {
+ const format = canonicalAttachmentFormat(value);
+ if (format) return format;
+ }
+ for (const value of [att.mimeType, att.mime, att.contentType]) {
+ const format = canonicalAttachmentFormat(value);
+ if (format) return format;
+ }
+ }
+ return canonicalAttachmentFormat(typeof att === 'string' ? att : (att?.src || att?.url));
+ };
+ const attachmentsMatchFormats = (requiredFormats, candidates) => {
+ if (!requiredFormats?.length) return true;
+ const observedFormats = candidates.map(attachmentFormat);
+ return observedFormats.length > 0
+ && observedFormats.every(format => format && requiredFormats.includes(format))
+ && requiredFormats.every(format => observedFormats.includes(format));
+ };
+
const canMatchSpecificTargets = (targets, candidates) => {
const matchFrom = (targetIdx, usedIndices) => {
if (targetIdx >= targets.length) return true;
@@ -3484,6 +3566,9 @@ export class Agent extends LoopDetector {
const videoCount = rawAttachments.filter(isVideoAttachment).length;
const gifCount = rawAttachments.filter(isGifAttachment).length;
const ordinaryVideoCount = rawAttachments.filter(att => isVideoAttachment(att) && !isGifAttachment(att)).length;
+ if (!attachmentsMatchFormats(parsed.requestedImageFormats, rawAttachments.filter(isImageAttachment))) return false;
+ if (!attachmentsMatchFormats(parsed.requestedVideoFormats,
+ rawAttachments.filter(att => isVideoAttachment(att) && !isGifAttachment(att)))) return false;
const imageAlternativeCounts = parsed.imageAlternativeCounts || [];
const videoAlternativeCounts = parsed.videoAlternativeCounts || [];
const gifAlternativeCounts = parsed.gifAlternativeCounts || [];
@@ -3777,6 +3862,7 @@ export class Agent extends LoopDetector {
const target = this._workflowMetadataValue(
requirement?.attachment ?? requirement?.target ?? requirement?.filename,
);
+ if (target && record?.uploadNameBindingAmbiguous === true) return false;
const relevantAttachments = target
? attachments.filter((attachment) => {
if (!attachment || typeof attachment !== 'object') return false;
diff --git a/test/run.js b/test/run.js
index 4333f88ec..df040e0a8 100644
--- a/test/run.js
+++ b/test/run.js
@@ -90817,6 +90817,10 @@ test('social publication workflow follows the live X or Bluesky destination and
'an explicitly excluded X destination was adopted'],
['Post on Bluesky, exclude X', ['bluesky'],
'an exclude clause adopted its forbidden X destination'],
+ ['Post only on Bluesky, no X', ['bluesky'],
+ 'a bare no-X clause adopted its forbidden destination'],
+ ['Post only on X, no Bluesky', ['twitter'],
+ 'a bare no-Bluesky clause adopted its forbidden destination'],
['Post on Bluesky rather than posting on X', ['bluesky'],
'rather-than clause adopted its expressly excluded X destination'],
['Post on Bluesky rather than on X', ['bluesky'],
@@ -92886,11 +92890,35 @@ test('social upload provenance reconciles only from complete composer snapshots'
});
assert.deepEqual(joined.attachments.map(item => item.name), ['a.png', 'c.png'],
AgentClass.name + ': reconciled active provenance did not join published media');
+ assert.equal(joined.uploadNameBindingAmbiguous, true,
+ AgentClass.name + ': same-type upload/DOM ordering was treated as a stable filename binding');
assert.equal(
agent._workflowSocialPublishedAttachmentObserved({ value: 'a.png and c.png' }, joined),
true,
AgentClass.name + ': active replacement media could not verify after reconciliation',
);
+ const ambiguousAltRecord = agent._workflowSocialRecordWithUploadedAttachmentNames({
+ attachments: [
+ { type: 'image', src: 'https://cdn.example/opaque-1', alt: 'Alt for a' },
+ { type: 'image', src: 'https://cdn.example/opaque-2', alt: 'Alt for c' },
+ ],
+ }, { uploadedAttachmentNames: ['a.png', 'c.png'] });
+ assert.equal(agent._workflowSocialPublishedAltTextObserved(
+ { attachment: 'a.png', value: 'Alt for a' }, ambiguousAltRecord,
+ ), false, AgentClass.name + ': upload order falsely bound targeted alt text to an opaque media card');
+ const stableAltRecord = agent._workflowSocialRecordWithUploadedAttachmentNames({
+ attachments: [
+ { type: 'image', src: 'https://cdn.example/c.png', alt: 'Alt for c' },
+ { type: 'image', src: 'https://cdn.example/a.png', alt: 'Alt for a' },
+ ],
+ }, { uploadedAttachmentNames: ['a.png', 'c.png'] });
+ assert.equal(stableAltRecord.uploadNameBindingAmbiguous, undefined,
+ AgentClass.name + ': filename evidence did not disambiguate reordered media cards');
+ assert.deepEqual(stableAltRecord.attachments.map(item => item.name), ['c.png', 'a.png'],
+ AgentClass.name + ': stable per-card filename evidence was ignored');
+ assert.equal(agent._workflowSocialPublishedAltTextObserved(
+ { attachment: 'a.png', value: 'Alt for a' }, stableAltRecord,
+ ), true, AgentClass.name + ': stable reordered filename/alt provenance did not verify');
}
});
@@ -92911,6 +92939,33 @@ test('upper-bound attachment qualifiers verify as maximum counts', () => {
const image = index => ({ type: 'image', src: `https://pbs.twimg.com/media/image${index}.png` });
const video = index => ({ type: 'video', src: `https://video.twimg.com/media/video${index}.mp4` });
const gif = index => ({ type: 'animated_gif', src: `https://video.twimg.com/tweet_video/gif${index}.mp4` });
+ const pngImage = agent._parseWorkflowAttachmentRequirement('one PNG image');
+ assert.equal(pngImage.isGeneric, true,
+ AgentClass.name + ': a PNG media qualifier was parsed as a filename');
+ assert.deepEqual(pngImage.specificTargets, [],
+ AgentClass.name + ': a PNG media qualifier created a specific target');
+ assert.deepEqual(pngImage.requestedImageFormats, ['png']);
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one PNG image' }, { attachments: [image(1)] }), true,
+ AgentClass.name + ': a PNG image did not satisfy its format-qualified contract');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one PNG image' }, { attachments: [{ type: 'image', src: 'https://cdn.example/media', name: 'source.png' }] }), true,
+ AgentClass.name + ': original upload provenance did not prove PNG format');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one PNG image' }, { attachments: [{ type: 'image', src: 'https://pbs.twimg.com/media/image1.jpg' }] }), false,
+ AgentClass.name + ': a JPEG image satisfied a PNG-only contract');
+ const mp4Video = agent._parseWorkflowAttachmentRequirement('one MP4 video');
+ assert.equal(mp4Video.isGeneric, true,
+ AgentClass.name + ': an MP4 media qualifier was parsed as a filename');
+ assert.deepEqual(mp4Video.specificTargets, [],
+ AgentClass.name + ': an MP4 media qualifier created a specific target');
+ assert.deepEqual(mp4Video.requestedVideoFormats, ['mp4']);
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one MP4 video' }, { attachments: [video(1)] }), true,
+ AgentClass.name + ': an MP4 video did not satisfy its format-qualified contract');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one MP4 video' }, { attachments: [{ type: 'video', src: 'https://video.twimg.com/media/video1.webm' }] }), false,
+ AgentClass.name + ': a WebM video satisfied an MP4-only contract');
assert.equal(agent._workflowSocialPublishedAttachmentObserved(
{ value: 'up to two images' }, { attachments: [] }), true,
AgentClass.name + ': zero images should satisfy "up to two images"');
@@ -93001,6 +93056,31 @@ test('upper-bound attachment qualifiers verify as maximum counts', () => {
assert.equal(agent._workflowSocialPublishedAttachmentObserved(
{ value: 'one image and no videos or GIFs' }, { attachments: [image(1), gif(1)] }), false,
AgentClass.name + ': embedded coordinated media negation allowed a GIF');
+ const imageWithoutArticleVideo = agent._parseWorkflowAttachmentRequirement('one image without a video');
+ assert.equal(imageWithoutArticleVideo.wantsVideo, false,
+ AgentClass.name + ': an article hid the negated video requirement');
+ assert.equal(imageWithoutArticleVideo.isVideoNegated, true,
+ AgentClass.name + ': without-a-video did not retain its negative subtype constraint');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one image without a video' }, { attachments: [image(1)] }), true,
+ AgentClass.name + ': an image-only post failed without-a-video verification');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'one image without a video' }, { attachments: [image(1), video(1)] }), false,
+ AgentClass.name + ': a forbidden video satisfied without-a-video verification');
+ const emptyOrVideo = agent._parseWorkflowAttachmentRequirement('either no images or one video');
+ assert.equal(emptyOrVideo.mediaAlternativeBranches.length, 2,
+ AgentClass.name + ': empty-or-video requirement did not preserve both branches');
+ assert.equal(emptyOrVideo.mediaAlternativeBranches[0].wantsNone, true,
+ AgentClass.name + ': leading either prevented parsing the negative media branch');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'either no images or one video' }, { attachments: [] }), true,
+ AgentClass.name + ': empty media did not satisfy the negative alternative');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'either no images or one video' }, { attachments: [video(1)] }), true,
+ AgentClass.name + ': one video did not satisfy the positive alternative');
+ assert.equal(agent._workflowSocialPublishedAttachmentObserved(
+ { value: 'either no images or one video' }, { attachments: [image(1), image(2)] }), false,
+ AgentClass.name + ': multiple images incorrectly satisfied an empty-or-video contract');
const boundedImages = agent._parseWorkflowAttachmentRequirement('between one and two images');
assert.equal(boundedImages.isGeneric, true,