| @@ -83,9 +83,11 @@ | ||
| 83 | 83 | |
| 84 | 84 | /** |
| 85 | 85 | * AI client instance |
| 86 | 86 | * |
| 87 | - * @var OpenAI_Client|Claude_Client | |
| 87 | + * Null when the generator was built for storage-only work. | |
| 88 | + * | |
| 89 | + * @var OpenAI_Client|Claude_Client|null | |
| 88 | 90 | */ |
| 89 | 91 | private $ai_client; |
| 90 | 92 | |
| 91 | 93 | /** |
| @@ -92,18 +94,31 @@ | ||
| 92 | 94 | * Constructor |
| 93 | 95 | * |
| 94 | 96 | * @param Settings|null $settings Settings instance |
| 95 | 97 | * @param OpenAI_Client|Claude_Client|null $ai_client AI client instance |
| 98 | + * @param bool $require_ai_client Whether a provider client is required. Pass | |
| 99 | + * false for storage-only use (list/export/ | |
| 100 | + * delete), which never calls a provider. | |
| 96 | 101 | */ |
| 97 | - public function __construct(?Settings $settings = null, $ai_client = null) { | |
| 102 | + public function __construct(?Settings $settings = null, $ai_client = null, bool $require_ai_client = true) { | |
| 98 | 103 | $this->settings = $settings ?? Settings::instance(); |
| 99 | 104 | |
| 100 | 105 | if ($ai_client) { |
| 101 | 106 | $this->ai_client = $ai_client; |
| 102 | - } else { | |
| 103 | - // Fallback to creating own client for backward compatibility | |
| 104 | - $this->init_ai_client(); | |
| 107 | + | |
| 108 | + return; | |
| 105 | 109 | } |
| 110 | + | |
| 111 | + // Read-only callers (listing, exporting and deleting saved briefs) only | |
| 112 | + // touch the database and never reach a provider. Constructing a client | |
| 113 | + // for them turns "no API key configured" — the default state of a fresh | |
| 114 | + // install — into a hard failure, so let them opt out. | |
| 115 | + if (!$require_ai_client) { | |
| 116 | + return; | |
| 117 | + } | |
| 118 | + | |
| 119 | + // Fallback to creating own client for backward compatibility | |
| 120 | + $this->init_ai_client(); | |
| 106 | 121 | } |
| 107 | 122 | |
| 108 | 123 | /** |
| 109 | 124 | * Initialize AI client based on available API keys |
| @@ -112,9 +127,9 @@ | ||
| 112 | 127 | * |
| 113 | 128 | * @throws \Exception On failure. |
| 114 | 129 | */ |
| 115 | 130 | private function init_ai_client(): void { |
| 116 | - $provider = $this->settings->get('ai_provider', 'openai'); | |
| 131 | + $provider = $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); | |
| 117 | 132 | |
| 118 | 133 | if ($provider === 'openai') { |
| 119 | 134 | $api_key = $this->settings->get('openai_api_key'); |
| 120 | 135 | if ($api_key) { |
| @@ -138,11 +153,31 @@ | ||
| 138 | 153 | if ($api_key) { |
| 139 | 154 | $model = $this->settings->get('openrouter_model', Settings::DEFAULT_OPENROUTER_MODEL); |
| 140 | 155 | $this->ai_client = new OpenRouter_Client($api_key, $model, self::AI_REQUEST_TIMEOUT); |
| 141 | 156 | } |
| 157 | + } elseif ($provider === 'openai_compatible') { | |
| 158 | + // Same client as OpenAI, different host — and the key is optional, | |
| 159 | + // so the URL and model id are what gate it (#721). The user's own | |
| 160 | + // timeout applies: a local model writing a brief on CPU is slow, | |
| 161 | + // and the setting exists for exactly that. | |
| 162 | + $base_url = (string) $this->settings->get('openai_compatible_base_url', ''); | |
| 163 | + $model = trim((string) $this->settings->get('openai_compatible_model', '')); | |
| 164 | + if ('' !== $base_url && '' !== $model) { | |
| 165 | + $this->ai_client = new OpenAI_Client( | |
| 166 | + (string) $this->settings->get('openai_compatible_api_key', ''), | |
| 167 | + $model, | |
| 168 | + (int) $this->settings->get('openai_compatible_timeout', Settings::DEFAULT_OPENAI_COMPATIBLE_TIMEOUT), | |
| 169 | + $base_url | |
| 170 | + ); | |
| 171 | + $this->ai_client->set_json_mode((bool) $this->settings->get('openai_compatible_json_mode', false)); | |
| 172 | + } | |
| 142 | 173 | } |
| 143 | 174 | |
| 144 | 175 | if (!$this->ai_client) { |
| 176 | + if ('openai_compatible' === $provider) { | |
| 177 | + throw new \Exception('Please set the base URL and model id for your OpenAI-compatible endpoint in ThinkRank settings.'); | |
| 178 | + } | |
| 179 | + | |
| 145 | 180 | throw new \Exception('Please configure your AI provider API key in ThinkRank settings.'); |
| 146 | 181 | } |
| 147 | 182 | } |
| 148 | 183 | |
| @@ -157,9 +192,15 @@ | ||
| 157 | 192 | return $this->ai_client->get_model(); |
| 158 | 193 | } |
| 159 | 194 | |
| 160 | 195 | // Fallback to settings |
| 161 | - $provider = $this->settings->get('ai_provider', 'openai'); | |
| 196 | + $provider = $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); | |
| 197 | + if (Settings::AI_PROVIDER_NONE === $provider) { | |
| 198 | + // No provider chosen, so there is no model to name. Reporting the | |
| 199 | + // OpenAI default here would attribute output to a provider the site | |
| 200 | + // never selected (#572). | |
| 201 | + return ''; | |
| 202 | + } | |
| 162 | 203 | if ($provider === 'claude') { |
| 163 | 204 | return $this->settings->get('claude_model', Settings::DEFAULT_CLAUDE_MODEL); |
| 164 | 205 | } elseif ($provider === 'gemini') { |
| 165 | 206 | return $this->settings->get('gemini_model', Settings::DEFAULT_GEMINI_MODEL); |
| @@ -164,8 +205,10 @@ | ||
| 164 | 205 | } elseif ($provider === 'gemini') { |
| 165 | 206 | return $this->settings->get('gemini_model', Settings::DEFAULT_GEMINI_MODEL); |
| 166 | 207 | } elseif ($provider === 'openrouter') { |
| 167 | 208 | return $this->settings->get('openrouter_model', Settings::DEFAULT_OPENROUTER_MODEL); |
| 209 | + } elseif ($provider === 'openai_compatible') { | |
| 210 | + return (string) $this->settings->get('openai_compatible_model', ''); | |
| 168 | 211 | } else { |
| 169 | 212 | return $this->settings->get('openai_model', Settings::DEFAULT_OPENAI_MODEL); |
| 170 | 213 | } |
| 171 | 214 | } |
| @@ -178,9 +221,9 @@ | ||
| 178 | 221 | * costliest configuration, where billed reasoning tokens (drawn from the |
| 179 | 222 | * same budget) are spent before any visible output (issue #286). |
| 180 | 223 | * |
| 181 | 224 | * A brief is a structured planning task, so 'low' is a provisional middle |
| 182 | - * ground — Brand Visibility uses 'minimal' for quick consumer-style answers. | |
| 225 | + * ground between 'minimal' and the model's default. | |
| 183 | 226 | * The level is filterable so a site can trade latency for more reasoning; |
| 184 | 227 | * returning '' opts out entirely and lets the model use its default effort. |
| 185 | 228 | * Only the GPT-5 family consumes this — o1/o3, gpt-4o and the non-OpenAI |
| 186 | 229 | * clients ignore an unrecognised option key. |
| @@ -212,9 +255,9 @@ | ||
| 212 | 255 | * |
| 213 | 256 | * @return string Current provider name |
| 214 | 257 | */ |
| 215 | 258 | private function get_current_provider(): string { |
| 216 | - return $this->settings->get('ai_provider', 'openai'); | |
| 259 | + return $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE); | |
| 217 | 260 | } |
| 218 | 261 | |
| 219 | 262 | /** |
| 220 | 263 | * Extract token usage from AI response |
| @@ -224,9 +267,9 @@ | ||
| 224 | 267 | */ |
| 225 | 268 | private function extract_token_usage(array $ai_response): int { |
| 226 | 269 | $provider = $this->get_current_provider(); |
| 227 | 270 | |
| 228 | - if ($provider === 'openai' || $provider === 'openrouter') { | |
| 271 | + if ($provider === 'openai' || $provider === 'openrouter' || $provider === 'openai_compatible') { | |
| 229 | 272 | // OpenAI-compatible format: response['usage']['total_tokens'] |
| 230 | 273 | return (int) ($ai_response['usage']['total_tokens'] ?? 0); |
| 231 | 274 | } elseif ($provider === 'claude') { |
| 232 | 275 | // Claude format: response['usage']['input_tokens'] + response['usage']['output_tokens'] |
| @@ -326,8 +369,11 @@ | ||
| 326 | 369 | // max_completion_tokens internally. Temperature is intentionally |
| 327 | 370 | // omitted: every client defaults it to 0.7, and reasoning models |
| 328 | 371 | // reject it outright, so passing it here was misleading no-op. |
| 329 | 372 | 'max_tokens' => $max_tokens, |
| 373 | + // The brief is one JSON object. Only a compatible endpoint | |
| 374 | + // with JSON mode on reads this; every other client ignores it. | |
| 375 | + 'json_object' => true, | |
| 330 | 376 | ]; |
| 331 | 377 | if ('' !== $reasoning_effort) { |
| 332 | 378 | $completion_options['reasoning_effort'] = $reasoning_effort; |
| 333 | 379 | } |
| @@ -339,9 +385,9 @@ | ||
| 339 | 385 | // truncation) BEFORE attempting text extraction. Otherwise a |
| 340 | 386 | // refusal — which OpenAI returns as HTTP 200 with content=null — |
| 341 | 387 | // slips past every isset() branch and gets serialized into the |
| 342 | 388 | // brief body instead of being reported to the user. |
| 343 | - $this->guard_against_non_answer($ai_response); | |
| 389 | + $this->guard_against_non_answer($ai_response, $max_tokens); | |
| 344 | 390 | |
| 345 | 391 | // Extract text content from AI response |
| 346 | 392 | $ai_text = ''; |
| 347 | 393 | |
| @@ -489,12 +535,13 @@ | ||
| 489 | 535 | * don't catch them here they fall through to the "unexpected format" path |
| 490 | 536 | * (or, historically, were serialized into the brief body). All messages |
| 491 | 537 | * start with "The AI " so the outer catch passes them through unchanged. |
| 492 | 538 | * |
| 493 | - * @param mixed $ai_response Raw response from the AI client. | |
| 539 | + * @param mixed $ai_response Raw response from the AI client. | |
| 540 | + * @param int $requested_tokens The max_tokens this request asked for; 0 when unknown. | |
| 494 | 541 | * @throws \Exception If the response is a refusal, policy block, or truncation. |
| 495 | 542 | */ |
| 496 | - private function guard_against_non_answer($ai_response): void { | |
| 543 | + private function guard_against_non_answer($ai_response, int $requested_tokens = 0): void { | |
| 497 | 544 | if (!is_array($ai_response)) { |
| 498 | 545 | return; |
| 499 | 546 | } |
| 500 | 547 | |
| @@ -506,18 +553,30 @@ | ||
| 506 | 553 | $message = $ai_response['choices'][0]['message']; |
| 507 | 554 | $finish = (string) ($ai_response['choices'][0]['finish_reason'] ?? ''); |
| 508 | 555 | |
| 509 | 556 | if (!empty($message['refusal'])) { |
| 510 | - throw new \Exception(sprintf( | |
| 557 | + throw new \Exception(esc_html(sprintf( | |
| 511 | 558 | 'The AI declined to generate this brief: %s', |
| 512 | 559 | (string) $message['refusal'] |
| 513 | - )); | |
| 560 | + ))); | |
| 514 | 561 | } |
| 515 | 562 | if ('content_filter' === $finish) { |
| 516 | 563 | throw new \Exception('The AI blocked this request under its content policy. Try a different topic or less sensitive keywords.'); |
| 517 | 564 | } |
| 565 | + // A self-hosted server can stop short of max_tokens because the | |
| 566 | + // prompt and the answer together filled its context window | |
| 567 | + // (Ollama loads models at 4096 by default). A bigger output budget | |
| 568 | + // cannot fix that, so say what can. | |
| 569 | + $completion_tokens = (int) ($ai_response['usage']['completion_tokens'] ?? 0); | |
| 570 | + if ('length' === $finish && $requested_tokens > 0 && $completion_tokens > 0 && $completion_tokens < $requested_tokens) { | |
| 571 | + throw new \Exception(esc_html(sprintf( | |
| 572 | + 'The AI stopped after %1$d tokens, short of the %2$d allowed, because the server ran out of context window before finishing the brief. Raise the context length on your AI server (for Ollama, set OLLAMA_CONTEXT_LENGTH to 16384 or more) and try again.', | |
| 573 | + $completion_tokens, | |
| 574 | + $requested_tokens | |
| 575 | + ))); | |
| 576 | + } | |
| 518 | 577 | if ('length' === $finish) { |
| 519 | - throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try a shorter content length or fewer competitor URLs.'); | |
| 578 | + throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); | |
| 520 | 579 | } |
| 521 | 580 | } |
| 522 | 581 | |
| 523 | 582 | // --- Claude (Messages) --- |
| @@ -526,9 +585,9 @@ | ||
| 526 | 585 | if ('refusal' === $stop_reason) { |
| 527 | 586 | throw new \Exception('The AI declined to generate this brief for this topic. Try a different topic or less sensitive keywords.'); |
| 528 | 587 | } |
| 529 | 588 | if ('max_tokens' === $stop_reason) { |
| 530 | - throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try a shorter content length or fewer competitor URLs.'); | |
| 589 | + throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); | |
| 531 | 590 | } |
| 532 | 591 | } |
| 533 | 592 | |
| 534 | 593 | // --- Gemini --- |
| @@ -536,12 +595,12 @@ | ||
| 536 | 595 | // promptFeedback.blockReason; a candidate can also finish on SAFETY or |
| 537 | 596 | // PROHIBITED_CONTENT, or be truncated at MAX_TOKENS. |
| 538 | 597 | $block_reason = (string) ($ai_response['promptFeedback']['blockReason'] ?? ''); |
| 539 | 598 | if ('' !== $block_reason) { |
| 540 | - throw new \Exception(sprintf( | |
| 599 | + throw new \Exception(esc_html(sprintf( | |
| 541 | 600 | 'The AI blocked this request under its content policy (%s). Try a different topic or less sensitive keywords.', |
| 542 | 601 | $block_reason |
| 543 | - )); | |
| 602 | + ))); | |
| 544 | 603 | } |
| 545 | 604 | $gemini_finish = (string) ($ai_response['candidates'][0]['finishReason'] ?? ''); |
| 546 | 605 | if (in_array($gemini_finish, ['SAFETY', 'PROHIBITED_CONTENT'], true)) { |
| 547 | 606 | throw new \Exception('The AI blocked this request under its content policy. Try a different topic or less sensitive keywords.'); |
| @@ -546,9 +605,9 @@ | ||
| 546 | 605 | if (in_array($gemini_finish, ['SAFETY', 'PROHIBITED_CONTENT'], true)) { |
| 547 | 606 | throw new \Exception('The AI blocked this request under its content policy. Try a different topic or less sensitive keywords.'); |
| 548 | 607 | } |
| 549 | 608 | if ('MAX_TOKENS' === $gemini_finish) { |
| 550 | - throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try a shorter content length or fewer competitor URLs.'); | |
| 609 | + throw new \Exception('The AI stopped at its output token limit before finishing the brief. Try fewer competitor URLs, or a model with a larger output limit. A shorter content length will not help: it asks for a smaller budget, not a smaller answer.'); | |
| 551 | 610 | } |
| 552 | 611 | } |
| 553 | 612 | |
| 554 | 613 | /** |
| @@ -642,9 +701,9 @@ | ||
| 642 | 701 | 'title' => $json_data['title_suggestions'] ?? [], |
| 643 | 702 | 'meta_description' => $json_data['meta_descriptions'][0] ?? '', |
| 644 | 703 | 'meta_descriptions' => $json_data['meta_descriptions'] ?? [], |
| 645 | 704 | 'url_slugs' => $json_data['url_slugs'] ?? [], |
| 646 | - 'outline' => $json_data['outline'] ?? [], | |
| 705 | + 'outline' => self::strip_outline_level_labels($json_data['outline'] ?? []), | |
| 647 | 706 | 'seo_recommendations' => [ |
| 648 | 707 | 'title_suggestions' => $json_data['title_suggestions'] ?? [], |
| 649 | 708 | 'meta_description' => $json_data['meta_descriptions'][0] ?? '', |
| 650 | 709 | 'meta_descriptions' => $json_data['meta_descriptions'] ?? [], |
| @@ -670,9 +729,9 @@ | ||
| 670 | 729 | ], |
| 671 | 730 | 'competitor_gaps' => $json_data['competitor_analysis']['content_gaps'] ?? [], |
| 672 | 731 | 'call_to_actions' => $json_data['call_to_actions'] ?? [], |
| 673 | 732 | 'writing_guidelines' => $json_data['writing_guidelines'] ?? [], |
| 674 | - 'content_body' => $json_data['content_body'] ?? '', | |
| 733 | + 'content_body' => self::strip_heading_level_labels((string) ($json_data['content_body'] ?? '')), | |
| 675 | 734 | 'estimated_word_count' => $this->get_word_count_estimate($original_params['content_length'] ?? 'medium'), |
| 676 | 735 | 'raw_response' => '', // Will be retrieved from ai_usage table |
| 677 | 736 | 'generation_params' => $original_params, |
| 678 | 737 | 'parsing_status' => 'success', |
| @@ -680,8 +739,118 @@ | ||
| 680 | 739 | ]; |
| 681 | 740 | } |
| 682 | 741 | |
| 683 | 742 | /** |
| 743 | + * Remove a leading level label from a heading string. | |
| 744 | + * | |
| 745 | + * The prompt's own JSON example labelled outline headings with their level | |
| 746 | + * (`"heading": "H1: Main Title"` next to a separate `"level": 1`), so the | |
| 747 | + * model often carried the convention into the drafted article and Pro's | |
| 748 | + * "Insert into post" wrote `<h2>H2: Real Heading</h2>` into published | |
| 749 | + * content. The prompt no longer does that, but a prompt change never fully | |
| 750 | + * binds a model — so the label is stripped here too (#410). | |
| 751 | + * | |
| 752 | + * Covers the label forms a model actually emits: `H2:`, `h3:`, `H2 -`, | |
| 753 | + * `H4.`, `H2)` and the en/em dash variants, optionally wrapped in markdown | |
| 754 | + * emphasis (`**H2:**`). The delimiter is anchored directly after the digit | |
| 755 | + * so `H10:` — a plausible heading in a numbered list — is left alone, and | |
| 756 | + * only a leading label is matched so body copy that mentions a level | |
| 757 | + * survives. Trailing emphasis is consumed only when the same marker opened | |
| 758 | + * the label, so `H2: *emphasised start*` keeps its asterisks. | |
| 759 | + * | |
| 760 | + * @since 2.0.1 | |
| 761 | + * | |
| 762 | + * @param string $heading Heading text. | |
| 763 | + * @return string Heading without its level prefix. | |
| 764 | + */ | |
| 765 | + public static function strip_level_label(string $heading): string { | |
| 766 | + // En dash and em dash as raw UTF-8 bytes, so the pattern needs no /u | |
| 767 | + // modifier and cannot blank a heading that is not valid UTF-8. | |
| 768 | + $delimiter = '(?:[:.)\-]|\xe2\x80\x93|\xe2\x80\x94)'; | |
| 769 | + $emphasis = '(\*{1,3}|_{1,3})'; | |
| 770 | + | |
| 771 | + $pattern = '/^\s*(?:' | |
| 772 | + . $emphasis . '\s*[Hh][1-6]\s*' . $delimiter . '\s*\1' | |
| 773 | + . '|[Hh][1-6]\s*' . $delimiter | |
| 774 | + . ')\s*/'; | |
| 775 | + | |
| 776 | + return (string) preg_replace($pattern, '', $heading); | |
| 777 | + } | |
| 778 | + | |
| 779 | + /** | |
| 780 | + * Strip a level label from a heading's inner HTML. | |
| 781 | + * | |
| 782 | + * A model drafting publish-ready HTML often wraps the heading text in an | |
| 783 | + * inline tag (`<h2><strong>H2: Real Heading</strong></h2>`). That pushes a | |
| 784 | + * `<` in front of the label, so the leading run of inline opening tags is | |
| 785 | + * set aside and re-attached around the cleaned text. | |
| 786 | + * | |
| 787 | + * @since 2.0.1 | |
| 788 | + * | |
| 789 | + * @param string $inner Heading inner HTML. | |
| 790 | + * @return string Inner HTML without the level prefix. | |
| 791 | + */ | |
| 792 | + private static function strip_inner_level_label(string $inner): string { | |
| 793 | + $prefix = ''; | |
| 794 | + | |
| 795 | + if (preg_match('/^(\s*(?:<(?:strong|em|b|i|span|mark|code|u)\b[^>]*>\s*)+)(.*)$/is', $inner, $parts)) { | |
| 796 | + $prefix = $parts[1]; | |
| 797 | + $inner = $parts[2]; | |
| 798 | + } | |
| 799 | + | |
| 800 | + return $prefix . self::strip_level_label($inner); | |
| 801 | + } | |
| 802 | + | |
| 803 | + /** | |
| 804 | + * Strip level labels from every heading in an outline. | |
| 805 | + * | |
| 806 | + * @since 2.0.1 | |
| 807 | + * | |
| 808 | + * @param mixed $outline Outline as returned by the model. | |
| 809 | + * @return array Outline with clean headings. | |
| 810 | + */ | |
| 811 | + public static function strip_outline_level_labels($outline): array { | |
| 812 | + if (!is_array($outline)) { | |
| 813 | + return []; | |
| 814 | + } | |
| 815 | + | |
| 816 | + foreach ($outline as $index => $section) { | |
| 817 | + if (is_array($section) && isset($section['heading']) && is_string($section['heading'])) { | |
| 818 | + $outline[$index]['heading'] = self::strip_level_label($section['heading']); | |
| 819 | + } elseif (is_string($section)) { | |
| 820 | + $outline[$index] = self::strip_level_label($section); | |
| 821 | + } | |
| 822 | + } | |
| 823 | + | |
| 824 | + return $outline; | |
| 825 | + } | |
| 826 | + | |
| 827 | + /** | |
| 828 | + * Strip level labels from the heading text inside drafted HTML. | |
| 829 | + * | |
| 830 | + * This is the path that reaches published post content, so it is the one | |
| 831 | + * that matters most. Only the text directly inside an <h1>-<h6> is touched. | |
| 832 | + * | |
| 833 | + * @since 2.0.1 | |
| 834 | + * | |
| 835 | + * @param string $html Drafted article body. | |
| 836 | + * @return string Body with clean headings. | |
| 837 | + */ | |
| 838 | + public static function strip_heading_level_labels(string $html): string { | |
| 839 | + if ('' === $html || false === stripos($html, '<h')) { | |
| 840 | + return $html; | |
| 841 | + } | |
| 842 | + | |
| 843 | + return (string) preg_replace_callback( | |
| 844 | + '/(<h([1-6])\b[^>]*>)(.*?)(<\/h\2>)/is', | |
| 845 | + static function (array $parts): string { | |
| 846 | + return $parts[1] . self::strip_inner_level_label($parts[3]) . $parts[4]; | |
| 847 | + }, | |
| 848 | + $html | |
| 849 | + ); | |
| 850 | + } | |
| 851 | + | |
| 852 | + /** | |
| 684 | 853 | * Create error response when JSON parsing fails |
| 685 | 854 | * |
| 686 | 855 | * @param string $ai_response Raw AI response |
| 687 | 856 | * @param array $original_params Original generation parameters |
| @@ -1144,8 +1313,20 @@ | ||
| 1144 | 1313 | if (false === $result) { |
| 1145 | 1314 | throw new \Exception('Failed to save content brief to database.'); |
| 1146 | 1315 | } |
| 1147 | 1316 | |
| 1317 | + /** | |
| 1318 | + * Fires after a content brief is persisted. | |
| 1319 | + * | |
| 1320 | + * Analytics listens to drop its cached overview so the brief counts | |
| 1321 | + * on the Usages page are not stale for a TTL. | |
| 1322 | + * | |
| 1323 | + * @since 2.2.1 | |
| 1324 | + * | |
| 1325 | + * @param int $brief_id Row id of the stored brief. | |
| 1326 | + */ | |
| 1327 | + do_action('thinkrank_content_brief_created', (int) $wpdb->insert_id); | |
| 1328 | + | |
| 1148 | 1329 | return $wpdb->insert_id; |
| 1149 | 1330 | } |
| 1150 | 1331 | |
| 1151 | 1332 | /** |
| @@ -1187,8 +1368,54 @@ | ||
| 1187 | 1368 | return is_string($rec) ? $rec : 'Image recommendation'; |
| 1188 | 1369 | }, $brief_data['visual_content']['image_recommendations']); |
| 1189 | 1370 | } |
| 1190 | 1371 | |
| 1372 | + return $this->sanitize_brief_output($brief_data); | |
| 1373 | + } | |
| 1374 | + | |
| 1375 | + /** | |
| 1376 | + * Strip untrusted markup out of brief fields before they leave the server. | |
| 1377 | + * | |
| 1378 | + * Brief content crosses a trust boundary: it is assembled by an external AI | |
| 1379 | + * provider from prompts that can include text fetched from competitor URLs. | |
| 1380 | + * It was previously copied out of the decoded JSON verbatim and rendered in | |
| 1381 | + * the admin SPA through dangerouslySetInnerHTML, so a malicious or | |
| 1382 | + * prompt-injected response could execute script in the admin origin (#365). | |
| 1383 | + * | |
| 1384 | + * Runs on the read path as well as generation, so briefs stored before this | |
| 1385 | + * fix are sanitized when they are loaded. | |
| 1386 | + * | |
| 1387 | + * @since 1.32.0 | |
| 1388 | + * | |
| 1389 | + * @param array $brief_data Brief data to sanitize. | |
| 1390 | + * @return array Sanitized brief data. | |
| 1391 | + */ | |
| 1392 | + private function sanitize_brief_output(array $brief_data): array { | |
| 1393 | + foreach ($brief_data as $key => $value) { | |
| 1394 | + // The raw provider response is debug output shown as plain text, and | |
| 1395 | + // the generation params are our own values — leave both intact. | |
| 1396 | + if ('raw_response' === $key || 'generation_params' === $key) { | |
| 1397 | + continue; | |
| 1398 | + } | |
| 1399 | + | |
| 1400 | + if ('content_body' === $key && is_string($value)) { | |
| 1401 | + // Deliberately HTML: it is the drafted article and is rendered as | |
| 1402 | + // markup. wp_kses_post() keeps normal post formatting while | |
| 1403 | + // dropping script/style/iframe, event-handler attributes and | |
| 1404 | + // javascript: URLs. | |
| 1405 | + $brief_data[$key] = wp_kses_post($value); | |
| 1406 | + continue; | |
| 1407 | + } | |
| 1408 | + | |
| 1409 | + if (is_array($value)) { | |
| 1410 | + $brief_data[$key] = $this->sanitize_brief_output($value); | |
| 1411 | + } elseif (is_string($value)) { | |
| 1412 | + // Every other field is plain text (headings, keywords, guidance). | |
| 1413 | + // Markdown emphasis markers are preserved; HTML tags are not. | |
| 1414 | + $brief_data[$key] = wp_strip_all_tags($value); | |
| 1415 | + } | |
| 1416 | + } | |
| 1417 | + | |
| 1191 | 1418 | return $brief_data; |
| 1192 | 1419 | } |
| 1193 | 1420 | |
| 1194 | 1421 | /** |
| @@ -1821,9 +2048,9 @@ | ||
| 1821 | 2048 | [ |
| 1822 | 2049 | 'user_id' => $user_id, |
| 1823 | 2050 | 'action' => $action, |
| 1824 | 2051 | 'tokens_used' => $tokens_used, |
| 1825 | - 'provider' => $this->settings->get('ai_provider', 'openai'), | |
| 2052 | + 'provider' => $this->settings->get('ai_provider', Settings::AI_PROVIDER_NONE), | |
| 1826 | 2053 | 'post_id' => $post_id, |
| 1827 | 2054 | 'metadata' => !empty($metadata) ? wp_json_encode($metadata) : null, |
| 1828 | 2055 | 'created_at' => current_time('mysql'), |
| 1829 | 2056 | ], |
| @@ -1828,8 +2055,17 @@ | ||
| 1828 | 2055 | 'created_at' => current_time('mysql'), |
| 1829 | 2056 | ], |
| 1830 | 2057 | ['%d', '%s', '%d', '%s', '%d', '%s', '%s'] |
| 1831 | 2058 | ); |
| 2059 | + | |
| 2060 | + /** | |
| 2061 | + * Fires after an AI usage row is recorded. | |
| 2062 | + * | |
| 2063 | + * @since 2.2.1 | |
| 2064 | + * | |
| 2065 | + * @param int $user_id User the usage was recorded against. | |
| 2066 | + */ | |
| 2067 | + do_action('thinkrank_ai_usage_logged', $user_id); | |
| 1832 | 2068 | |
| 1833 | 2069 | return $wpdb->insert_id; |
| 1834 | 2070 | } |
| 1835 | 2071 | |
