From 037a963547df6973ed418b332800dfe5c25444eb Mon Sep 17 00:00:00 2001 From: cflmy Date: Tue, 27 Oct 2026 14:14:17 +0800 Subject: [PATCH] feat(workbench): emit blob_ref for PDF and image ResourceView Client visual preview consumes the blob; extract/OCR capabilities stay false so QA-10 honesty holds without cutting the preview path. Co-authored-by: Cursor --- src/workbench/resource_view.rs | 192 ++++++++++++++++++++++++++++++++- 1 file changed, 191 insertions(+), 1 deletion(-) diff --git a/src/workbench/resource_view.rs b/src/workbench/resource_view.rs index 5bf6f79..95ab022 100644 --- a/src/workbench/resource_view.rs +++ b/src/workbench/resource_view.rs @@ -124,6 +124,13 @@ pub fn build_resource_view(req: &ResourceViewRequest) -> Result { })); } + // PDF / image: framework supplies authorized blob_ref for client visual + // preview (PDF.js / ImagePreview). Extract and OCR stay independently + // unavailable — never claim text was extracted (F16 / QA-10). + if let Some(view) = binary_blob_view(&rel, media, kind, &digest, &title, size) { + return Ok(view); + } + let truncated = size > req.max_bytes; let slice = if truncated { &raw[..req.max_bytes] @@ -163,7 +170,8 @@ pub fn build_resource_view(req: &ResourceViewRequest) -> Result { "completeness": "partial", "no_exec": true, "diagnostics": [{"code": "unsupported_format", "message": other}], - "next_cursor": Value::Null + "next_cursor": Value::Null, + "capabilities": {"read_only": true, "executed": false, "kind": kind, "preview": false, "extract": false} })); } }; @@ -218,6 +226,22 @@ fn media_type_for(path: &str) -> &'static str { "application/json" } else if lower.ends_with(".csv") { "text/csv" + } else if lower.ends_with(".pdf") { + "application/pdf" + } else if lower.ends_with(".png") { + "image/png" + } else if lower.ends_with(".jpg") || lower.ends_with(".jpeg") { + "image/jpeg" + } else if lower.ends_with(".webp") { + "image/webp" + } else if lower.ends_with(".gif") { + "image/gif" + } else if lower.ends_with(".docx") { + "application/vnd.openxmlformats-officedocument.wordprocessingml.document" + } else if lower.ends_with(".xlsx") { + "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet" + } else if lower.ends_with(".pptx") { + "application/vnd.openxmlformats-officedocument.presentationml.presentation" } else if lower.ends_with(".rs") || lower.ends_with(".ts") || lower.ends_with(".tsx") @@ -249,6 +273,16 @@ fn kind_for(path: &str, media: &str) -> &'static str { "json" } else if media == "text/csv" { "csv" + } else if media == "application/pdf" { + "pdf" + } else if media.starts_with("image/") { + "image" + } else if media.contains("wordprocessingml") { + "docx" + } else if media.contains("spreadsheetml") { + "xlsx" + } else if media.contains("presentationml") { + "pptx" } else if media == "text/x-code" { "code" } else if media.starts_with("text/") { @@ -258,6 +292,109 @@ fn kind_for(path: &str, media: &str) -> &'static str { } } +/// PDF/image → `blob_ref` for client visual preview; Office stays conversion-gated. +/// Extract/OCR capabilities remain false and must not be implied by a visible page. +fn binary_blob_view( + rel: &str, + media: &str, + kind: &str, + digest: &str, + title: &str, + size: usize, +) -> Option { + match kind { + "pdf" | "image" => { + let (renderer_id, diagnostics) = if kind == "pdf" { + ( + "pdf_blob", + vec![ + json!({"code": "extract.unavailable", "message": "pdf text extract unavailable"}), + json!({"code": "ocr.unavailable", "message": "ocr not claimed; visual page ≠ extracted text"}), + ], + ) + } else { + ( + "image_blob", + vec![ + json!({"code": "extract.unavailable", "message": "image/ocr extract unavailable"}), + json!({"code": "vision.unclaimed", "message": "image preview ≠ model vision support"}), + ], + ) + }; + Some(json!({ + "schema": "ResourceViewV1", + "schema_version": 1, + "resource": resource_ref(rel, media, digest, title, size), + "renderer": {"id": renderer_id, "version": "1", "trust": "blob"}, + "render": { + "format": "blob_ref", + "fragment": Value::Null, + "text": Value::Null, + "blob_ref": { + "path": rel, + "digest": digest, + "media_type": media, + "byte_length": size, + "title": title + } + }, + "outline": [], + "locators": if kind == "pdf" { + json!([{"id": "page-1", "kind": "page", "locator": {"page": 1}}]) + } else { + json!([]) + }, + "completeness": "partial", + "no_exec": true, + "diagnostics": diagnostics, + "next_cursor": Value::Null, + "capabilities": { + "read_only": true, + "executed": false, + "kind": kind, + "visual_preview": "client", + "extract": false, + "ocr": false + } + })) + } + "docx" | "xlsx" | "pptx" => { + let text = format!( + "Office 文档「{title}」({media}, {size} bytes)\n\n转换插件未安装;不会伪造预览或提取结果。可下载原件。能力闭环仍要求转换插件落地后按派生格式预览,本卡不是终态替代方案。" + ); + Some(json!({ + "schema": "ResourceViewV1", + "schema_version": 1, + "resource": resource_ref(rel, media, digest, title, size), + "renderer": {"id": "office_conversion_pending", "version": "1", "trust": "none"}, + "render": { + "format": "sanitized_html", + "fragment": wrap_pre(&escape_html(&text)) + }, + "outline": [], + "locators": [], + "completeness": "partial", + "no_exec": true, + "diagnostics": [ + {"code": "conversion.unavailable", "message": "office conversion plugin not installed"}, + {"code": "extract.unavailable", "message": format!("{kind} extract unavailable")} + ], + "next_cursor": Value::Null, + "capabilities": { + "read_only": true, + "executed": false, + "kind": kind, + "visual_preview": false, + "extract": false, + "ocr": false, + "conversion": false + } + })) + } + _ => None, + } +} + const JSON_OUTLINE_MAX: usize = 40; const CSV_MAX_ROWS: usize = 200; const CSV_MAX_COLS: usize = 32; @@ -677,4 +814,57 @@ mod tests { assert!(!frag.contains(" = view["diagnostics"] + .as_array() + .unwrap() + .iter() + .filter_map(|d| d["code"].as_str()) + .collect(); + assert!(codes.contains(&"ocr.unavailable")); + assert!(codes.contains(&"extract.unavailable")); + assert!(!codes.iter().any(|c| c.contains("preview.unavailable"))); + let _ = fs::remove_dir_all(&dir); + } + + #[test] + fn image_blob_ref_for_client_preview() { + let dir = scratch_dir("img"); + let path = dir.join("pic.png"); + fs::write(&path, b"\x89PNG\r\n\x1a\n").unwrap(); + let req = ResourceViewRequest { + workspace_root: dir.clone(), + path: "pic.png".into(), + revision: None, + max_bytes: 10_000, + }; + let view = build_resource_view(&req).unwrap(); + assert_eq!(view["renderer"]["id"], "image_blob"); + assert_eq!(view["render"]["format"], "blob_ref"); + assert_eq!(view["resource"]["kind"], "image"); + assert_eq!(view["capabilities"]["visual_preview"], "client"); + assert_eq!(view["capabilities"]["extract"], false); + let _ = fs::remove_dir_all(&dir); + } }