@biaoo/tiangong-wiki 0.3.8 → 0.3.10

package/README.md

@@ -117,6 +117,8 @@ tiangong-wiki dashboard                               # open dashboard in browse
 
 > Environment variables are managed via `.wiki.env` (created by `tiangong-wiki setup`). The CLI prefers the nearest local `.wiki.env`, then falls back to the global default workspace config. See [references/troubleshooting.md](./references/troubleshooting.md) for the full reference. For a centralized Linux + `systemd` + Nginx deployment, see [references/centralized-service-deployment.md](./references/centralized-service-deployment.md). That deployment guide now also includes Git repository / GitHub remote setup for daemon-side commit and optional auto-push.
 
+For document-heavy vaults, set `WIKI_PARSER_SKILLS=document-granular-decompose` and configure `UNSTRUCTURED_API_BASE_URL` plus `UNSTRUCTURED_AUTH_TOKEN` to prefer the TianGong Unstructure parser. `document-granular-decompose` is a broad document/image parser for PDF, Word, PowerPoint, Excel, Markdown, and common image formats; the skill's `SKILL.md` remains the source of truth for the exact extension allowlist. The wiki workflow uses `return_txt=true` and consumes the plain `txt` text as the agent input; `UNSTRUCTURED_PROVIDER` and `UNSTRUCTURED_MODEL` are optional overrides.
+
 ## MCP Server
 
 Tiangong Wiki ships a separate MCP adapter that talks to the daemon over HTTP. It uses the MCP Streamable HTTP transport, not stdio.

package/README.zh-CN.md

@@ -117,6 +117,8 @@ tiangong-wiki dashboard                               # 在浏览器中打开仪
 
 > 环境变量通过 `.wiki.env` 管理（由 `tiangong-wiki setup` 创建）。CLI 会优先使用最近的本地 `.wiki.env`，找不到时再 fallback 到全局默认工作区配置。完整参考见 [references/troubleshooting.md](./references/troubleshooting.md)。如需部署中心化服务（Linux + `systemd` + Nginx），见 [references/centralized-service-deployment.md](./references/centralized-service-deployment.md)。该部署文档现在也包含了 Git 仓库初始化、GitHub remote 配置和 daemon 自动 push 的 Git 配置说明。
 
+如果 vault 里以文档解析为主，可设置 `WIKI_PARSER_SKILLS=document-granular-decompose`，并配置 `UNSTRUCTURED_API_BASE_URL` 与 `UNSTRUCTURED_AUTH_TOKEN`，让 workflow 优先使用 TianGong Unstructure parser。`document-granular-decompose` 是覆盖范围更宽的文档/图片解析器，适用于 PDF、Word、PowerPoint、Excel、Markdown 和常见图片格式；精确扩展名 allowlist 以该 skill 自身的 `SKILL.md` 为准。wiki workflow 会使用 `return_txt=true`，并把返回的纯 `txt` 文本作为 agent 主输入；`UNSTRUCTURED_PROVIDER` 和 `UNSTRUCTURED_MODEL` 只是可选覆盖项。
+
 ## MCP Server
 
 Tiangong Wiki 提供了独立的 MCP 适配层，通过 HTTP 调用 daemon。它使用的是 MCP 的 Streamable HTTP 传输，不是 stdio。

package/dist/core/workflow-context.js

@@ -83,6 +83,8 @@ export function buildVaultWorkflowPrompt(input) {
         "",
         "1. Read queue-item.json next to RESULT_JSON_PATH.",
         "2. Read the target vault file at VAULT_FILE_PATH. Refer to `references/vault-to-wiki-instruction.md` (Phase 1) in the wiki package for file-type-specific reading strategies, parser skill discovery, image handling, and metadata utilization.",
+        "   - If `WIKI_PARSER_SKILLS` includes `document-granular-decompose` and `UNSTRUCTURED_API_BASE_URL` plus `UNSTRUCTURED_AUTH_TOKEN` are set, prefer that skill for supported document/image files before the legacy type-specific parser skills. This includes PDF, Word, PowerPoint, Excel, Markdown, and common image formats; use the skill's `SKILL.md` for the exact extension allowlist.",
+        "   - When using `document-granular-decompose`, request `return_txt=true`, treat the pure text extracted from `response.txt`/`txt` as the main input, and keep raw JSON only for debugging or page-number evidence.",
         "3. Discover the current page type ontology via `tiangong-wiki type list` and `tiangong-wiki type show <type>`. Do not assume any type, template, or default target type.",
         "4. Search the existing wiki for overlapping or related content:",
         "   - Use `tiangong-wiki fts` and `tiangong-wiki search` with key terms from the source.",

package/dist/core/workspace-skills.js

@@ -8,14 +8,21 @@ import { copyDirectoryContentsSync, ensureDirSync, pathExistsSync, readTextFileS
 import { resolveRuntimePaths } from "./paths.js";
 import { toOffsetIso } from "../utils/time.js";
 export const PARSER_SKILL_SOURCE = "https://github.com/anthropics/skills";
+export const TIANGONG_PARSER_SKILL_SOURCE = "https://github.com/tiangong-ai/skills";
 export const MANAGED_SKILL_STATE_FILE = ".tiangong-wiki-skill.json";
 export const OPTIONAL_PARSER_SKILLS = [
-    { name: "pdf", summary: "Process PDF files" },
-    { name: "docx", summary: "Process DOCX files" },
-    { name: "pptx", summary: "Process PPTX files" },
-    { name: "xlsx", summary: "Process XLSX/CSV files" },
+    { name: "pdf", summary: "Process PDF files", source: PARSER_SKILL_SOURCE },
+    { name: "docx", summary: "Process DOCX files", source: PARSER_SKILL_SOURCE },
+    { name: "pptx", summary: "Process PPTX files", source: PARSER_SKILL_SOURCE },
+    { name: "xlsx", summary: "Process XLSX/CSV files", source: PARSER_SKILL_SOURCE },
+    {
+        name: "document-granular-decompose",
+        summary: "Process documents through TianGong Unstructure and return plain fulltext",
+        source: TIANGONG_PARSER_SKILL_SOURCE,
+    },
 ];
 const OPTIONAL_PARSER_SKILL_NAMES = new Set(OPTIONAL_PARSER_SKILLS.map((skill) => skill.name));
+const OPTIONAL_PARSER_SKILL_BY_NAME = new Map(OPTIONAL_PARSER_SKILLS.map((skill) => [skill.name, skill]));
 const MANAGED_SKILL_SOURCE_KINDS = new Set([
     "workspace-package",
     "curated-parser",
@@ -228,11 +235,15 @@ function normalizeManagedSource(rawSource) {
     return source;
 }
 function createParserDescriptor(workspaceRoot, name, configured) {
+    const definition = OPTIONAL_PARSER_SKILL_BY_NAME.get(name);
+    if (!definition) {
+        throw new AppError(`Unsupported parser skill: ${name}`, "config");
+    }
     return {
         name,
         sourceKind: "curated-parser",
         configured,
-        source: PARSER_SKILL_SOURCE,
+        source: definition.source,
         skillPath: resolveWorkspaceSkillPath(workspaceRoot, name),
     };
 }
@@ -656,7 +667,11 @@ export function ensureWikiSkillInstall(wikiPath, packageRoot) {
     };
 }
 export function buildParserSkillInstallInvocation(skillName) {
-    return buildExternalSkillInstallInvocation(PARSER_SKILL_SOURCE, skillName);
+    const definition = OPTIONAL_PARSER_SKILL_BY_NAME.get(skillName);
+    if (!definition) {
+        throw new AppError(`Unsupported parser skill: ${skillName}`, "config");
+    }
+    return buildExternalSkillInstallInvocation(definition.source, skillName);
 }
 export function installParserSkill(skillName, workspaceRoot, options = {}) {
     const installed = installManagedExternalSkill(createParserDescriptor(workspaceRoot, skillName, true), {