DeepWrite for Official Documents

The Specific Problem: Challenges in Formal Document Writing

Creating standardized official documents, formal reports, and speeches is more than text compilation. It involves several concrete and tedious challenges:
* Strict Formatting and Norms: Different document types (notices, requests for instructions, minutes) have specific structures, terminologies, and layout requirements.
* High Demand for Substantiated Content: Policy bases, data support, and case references require authoritative sources; fabrication from memory is unacceptable.
* Uncertain Workflow: The process from drafting to finalization may require outline confirmation, multiple revisions, and fact-checking, with the workflow varying based on material complexity.
* Standardized Deliverables: Typically requires generation of a .docx document with fixed formatting; complex scenarios may also need traceable auxiliary reports.

How "DeepWrite for Official Documents" Works

This skill is not a fixed script but an intelligent writing framework that dynamically combines multiple modes based on task complexity. Its core workflow is as follows:

1. Intelligent Task Routing and Process Selection
When a user presents a writing need (e.g., "write an annual work summary"), the skill first assesses the task type and complexity.
* Simple tasks (e.g., an internal notice): May skip search and generate a Word document directly based on built-in standards.
* Complex tasks (e.g., a research report): Enter a more rigorous process, potentially including outline confirmation, material retrieval, and multiple review cycles.

2. Five Core Capability Modes
The skill combines the following modes to handle different stages:
* Tool Wrapper: Encapsulates tool chains for official document outline generation, DeepWrite Trustworthy Search, Word formatting, red-header document generation, etc.
* Generator: Generates content conforming to requirements based on document type standards, user materials, and retrieved materials.
* Reviewer: Automatically checks formatting, logic, source credibility, and potential risks according to a review checklist.
* Inversion: Actively asks the user for clarification when information is insufficient or the task is complex.
* Pipeline: Executes a strict workflow with checkpoints for materials requiring policy bases, long-form complexity, or red-header delivery.

3. Key Steps: On-Demand Search and Trustworthy Tracing
* On-Demand Activation: The DeepWrite Trustworthy Search function (configured via the DKNOWC_API_KEY environment variable) is only activated and guided when the task requires policy bases, authoritative data, or case references. Simple writing tasks that do not require search proceed without configuration.
* Scheme-First Approach: Before searching, the skill designs a search plan with multiple retrieval items and only executes serial searches after user confirmation.
* Standardized Delivery: The primary deliverable is a .docx Word document. If searches were performed, an additional Trustworthy Traceability Report HTML file is generated, converting citation markers like [1] in the text into clickable source links to ensure verifiability.

Applicable Boundaries and Important Notes

• Task Scope: Focused on official documents, transactional documents, briefing materials, and similar scenarios. Simple short texts (e.g., a few-line notice) can also be handled, but the deliverable remains a Word document.
• Deliverables: Defaults to a clean Word document (text does not include source annotations). A red-header formatted version is generated only when explicitly requested by the user. The current version does not support automatic PDF output.
• Personal Memory: The skill can maintain a local personal material library and writing preferences. User's private materials are used only for the current session by default; they are only saved to the local material library upon explicit confirmation and are never uploaded.
• Human Oversight Necessity: While the skill assists with formatting, retrieval, and generation, the final accuracy and policy applicability of the material must still be verified and confirmed by the user according to actual circumstances, especially for external materials obtained through search.