This entry is retired and its content has moved to dsh-doc-tidy, so it should not be loaded.
guzhou079-arch/deepseek-harness-android
Bundles a set of ready-to-use skills for running an AI agent on an Android phone, covering documents, data, writing, research and phone automation.
- 27 skills
- 110 stars
The 5 most installed
Backs up the whole system, memory and sessions, lists snapshots, and restores a chosen backup or moves it to another device.
Makes the agent talk and work in a restrained Claude-like style, asking before acting and verifying its own work.
Reviews code across security, concurrency, memory and compatibility, generates refactoring diffs, and writes unit tests with mocks.
Splits a large job into roles and a team structure, and sets up sub-agents on the phone to carry it out.
The other 22 skills
Cleans and pivots data, aggregates metrics, finds outliers, and draws charts inside the chat.
Researches technical options, compares competitors across many dimensions, and drafts architecture proposals and decision records.
Turns code, system architecture or business logic into clear diagrams rendered right in the chat.
Generates properly formatted Word .docx documents with headings, numbering, tables and margins that open in WPS or Office.
Indexes local code, docs, configs and logs on the phone and answers questions about them quickly.
Reads, summarizes and renames Excel, Word and PowerPoint files without opening WPS, using built-in tools only.
Aligns messy multi-sheet Excel tables, merges differing fields, inserts formulas and exports a clean .xlsx file.
Browses Taobao Flash Sale in a virtual screen, recommends shops and items, and fills an order up to checkout for you to pay.
Extracts and copies verification codes, logs payments from WeChat, Alipay and banks, gathers parcel codes, and manages scheduled tasks.
Polishes work emails in different tones, reviews long email threads, and drafts structured replies.
Turns long meeting transcripts into a short summary, a view-by-view table and a to-do list with owners and deadlines.
Saves key facts, preferences and lessons from chats into a local memory database and recalls them in later sessions.
Controls apps on the Android phone by reading the screen, tapping, typing, gestures, files, clipboard and notifications.
Has two or more models from different vendors review the same output independently, then merges agreement and disagreement.
Breaks goals into estimated task cards, draws Gantt charts and critical paths, and writes weekly project reports with risks.
Generates editable PowerPoint .pptx presentations from an outline or brief, ready to open in WPS or PowerPoint.
Writes short-video storyboards, narration and visual direction in a paper-collage, hand-drawn or minimal style.
Sets the phone model's thinking level to low, high or max and switches it automatically based on task difficulty.
Turns photos, clips and audio into a finished mp4 with transitions, subtitles, music, narration and a cover frame.
Handles voice on the phone: speech-to-text input, voice state listening, and spoken replies through text-to-speech.
Scrapes public web pages and logged-in app data, cleaning results into structured JSON or Markdown.
Writes stories, articles, reports and emails in a more human style, with a script that flags AI-sounding text.