with▌

Persistent markdown-based planning system for organizing complex multi-step tasks and research projects. \n \n Creates three working files ( task_plan.md , findings.md , progress.md ) in your project directory to serve as persistent \"working memory\" across agent commands \n Enforces the 2-Action Rule: save key findings to disk immediately after viewing/searching to prevent multimodal information loss \n Includes a 3-Strike Error Protocol for systematic debugging: diagnose and fix on attempt 1,

像 Manus 一樣工作：用持久化的 Markdown 檔案作為你的「磁碟工作記憶」。

Work like Manus: Use persistent markdown files as your "working memory on disk."

Comprehensive guide for Medusa backend architecture, workflows, and critical implementation rules. \n \n Covers six rule categories (architecture, type safety, business logic placement, imports, data access, file organization) with specific anti-patterns and enforcement checks \n Enforces strict layer separation: Module → Workflow → API Route → Frontend, with workflows required for all mutations and only GET/POST/DELETE HTTP methods allowed \n Includes critical data handling rules: prices stored

Persistent file-based planning system for organizing complex multi-step tasks and maintaining context across sessions. \n \n Creates three markdown files (task_plan.md, findings.md, progress.md) in your project directory to serve as persistent working memory \n Includes session recovery via script to sync context after /clear commands and resume interrupted work \n Enforces structured workflows: plan first, update after each phase, log all errors, and never repeat failed actions \n Provides temp

SQLite and offline-first data layer architecture for Flutter apps with structured local persistence. \n \n Separates data layer into stateless Services (raw data wrappers) and business-logic Repositories (single source of truth per domain entity) \n Implements offline-first synchronization: reads yield cached local data immediately then fetch remote updates; writes save locally first, then attempt API sync with background queue fallback \n Covers SQLite setup with safe parameterized queries, Dom

Browser automation with Playwright MCP for web navigation, form interaction, and data extraction. \n \n Supports navigation, element interaction (click, type, select), screenshots, and accessibility snapshots that return element references for precise targeting \n Includes JavaScript execution via browser_evaluate and multi-step atomic operations through browser_run_code for complex workflows \n Requires --shared-browser-context flag to maintain browser state across sequential commands; server r

File-based task organization and progress tracking for complex multi-step projects. \n \n Creates and maintains three persistent markdown files (task_plan.md, findings.md, progress.md) to preserve context across sessions and tool calls \n Automatically detects active plans on startup and prompts context recovery; supports session restoration after /clear via catchup script \n Enforces structured workflow: plan phases upfront, log discoveries and errors immediately, update progress after each act

Clone real voices from online video or design voices from photos, then roleplay as that person with synthetic speech. \n \n Two workflows: extract voice from public video (interviews, speeches) by name, or generate a matching voice from an uploaded image of an unrecognizable person \n Requires ffmpeg , yt-dlp , the tts skill, and a Noiz API key; includes setup verification and dependency installation steps \n Built-in ethical guardrails: agent must refuse requests targeting non-consenting privat

$23