Cursor workflow with persistent memory.
Your go-to toolkit for creators in the AI era.
Persistent memory so the AI works the way you do.
Respect AI capability and leave room for evolution.
Lower cognitive load and a smooth, user-friendly experience.
- Flexible workflow: Compose your own flow—rigorous when needed, light when not
- Persistent memory bank: Learns your judgment, taste, and responsibility in the project and applies them in every new conversation
- Human in the loop: Key decisions follow your lead—optional when you want, never overstepping when you don’t
- Self-iterate: Run low-risk, audit-driven improvements on heartbeat cycles so the system keeps getting more stable and accurate
- Ready to use: Use the install script to add LingXi to your project, then run
/initto understand project context and bootstrap LingXi workflow
Run one of the following commands from your project root to install LingXi into your project:
- Linux / macOS / Git Bash:
curl -fsSL https://raw.githubusercontent.com/tower1229/LingXi/main/install/bash.sh | bash - Windows PowerShell:
irm https://raw.githubusercontent.com/tower1229/LingXi/main/install/powershell.ps1 | iex
After install, open the project in Cursor and run /init once to build project context and generate optional memory candidates (write is gated by your explicit choice).
We recommend running /init first to understand the existing project and prepare optional memory candidates; then use the workflow skills or helper commands below.
The core workflow is driven by Skills (task, vet, plan, build, review). Invoke them by typing /task, /plan, /build, /review, or /vet (Cursor shows the matching skill) or by natural language (e.g. “create a task document”, “plan task 001”).
Use these in lifecycle order:
| Skill | Usage | Description |
|---|---|---|
| task | /task <description> or “create task…”Examples: /task Add user login with email and phone/task Improve homepage load, target LCP < 1s |
Create task doc Auto task id (001, 002...) and title; creates: .cursor/.lingxi/tasks/001.task.<title>.mdThis is the core document for the whole workflow, including refined requirements, technical approach, and acceptance criteria. |
| vet | /vet [taskId] or “review task doc”Examples: /vet 001/vet (latest task) |
Review task doc (optional) Multi-dimension review of the task doc to improve quality. Optional, can be run multiple times. No file output; results and suggestions in chat only. |
| plan | /plan [taskId] or “plan task…”Examples: /plan 001/plan (latest task) |
Task planning (optional) Generate plan and test-case docs from the task doc. For complex tasks; simple ones can skip. Tip: Works with Cursor’s plan mode. |
| build | /build [taskId] or “implement task…”Examples: /build 001/build (latest task) |
Run build (optional) Two modes: - Plan-driven: Follow plan when present (recommended) - Task-driven: Agent decides from task doc when no plan Tip: In plan mode you can use its built-in build and skip LingXi’s build skill. |
| review | /review [taskId] or “review delivery”Examples: /review 001/review (latest task) |
Review delivery Multi-dimension review and report: Core: functionality, test coverage, architecture, maintainability, regression Optional: doc consistency, security, performance, E2E Tests: unit, integration, E2E when applicable |
| Command | Usage | Description |
|---|---|---|
/remember |
/remember <description>Examples: /remember Capture the lesson from that bug/remember Always use X for Y |
Write to memory (any time) No task id needed. Write judgments, tradeoffs, runbooks, or checks to memory/project/ (or memory/share/ for team-level) for later retrieval.Use when: - Stating a principle or decision - Extracting from recent conversation - Giving keywords so the system can find and extract the right content Session distillation is automatic: when you start a new conversation and it has been more than 30 minutes since the last run, LingXi enqueues up to 3 finished sessions for background distillation (source=heartbeat); no command needed. |
/init |
/init |
Initialize project context (first use) Guided understanding of an existing project, producing memory candidates with explicit write gating. Internal workspace bootstrap may run as a preflight step. Recommended when first using LingXi in a project. |
LingXi uses a designated share directory for team knowledge that can be reused across projects:
- Share directory:
.cursor/.lingxi/memory/share/(recommended as a git submodule). Team-level memory (apply=team) is written here; project-level memory is written tomemory/project/.
1) Add share repo (submodule)
git submodule add <shareRepoUrl> .cursor/.lingxi/memory/share2) Update share repo
git submodule update --remote --merge3) Sync memory index (after adding shared notes)
In Cursor, run the memory-govern skill (e.g. type /memory-govern) to sync INDEX with notes (removes orphan index rows and lets the model complete INDEX lines for unindexed notes).