# AI-Assisted Development NeuG provides AI-assisted skills to streamline your development workflow. These skills are integrated with your coding agent (Cursor, Qoder, Qwen Code, etc.) as agent skills. You can use slash commands to invoke them. ## Quick Start **1. Initialize skills for your coding agent:** ```bash ./scripts/init_skills.sh # e.g., qoder, qwen ``` > **Note:** Cursor is the default agent and does not require initialization. See [Supported Agents](#supported-agents) for full list. **2. Use skills in your chat box with a `/` prefix:** ``` /create-issue I encountered a segfault when running this query... ``` ## Common Skills ### `/create-issue` - Report Bugs or Request Features Create GitHub issues directly from your IDE with full context. **Example: Report a Bug** ``` /create-issue Type: Bug Assignee: @who Parent Issue: #42 The query fails with segfault. Terminal output: E20260112 10:23:45.123456 12345 executor.cc:271] Query execution failed Segmentation fault (core dumped) ``` **Example: Request a Feature** ``` /create-issue Type: Feature Assignee: @me I need support for OPTIONAL MATCH with multiple patterns. Currently only single pattern is supported. ``` **Tips:** - Copy-paste terminal error output directly - the agent will extract key information - Specify `Type: Bug` or `Type: Feature` to skip auto-detection - Use `Assignee: @me` or `Assignee: @username` to pre-assign - Use `Parent Issue: #id` to link as sub-issue ### `/create-pr` - Submit Pull Requests Create PRs with auto-generated summaries linked to related issues. **Example** ``` /create-pr Fixes: #42 Reviewers: @who This fixes the segfault issue in query execution. Added null pointer check before accessing result buffer. ``` **Tips:** - Use `Fixes: #id` to auto-close the issue when PR merges - Use `Reviewers: @user1, @user2` to request specific reviewers - The agent auto-generates PR title and summary from your changes ### `/update-with-comments` - Address PR Feedback Automatically fetch and apply PR review comments. **Example** ``` /update-with-comments ``` Or specify a PR: ``` /update-with-comments #123 ``` The agent will: 1. Fetch all review comments from the PR 2. Apply requested changes to your code 3. Show a summary of modifications 4. Commit and push after your review ## Spec-Driven Workflow For larger features, we provide a structured workflow to plan before implementing. ### Step 1: `/speckit.specify` - Define the Feature Create a feature specification from natural language. **Example** ``` /speckit.specify Add graph algorithm extension support for NeuG. Users should be able to register custom algorithms and execute them via Cypher. ``` The agent will: 1. Create a feature branch (e.g., `001-graph-algo-extension`) 2. Generate a specification document in `specs/001-graph-algo-extension/spec.md` 3. Create a GitHub issue to track the feature 4. Ask clarifying questions if needed (max 3) ### Step 2: `/speckit.plan` - Plan the Implementation Generate a technical plan from the specification. **Example** ``` /speckit.plan 001-graph-algo-extension ``` The agent will: 1. Analyze the spec and codebase structure 2. Plan project structure, data models, and algorithms 3. Generate `plan.md` with implementation details 4. Create a linked GitHub issue ### Step 3: `/speckit.tasks` - Break Down into Tasks Generate actionable tasks from the plan. **Example** ``` /speckit.tasks 001-graph-algo-extension ``` The agent will: 1. Create module-level breakdown 2. Generate task files for each module 3. Each task is sized for independent testing ### Sync to GitHub After creating specs/plans/tasks, sync them to GitHub: ``` /sync-modules 001-graph-algo-extension # Sync module issues /sync-tasks 001-graph-algo-extension # Sync task issues ``` ## Setup ### Prerequisites Agent skills are implemented by markdown files and can be invoked by coding agent automatically or user explicitly. To avoid accidental invocation, we set `disable-model-invocation: true` for all skills by default. You can only invoke them by slash commands. You can also customize the skills files to enable model invocation. ### Installing Skills We provide agent skills in Cursor format at `.cursor/skills/`. To convert for your agent: ```bash ./scripts/init_skills.sh ``` ### Supported Agents | Shortcut | Agent | Documents | |------------|---------------------|----------------| | cursor | Cursor (default) | [Cursor Skills](https://cursor.com/docs/context/skills) | | claude | Claude Code | [Claude Code Skills](https://code.claude.com/docs/en/skills) | | codebuddy | CodeBuddy | [CodeBuddy Skills](https://www.codebuddy.ai/docs/ide/Features/Skills) | | codex | Codex | [Codex Skills](https://developers.openai.com/codex/skills/) | | gemini | Gemini CLI | [Gemini CLI Skills](https://geminicli.com/docs/cli/skills/) | | kilocode | Kilocode | [Kilocode Skills](https://kilo.ai/docs/customize/skills) | | opencode | OpenCode | [OpenCode Skills](https://opencode.ai/docs/skills/) | | qoder | Qoder | [Qoder Skills](https://docs.qoder.com/extensions/skills) | | qwen | Qwen Code | [Qwen Code Skills](https://qwenlm.github.io/qwen-code-docs/en/users/features/skills/) | | roo | RooCode | [RooCode Skills](https://docs.roocode.com/features/skills) | | windsurf | Windsurf | [Windsurf Skills](https://docs.windsurf.com/windsurf/cascade/skills) | Note: Different agents might have minor differences in the Agent Skills standard. You can refer to the documents for more details. **Example:** ```bash ./scripts/init_skills.sh qoder ``` Or manually specify output path: ```bash ./scripts/init_skills.sh --output=.qwen/skills ``` ## Skills Reference | Skills | Purpose | |---------|---------| | `/create-issue` | Create bug reports or feature requests | | `/create-pr` | Submit pull requests with auto-summary | | `/update-with-comments` | Apply PR review feedback | | `/speckit.specify` | Define feature specifications | | `/speckit.plan` | Create implementation plans | | `/speckit.tasks` | Break plans into tasks | | `/sync-modules` | Sync module issues to GitHub | | `/sync-tasks` | Sync task issues to GitHub | | `/generate_testcase` | Generate test cases |