TRAE Skills from Claude Code Plugins

中文说明 / Chinese Documentation

A collection of skills adapted for TRAE_CN SOLO, sourced from well-known Claude Code plugins (superpowers, oh-my-claudecode, get-shit-done, gstack) and custom-assembled pipelines.

TRAE_CN is a free vibe coding application that provides powerful models for free and supports Rules & Skills. However, TRAE_CN does not provide users with general, powerful skills to fully leverage model capabilities and lower the usage barrier. Therefore, I selected the most useful skills from well-known Claude Code plugins (superpowers, oh-my-claudecode, get-shit-done, gstack), adapted them for TRAE_CN SOLO, and also assembled a custom AI pipeline skill set. You can download and import them yourself.

---

Directory Structure

├── ai-pipeline（一个使用Ralph保持运行的持续开发循环）/
├── frontend（前端及幻灯片）/
├── GSD技能移植/
├── gstack技能移植/
├── oh-my-claudecode技能移植/
├── ppt-master（AI生成PPT）/
├── superpowers技能移植/
└── taste-skill（反AI俗套前端设计）/

---

ai-pipeline (A Continuous Development Loop Using Ralph)

A custom-assembled automated engineering pipeline from idea to production-grade code. The pipeline runs in a goal-driven loop until acceptance criteria are met.

Skill	Description
ai-pipeline	Full AI engineering pipeline orchestrator: blueprint → ADR → eval-harness → self-improve → slop-cleaner, loops until goal is met
ai-slop-cleaner	AI-generated code cleaner: regression-safe deletion-first workflow that removes dead code, merges duplicates, reduces complexity without changing behavior
architecture-decision-records	Architecture Decision Records (ADR): auto-captures architectural decisions as structured documents with context, alternatives, and rationale
blueprint	Blueprint generator: turns a one-line objective into a step-by-step construction plan with self-contained context briefs, dependency graphs, and parallel step detection
eval-harness	Evaluation framework: implements eval-driven development (EDD) with pass/fail criteria, pass@k reliability metrics, and regression test suites
external-research	External research: searches GitHub, arXiv, technical blogs and forums when optimization plateaus, injecting actionable improvement ideas
self-improve	Self-improvement engine: tournament-style auto-evolution with parallel plan competition, winner merging, continuous optimization until benchmark targets are met

frontend (Frontend & Slides)

Frontend development skills covering design, patterns, and presentation creation. v4.1 Update: All three skills now include a mandatory "Phase 0: Style Discovery & Confirmation" workflow powered by Refero Styles. Before starting any implementation, the LLM will browse Refero's curated style library, extract design tokens (colors, typography, spacing, components), generate visual previews for user comparison, and only begin work after the user confirms their preferred style direction.

Skill	Description
frontend-design	Frontend design: creates distinctive, production-grade interfaces with strong visual direction. New: Mandatory Refero style discovery → preview generation → user selection → style lock before implementation
frontend-patterns	Frontend patterns: React/Next.js component patterns, state management, performance optimization. New: Visual style discovery when work involves UI rendering; Refero token-to-component mapping with role preservation rules
frontend-slides	Frontend slides: creates zero-dependency HTML presentations. New: Dual-source style discovery (Refero first + built-in presets fallback); mixed strategy supported

ppt-master (AI-Generated Natively Editable PPTX)

Sourced from ppt-master by Hugo He. An AI-driven presentation generation system that converts source documents (PDF/DOCX/URL/Markdown) into natively editable PPTX with real PowerPoint shapes (DrawingML) through multi-role collaboration (Strategist → Image_Generator → Executor). Unlike image-based PPT tools, every element in the output is directly clickable and editable in PowerPoint.

Prerequisites: Python 3.10+ and pip install -r requirements.txt (auto-detected on first use).

Skill	Description
ppt-master	AI-driven PPT generation: source document → multi-role collaboration → SVG pages → natively editable PPTX. Includes 7-step pipeline, 10 standalone workflows, 30+ Python scripts, brand/layout/chart/icon templates

Key features:

• Real PowerPoint — every element is natively editable (not flattened images)
• Multi-source input — PDF, DOCX, Excel, PPTX, web URLs, Markdown
• Multi-role pipeline — Strategist (8 confirmations) → Image Generator → Executor (SVG) → Post-processing → PPTX export
• Template system — brands, layouts, decks, charts, icons
• Standalone workflows — topic research, template fill, live preview, visual review, audio narration, animation customization, and more
• Progressive disclosure — SKILL.md stays concise (~345 lines); detailed role definitions and technical specs are loaded on demand from references/ and workflows/

taste-skill (Anti-Slop Frontend Design)

Sourced from Leonxlnx/taste-skill. A collection of portable Agent Skills that upgrade AI-built interfaces: stronger layout, typography, motion, and spacing instead of boilerplate-looking UIs. These skills systematically break the statistical biases that make LLM-generated frontends look generic and templated.

Skill	Description
taste-skill	Core anti-slop frontend skill: reads the brief, infers the design direction via three dials (Design Variance / Motion Intensity / Visual Density), and ships interfaces that do not look templated. Real design systems when applicable, audit-first on redesigns, strict pre-flight check
redesign-skill	Existing project upgrade: audits current design, identifies generic AI patterns, and applies high-end design standards without breaking functionality. Works with any CSS framework or vanilla CSS
output-skill	Full-output enforcement: overrides default LLM truncation behavior, bans placeholder patterns, and handles token-limit splits cleanly
soft-skill	High-end visual design: teaches the AI to design like a premium agency — precise fonts, spacing, shadows, card structures, and animations that make a website feel expensive
minimalist-skill	Minimalist UI: editorial-style interfaces (Notion/Linear vibes) with warm monochrome palette, typographic contrast, flat bento grids, muted pastels. No gradients, no heavy shadows
brutalist-skill	Industrial brutalist UI: raw mechanical interfaces fusing Swiss typographic print with military terminal aesthetics. Rigid grids, extreme type scale contrast, utilitarian color, analog degradation effects

oh-my-claudecode Skills Port

Sourced from the oh-my-claudecode plugin, providing full-lifecycle skills from requirements analysis to code delivery.

Skill	Description
autopilot	Full autonomous execution from idea to working code — requirements analysis, technical design, planning, parallel implementation, QA cycling, multi-perspective validation
autoresearch	Stateful single-mission improvement loop with strict evaluator contract, markdown decision logs, and max-runtime stop behavior
debug	Debug diagnostics: diagnoses session or repo state problems using logs, traces, and state inspection, distinguishing symptoms from root causes
deep-dive	Deep dive: 2-stage pipeline — trace (3 parallel causal investigation lanes) → deep-interview (requirements crystallization with 3-point injection)
deep-interview	Deep interview: Socratic questioning with mathematical ambiguity gating, refuses to proceed until ambiguity drops below threshold
deepinit	Deep initialization: creates hierarchical AGENTS.md documentation across the entire codebase for AI agent comprehension
external-context	External context: spawns 2-5 parallel document-specialist agents for external web searches and documentation lookup
learner	Learner: extracts reusable learned skills from the current conversation, capturing non-obvious workarounds and hidden gotchas
omc-plan	Strategic planning: intelligent interactive planning with auto-detection of interview needs, supports consensus mode (Planner/Architect/Critic loop)
ralph	Ralph persistence loop: PRD-driven persistent execution loop, story-by-story verification until all acceptance criteria pass, with mandatory review and code cleanup
ralplan	Consensus planning entrypoint: auto-gates vague ralph/autopilot/team requests before execution, ensuring plans are well-validated
release	Release assistant: analyzes repo release rules, caches them, then guides the release process
remember	Remember: reviews reusable project knowledge and decides what belongs in project memory, notepad, or durable docs
sciomc	Scientific analysis: orchestrates parallel scientist agents for comprehensive analysis with AUTO mode
skillify	Skillify: turns repeatable workflows from the current session into reusable skill drafts
trace	Trace: evidence-driven tracing lane that orchestrates competing tracer hypotheses for causal investigation
ultraqa	QA cycling: test → verify → fix → repeat workflow until goal is met
ultrawork	Parallel execution: high-throughput parallel task execution engine for simultaneous independent tasks
verify	Verify: verifies that a change really works before claiming completion, providing confidence in features, fixes, or refactors
visual-verdict	Visual verdict: structured visual QA for screenshot-to-reference comparisons with deterministic pass/fail guidance
wiki	Wiki: persistent markdown knowledge base that compounds across sessions (Karpathy model)
writing-plans	Writing plans: writes multi-step implementation plans from specs or requirements before touching code
writing-skills	Writing skills: creates new skills, edits existing skills, or verifies skills work before deployment

superpowers Skills Port

Sourced from the superpowers plugin, providing development workflow and collaboration enhancement skills.

Skill	Description
brainstorming	Brainstorming: MUST use before any creative work — explores user intent, requirements, and design before implementation
dispatching-parallel-agents	Dispatching parallel agents: dispatches agents in parallel when facing 2+ independent tasks without shared state or sequential dependencies
executing-plans	Executing plans: executes written implementation plans in a separate session with review checkpoints
finishing-a-development-branch	Finishing a development branch: guides completion of development work by presenting structured options for merge, PR, or cleanup
receiving-code-review	Receiving code review: requires technical rigor and verification when receiving code review feedback, not performative agreement or blind implementation
requesting-code-review	Requesting code review: verifies work meets requirements when completing tasks, implementing features, or before merging
self-improving-agent	Self-improving agent: universal self-improving agent that learns from ALL skill experiences using multi-memory architecture (semantic + episodic + working)
skill-creator	Skill creator: mandatory tool for creating new skills
subagent-driven-development	Subagent-driven development: executes implementation plans with independent tasks in the current session
systematic-debugging	Systematic debugging: systematic investigation before proposing fixes when encountering bugs, test failures, or unexpected behavior
test-driven-development	Test-driven development: writes tests before implementation code for any feature or bugfix
using-git-worktrees	Using git worktrees: creates isolated git worktrees for feature work that needs isolation from the current workspace
using-superpowers	Using superpowers: establishes how to find and use skills at the start of any conversation

GSD (Get-Shit-Done) Skills Port

Sourced from the get-shit-done plugin, providing a complete project management workflow from initialization to delivery. GSD implements a structured discuss → plan → execute → verify → ship cycle that complements SOLO Coder's Plan/Spec modes.

Core Workflow

Skill	Description
gsd-new-project	Initialize new project with deep context collection: questioning → research → requirements → roadmap
gsd-discuss-phase	Collect phase context through adaptive questioning before planning
gsd-plan-phase	Create detailed phase plans (PLAN.md) with research and verification loops
gsd-execute-phase	Execute all phase plans using wave-based parallelization
gsd-verify-work	Verify built features through conversational UAT testing
gsd-ship	Create PR, run reviews, and prepare for merge after verification passes

Project Management

Skill	Description
gsd-autonomous	Autonomously run all remaining milestone phases: discuss → plan → execute per phase
gsd-manager	Unified project manager: status, routing, and coordination across all GSD operations
gsd-new-milestone	Add a new milestone to an existing project
gsd-complete-milestone	Complete and archive a milestone with audit
gsd-milestone-summary	Generate milestone summary report
gsd-progress	Show current project progress and status
gsd-health	Diagnose project health and detect anomalies
gsd-stats	Display project statistics: phases, plans, requirements, git metrics

Code Quality

Skill	Description
gsd-code-review	Cross-AI code review with structured feedback
gsd-debug	Systematic debugging using scientific method with persistent state
gsd-forensics	Post-mortem investigation of failed GSD workflow executions
gsd-audit-milestone	Audit milestone completeness and quality
gsd-audit-uat	Cross-phase audit of all pending UAT and verification items
gsd-audit-fix	Fix issues found during audits

Quick Actions

Skill	Description
gsd-fast	Inline execution of simple tasks — no sub-agents, no planning overhead
gsd-quick	Quick execution with minimal planning for small-to-medium tasks
gsd-capture	Capture ideas, tasks, notes, and seeds to their target locations
gsd-help	Display complete GSD command reference

Specialized Phases

Skill	Description
gsd-ui-phase	UI-focused phase: design, prototype, implement, and review user interfaces
gsd-ai-integration-phase	AI integration phase: plan and implement AI/ML features
gsd-secure-phase	Security phase: audit and harden application security
gsd-mvp-phase	MVP phase: build minimum viable product with vertical slices
gsd-spec-phase	Specification phase: create detailed technical specifications
gsd-validate-phase	Validation phase: automated testing and validation

Knowledge & Configuration

Skill	Description
gsd-map-codebase	Analyze codebase with parallel mapping agents, generate structured documentation
gsd-graphify	Build, query, and inspect project knowledge graphs
gsd-extract-learnings	Extract decisions, lessons, patterns from completed phase artifacts
gsd-ingest-docs	Ingest external documents (PRD, ADR) into GSD planning context
gsd-config	Configure GSD settings: workflow toggles, advanced knobs, integrations
gsd-settings	Interactive GSD settings configuration

Workflow Control

Skill	Description
gsd-pause-work	Pause current work session with state preservation
gsd-resume-work	Resume a paused work session
gsd-undo	Safe git rollback of GSD phase or plan commits with dependency checking
gsd-pr-branch	Create and manage PR branches for GSD phases
gsd-cleanup	Clean up completed GSD artifacts and temporary files
gsd-update	Update GSD to the latest version

Additional Tools

Skill	Description
gsd-explore	Explore project structure and understand codebase
gsd-sketch	Explore UI/design ideas with disposable HTML prototypes
gsd-spike	Time-boxed technical investigation for risk reduction
gsd-review	Cross-AI review of project deliverables
gsd-review-backlog	Review and prioritize the project backlog
gsd-add-tests	Add tests to the project
gsd-docs-update	Update project documentation
gsd-eval-review	Evaluate and review project quality
gsd-plan-review-convergence	Converge plan reviews from multiple AI perspectives
gsd-ultraplan-phase	Ultra-detailed planning with maximum context
gsd-thread	Manage conversation threads across GSD sessions
gsd-workspace	Workspace management and environment setup
gsd-workstreams	Manage parallel workstreams within a project
gsd-inbox	Manage incoming tasks and requests
gsd-import	Import external project data into GSD
gsd-profile-user	Configure user profile and preferences
gsd-phase	Generic phase operations
gsd-ns-context	Namespace context management
gsd-ns-ideate	Namespace ideation and brainstorming
gsd-ns-manage	Namespace management operations
gsd-ns-project	Namespace project operations
gsd-ns-review	Namespace review operations
gsd-ns-workflow	Namespace workflow operations

gstack Skills Port

Sourced from the gstack plugin by Garry Tan, providing a comprehensive engineering workflow from planning through deployment. gstack implements a structured ship cycle with canary monitoring, multi-perspective plan reviews, and design consultation — built for teams that ship fast with confidence.

Core Workflow

Skill	Description
gstack	Root router skill: session initialization, skill routing, completion protocol, and self-improvement
browse	Web browsing: search and fetch web content for research and context gathering
ship	Ship workflow: 16-step release process from branch creation through deployment verification
investigate	Systematic debugging: 5-stage investigation workflow with "no root cause, no fix" principle
qa	QA testing: 10-stage quality assurance workflow with health scoring rules
qa-only	QA-only mode: run quality checks without full ship workflow
canary	Canary monitoring: post-deployment canary watch with automated rollback triggers
careful	Careful mode: enhanced review and validation for high-stakes changes
review	Pre-land PR review: Review Army expert dispatch for multi-perspective code review
land-and-deploy	Land and deploy: merge verification and deployment confirmation
health	Health dashboard: 6-dimension code quality assessment

Planning & Review

Skill	Description
autoplan	Auto-planning: generates structured plans with dependency analysis and parallel step detection
plan-ceo-review	CEO/Founder mode plan review: 4 review modes (Founder/VC/Operator/All)
plan-eng-review	Engineering plan review: technical feasibility, architecture, and implementation assessment
plan-design-review	Design plan review: UX, visual hierarchy, and interaction quality assessment
plan-devex-review	Developer experience review: DX real-time audit for tooling and workflow quality
plan-tune	Plan tuning: iterative plan refinement based on review feedback

Design

Skill	Description
design-consultation	Design consultation: expert design feedback and recommendations
design-html	HTML design: Pretext text layout engine for production-grade HTML generation
design-review	Design review: visual and interaction quality assessment
design-shotgun	Design shotgun: rapid multi-variant design exploration

Office & Retrospective

Skill	Description
office-hours	Office hours: structured Q&A and decision-making sessions
retro	Weekly engineering retrospective with global retro and comparison modes
devex-review	DX review: real-time developer experience audit

Security & Documentation

Skill	Description
cso	Chief Security Officer: comprehensive security review and compliance check
document-generate	Document generation: Diataxis framework documentation creation
document-release	Document release: post-ship documentation updates
learn	Project learning: capture and manage project knowledge
scrape	Web scraping: extract data from web pages

Context & Safety

Skill	Description
context-save	Context save: preserve session context for later restoration
context-restore	Context restore: resume from saved session context
freeze	Freeze: restrict editing scope to prevent unintended changes
unfreeze	Unfreeze: release editing restrictions
guard	Guard mode: full safety mode with comprehensive protection rules

Hooks Integration (v5.1 Update)

The following gstack skills now include hook adaptations for TRAE IDE's solo mode, enabling automatic enforcement through the hooks mechanism:

• careful — PreToolUse hook checks destructive commands before execution
• freeze — PreToolUse hook enforces edit boundaries on Write/Edit operations
• guard — SessionStart hook injects comprehensive protection rules
• investigate — SessionStart hook loads systematic debugging context

Hook configuration and PowerShell scripts are packaged in hooks-config.zip. Extract to .trae/hooks/ and configure .trae/hooks.json to enable automatic safety enforcement.

Benchmarking

Skill	Description
benchmark	Benchmark: web performance regression detection
benchmark-models	Benchmark models: model performance comparison and evaluation

Setup & Utilities

Skill	Description
gstack-codex	OpenAI Codex CLI wrapper for alternative model access
gstack-gstack-upgrade	gstack upgrade: update to latest gstack version
gstack-hackernews-frontpage	Hacker News frontpage: fetch and analyze HN stories
gstack-landing-report	Landing report: version queue status and release tracking
gstack-make-pdf	Make PDF: convert Markdown to PDF
gstack-open-gstack-browser	Open gstack browser: launch browser for web interaction
gstack-pair-agent	Pair agent: remote agent pairing collaboration
gstack-setup-browser-cookies	Setup browser cookies: configure browser authentication
gstack-setup-deploy	Setup deploy: configure deployment parameters
gstack-setup-gbrain	Setup gbrain: configure knowledge base integration
gstack-skillify	Skillify: browser skill encoding and creation
gstack-sync-gbrain	Sync gbrain: synchronize knowledge base data

OpenClaw

Skill	Description
gstack-openclaw-ceo-review	OpenClaw CEO review: founder-perspective plan review
gstack-openclaw-investigate	OpenClaw investigate: systematic problem investigation
gstack-openclaw-office-hours	OpenClaw office hours: structured Q&A
gstack-openclaw-retro	OpenClaw retro: engineering retrospective

---

How to Use

1. Download the desired skill .zip file
2. Open Skill Management in TRAE_CN SOLO
3. Click Upload Skill and select the downloaded zip file
4. The skill will be automatically parsed and installed

Note: Each zip file contains a root-level SKILL.md file, conforming to the official TRAE skill format specification.

Credits

• superpowers — Claude Code plugin
• oh-my-claudecode — Claude Code plugin
• get-shit-done — Claude Code plugin
• gstack — Claude Code plugin (by Garry Tan)
• ppt-master — AI-driven PPT generation (by Hugo He)

Changelog

v5.1 — gstack Skills: Hooks Adaptation for TRAE IDE Solo Mode

• Added hooks integration for 4 core gstack skills: careful, freeze, guard, investigate
• careful now enforces destructive command checks via PreToolUse:RunCommand hook
• freeze now enforces edit boundary checks via PreToolUse:Edit and PreToolUse:Write hooks
• guard now injects comprehensive protection rules via SessionStart hook
• investigate now loads systematic debugging context via SessionStart hook
• Packaged hooks-config.zip containing:
  • hooks.json — Project-level hook configuration with event matchers and timeouts
  • PowerShell hook scripts: check-careful.ps1, check-freeze.ps1, session-start.ps1
• Updated SKILL.md for all 4 skills with hook adaptation notes and TRAE-specific instructions
• Hooks enable automatic safety enforcement without manual skill invocation

v5.0 — ppt-master: AI-Generated Natively Editable PPTX

• Added ppt-master skill — AI-driven presentation generation system
• Converts source documents (PDF/DOCX/URL/Markdown) into natively editable PPTX with real PowerPoint shapes
• 7-step pipeline: source processing → project init → template → strategist (8 confirmations) → image acquisition → executor (SVG) → post-processing & export
• 10 standalone workflows: topic research, template fill, create template/brand, resume execute, verify charts, customize animations, live preview, visual review, audio narration
• 30+ Python scripts for source conversion, image generation, SVG processing, and PPTX export
• Template system: brands (Anthropic, Google), layouts (7 types), decks (8 types including Chinese brands), charts (50+ SVG), icons (3 libraries)
• Progressive disclosure architecture: SKILL.md ~345 lines, detailed specs loaded on demand
• Auto-detection of Python environment and dependencies on first use
• Bilingual: Chinese frontmatter/description, English instructions (matching original project)

v4.1 — Frontend Skills: Refero Style Discovery Integration

• All three frontend skills now include a mandatory Phase 0: Style Discovery & Confirmation workflow
• Powered by Refero Styles — curated design references extracted from real product pages
• Workflow: analyze requirements → browse Refero style library → extract design tokens → generate visual previews → user selects → style lock → begin implementation
• frontend-design: Full 6-step Phase 0 with UUID reference table, preview template, style lock format, Refero MCP optional enhancement, and fallback strategy
• frontend-patterns: Style discovery for visual work (skippable for logic-only tasks); new "Implementing Components with Refero Tokens" section with token mapping rules and role preservation
• frontend-slides: Dual-source style discovery (Refero first + built-in STYLE_PRESETS fallback); mixed strategy for combining Refero tokens with preset layouts/animations
• Style lock mechanism prevents implementation drift from user-confirmed direction
• Token role preservation rules: CTA colors stay CTA-only, border radii keep their specified elements, etc.

v4.0 — gstack Skills Port (52 New Skills)

• Added 52 skills from the gstack plugin (by Garry Tan)
• Complete engineering workflow: planning → review → ship → canary monitoring
• Multi-perspective plan reviews: CEO, engineering, design, and DX review modes
• Design consultation and multi-variant design exploration
• Security review (CSO), documentation generation, and context management
• All 52 zip packages verified through format compliance checks

v3.0 — GSD Skills Port (66 New Skills)

• Added 66 skills from the get-shit-done (GSD) plugin
• Complete project management workflow: discuss → plan → execute → verify → ship
• Path adaptation: .planning/ → .trae/gsd/planning/
• Tool adaptation: gsd-sdk/gsd-tools CLI references annotated with TRAE alternatives
• Language unification: All descriptions and core instructions in Chinese
• SOLO mode adaptation: Each skill includes relationship notes with SOLO Coder Plan/Spec modes
• All 66 zip packages verified through format compliance checks

v2.0 — Deep SOLO Mode Adaptation

• Path adaptation: Unified .omc/, .claude/ paths to .trae/
• Tool adaptation: Replaced Claude Code-specific tool references with TRAE SOLO-compatible alternatives
• Language unification: All skill descriptions and core instructions in Chinese
• SOLO mode adaptation: Each skill includes relationship notes with SOLO Coder's Plan/Spec modes
• Spec compliance: Added "Not Applicable Scenarios", "Output Specification", and "Dependencies" sections
• Lossless functionality: All core instructions, data structures, parameter thresholds, and code examples are fully preserved

License

MIT