Professional AI Agent Skill Toolkit — Cross-platform workflow skills for OpenCode, Claude Code, Cursor, Codex
Opc_Kit is a cross-platform AI Agent skill toolkit. Each skill is a meticulously designed, rigor validated professional workflow that helps product managers, developers, and designers efficiently complete complex tasks.
Core Philosophy:
- 📐 Structured Assurance — Mandatory templates and validation ensure output quality
- ⚡ Efficiency Optimization — Auto-inference + on-demand interaction reduces repetitive work
- 🔗 Traceability — Bidirectional traceability ensures every decision has a source
- 🎓 Professional Perspective — Senior expert mindset frameworks and industry best practices
- 🌐 Cross-Platform Compatible — Platform-agnostic design, auto-adapts to mainstream Agent toolchains
Automatically identifies user intent (create/update/validate) without manual workflow specification.
- Coaching Mode (Default) — Auto-inference → Complete draft → Self-review → On-demand interaction (~5-10 min)
- Fast Mode — Quick generation of complete document → Reverse question supplement (~2-3 min)
- First Principles Validation — 5+ fundamental questions ensure document foundation correctness
- Logical Completeness Review — US↔FR bidirectional traceability, Tracking↔Metrics traceability
- Boundary & Risk Scanning — Proactively identifies exception flows, boundary conditions, external dependencies
- Fixed skeleton: Problem Description, Goal Definition, Target Users, User Stories, Feature Flowchart, Feature List, Feature Details, Tracking Design, Future Improvements, Risks & Dependencies
- Auto-generated: Decision Log, Glossary, Assumption Index
Intelligent detection of conversation language (Chinese/English), auto-generates PRD in the matching language. User explicit override takes highest priority.
7-dimension quantitative scoring (max 100 points) with production-ready benchmarks (70+ = ready, 85+ = excellent).
Auto-creates task list at start, real-time progress updates ("✅ Chapters 1-3 completed. Moving to User Stories...").
Each interaction provides 1-3 carefully considered recommendations with rationale, guiding decisions not fill-in-the-blanks.
Auto-infers target platform (iOS/Android/Web/Mini Program/Admin Backend) from user input using confidence-level strategy. High confidence: direct use. Medium: use + flag. Low: one-line confirm.
PRD generation follows 5-batch incremental write (Ch1-3 → Ch4-5 → Ch6-7 → Ch8-10 → Ch11-13). Real-time progress notifications after each batch. Session-interrupt recovery via temporary checkpoint tracking (front matter removed after completion).
Note: Checkpoint metadata (generate_progress) is stored in YAML front matter during generation and automatically removed when PRD generation completes. Final PRDs contain no front matter; business metadata lives only in the
## Metadatasection.
Requirements-first, prototype-after workflow. Validates PRD status before prototype design, warning users about rework risk when skipping PRD.
| Skill | Language | Purpose | Version | Install Command |
|---|---|---|---|---|
| create-prd | EN/CN | PRD creation/update/validation | v2.2.2 | npx skills add sacrtap/Opc_Kit --skill create-prd |
No installation required. The skill auto-creates docs/specs/ directory if absent.
Node.js (optional): Only needed for running the PRD validation script validate-prd.js.
# Install all skills
npx skills add sacrtap/Opc_Kit
# Install specific skill
npx skills add sacrtap/Opc_Kit --skill create-prd
# Update specific skill to latest version
npx skills update create-prd
# List available skills before installing
npx skills add sacrtap/Opc_Kit --list# Clone repository
git clone https://github.com/sacrtap/Opc_Kit.git
# Copy skill to your project's .agents/skills/ directory
cp -r Opc_Kit/create-prd /your/project/.agents/skills/Activate the skill in your AI Agent:
/create-prd Create a PRD for user authentication feature
Or directly describe your requirement:
/create-prd Help me write a PRD for: users can create and share collections on the platform
# Fast Mode (skip interaction, direct generation, ~2-3 min)
fast
# Coaching Mode (on-demand interaction, progressive refinement, ~5-10 min)
coaching
create-prd is a professional PRD (Product Requirements Document) writing assistant, providing complete workflow support from creation, update to validation.
| Intent | Function | Trigger Signals |
|---|---|---|
| create | Create new PRD | "New PRD", "write requirements doc", "create product requirements" |
| update | Update existing PRD | "Update/modify existing PRD", "PRD change", "add features to existing doc" |
| validate | Validate PRD completeness | "Validate/check PRD", "review requirements doc completeness" |
- Fixed Template — 12-chapter skeleton + 3 auto-generated, immutable order
- Strict Validation — US↔FR bidirectional traceability, 18+ criteria checklist
- Mandatory Change Log — Every update must append change records
- Exception-Covered Flowcharts — Every API call has failure + timeout branches
- Testable Acceptance Criteria — Quantifiable, executable judgment conditions
- Auto Language Detection — Chinese/English bilingual PRD generation
- Cross-Agent Compatible — Works on OpenCode, Claude Code, Cursor, Codex
- Incremental Write + Progress — 5-batch write with real-time notifications and checkpoint recovery
- Platform Inference — Auto-detects target ecosystem from user input, zero-interaction fill
- Prototype Constraint — Requirements-first workflow, validates PRD status before prototype design
- Cross-Engine Flowcharts — Mermaid syntax compatible with Zed, VS Code, GitHub (no quotes, no circles, no HTML)
User: /create-prd Help me write a PRD for user collection feature
Skill:
1. Deep reasoning: Analyze user motivation, business value, technical feasibility
2. Auto inference: Generate 12-chapter complete draft
3. On-demand interaction: Real-time confirmation for critical assumptions
4. Strict validation: US↔FR bidirectional traceability, flowchart exception check
5. Output: High-quality executable PRD + assumption index + quality score
| Dimension | Traditional Approach | create-prd Skill |
|---|---|---|
| Structural Completeness | Depends on author experience, prone to omission | Mandatory 12-chapter skeleton + bidirectional traceability validation |
| Quality Assurance | No automatic validation mechanism | First principles validation + boundary risk scanning |
| Efficiency | Extensive fill-in-the-blank Q&A | Inference + on-demand interaction, recommendation-driven decision |
| Traceability | Features separated from requirements | US↔FR bidirectional traceability, every feature has a source |
| Change Management | No version records | Mandatory change log + key update annotations |
| Professionalism | Generic templates | Senior PM perspective + industry best practices |
- Validation Script —
scripts/validate-prd.jsprogrammatically checks 18+ quality criteria - Quality Score — 7-dimension scoring (max 100 points), see Quality Scoring section
- Peer-Reviewed Methodology — Amazon PRFAQ, First Principles, YAGNI, MoSCoW frameworks
- Bidirectional Traceability — US↔FR 1:1 mapping, no orphaned requirements
- Exception Path Coverage — Every API call and external dependency has failure + timeout branches
- Version History — Mandatory changelog for every update, tracked in PRD metadata
Benchmark: 70+ points = production-ready, 85+ = excellent
Every PRD is scored on 7 dimensions (max 100 points):
| Dimension | Weight | Max Points | Description |
|---|---|---|---|
| Completeness | 30% | 30 | All 13 chapters present and filled |
| Traceability | 25% | 25 | US↔FR 1:1 mapping, no orphaned requirements |
| Testability | 15% | 15 | Acceptance criteria are executable and quantifiable |
| Exception Coverage | 15% | 15 | Failure paths documented for all external calls |
| Assumption Coverage | 5% | 5 | All inferences tagged and summarized |
| Review Completeness | 5% | 5 | Three review steps executed with fixed format |
| Product Thinking | 5% | 5 | Why Now, differentiation, user segmentation |
Benchmark Scores:
- 70+ points = production-ready
- 85+ points = excellent
- < 70 points = needs revision before sharing
Generated PRDs integrate with common product development tools:
- Figma: Include prototype URLs in PRD metadata (
metadata.prototype) - Jira: Export feature list as Jira Epic/Story structure (F-x.x → Story ID mapping)
- GitHub: Link to related issues/PRs in PRD metadata (
metadata.related-docs) - Confluence: Mermaid flowcharts render natively in Confluence markdown
- Notion: PRD markdown imports cleanly with table and heading support
- Analytics: Tracking events compatible with Amplitude, Mixpanel, Segment
This skill is platform-agnostic, auto-adapting to mainstream AI Coding Agents:
| Operation | OpenCode | Claude Code | Cursor | Codex |
|---|---|---|---|---|
| File Read | read |
Read |
Built-in | read_file |
| File Create/Overwrite | write |
Write |
Built-in | write_file |
| Precise Edit | edit |
Edit |
Built-in | edit_file |
| File Lookup | glob |
Glob |
Built-in | glob |
| Content Search | grep |
Grep |
Built-in | search |
| Large File Processing | ctx_execute_file |
Bash |
Bash |
run_shell |
| Task List Management | todowrite |
TodoWrite |
Built-in | Built-in |
| Subagent Dispatch | task |
Task |
❌ Not supported | ❌ Not supported |
When a specific tool is unavailable:
- Large file analysis → Use
bash/run_shellwithhead/tail/grepto read key sections - Subagents → Execute tasks sequentially in current session, report progress after each
- Task list → Outline plan in text at session start, mark completion step by step
We welcome high-quality skill contributions!
- Fork this repository
- Create new skill folder (e.g.,
my-skill/) - Write SKILL.md following this structure:
- Frontmatter (name, description, license, metadata)
- Trust & Quality
- Quick Start
- When to Use / Intent Recognition
- Core Principles
- Quality Scoring
- Template + Validation
- Assets & Scripts / Integrations
- Submit PR with skill usage examples
- ✅ Fixed template + mandatory validation mechanism
- ✅ Recommendation-driven interaction (not fill-in-the-blank Q&A)
- ✅ Bidirectional traceability / traceability assurance
- ✅ Professional perspective + industry best practices
- ✅ Cross-platform compatible (no Agent toolchain lock-in)
- ✅ Complete documentation + usage examples
See CONTRIBUTING.md for details.
MIT © sacrtap | create-prd v2.2.2
- GitHub Issues: Report issues or request features
- Discussions: Share use cases
Opc_Kit — Empower AI Agents to become true product workflow experts, not simple Q&A machines.