Project Skill Writer

Build project-level skills with deterministic scaffolding, then refine with domain evidence.

L1: Create or Update Project Skills

• Create new project skills under the discovered project skill directory
• Update existing skill workflows, triggers, and exclusions
• Normalize path style, language style, and output structure
• Delegate final drafting to skill-creator when needed
• Reject global-scope output targets

L2: Project Workflow Contract

1. If repository is large, require user-selected scope before deep analysis
2. Detect project skill path by loading Path Discovery
3. Gather project evidence from selected scope only
4. Check global prerequisite: skill-creator; if missing, ask user to install before continuing
5. Delegate drafting to skill-creator and enforce project-only output path
6. For Trae / Trae-CN targets, apply Trae Skill Best Practices
7. If target is not a skill request, route to the matching writer skill

L3: Category Details

Category A: New Skill Creation

• Identify one domain per skill
• Keep triggers focused on user intent, not just keywords
• Keep file paths project-relative only

Category B: Existing Skill Update

• Preserve valid frontmatter fields
• Replace weak workflow steps with executable steps
• Move long vendor/runtime details into references

Category C: Quality Gate

• Ensure single-language consistency
• Ensure no absolute paths
• Ensure clear activation and non-activation sections

Category C2: Prerequisites Gate

• Require global availability of skill-creator
• If missing, prompt user with installation guidance before execution
• Suggested commands:
  • npx skills add skill-creator -g -y

Category D: Reference Loading

• Load references/path-discovery.md for output path discovery and validation
• Load references/advanced-patterns.md only for complex structuring decisions
• Load references/trae-target-mode.md only when runtime target is Trae or Trae-CN
• Load references/trae-skill-best-practices.md only when runtime target is Trae or Trae-CN

Category E: Target Routing

• Skills request: handle in this skill
• Agents request: route to project-agent-writer
• Rules request: route to project-rules-writer

Category F: Claude Code Practices

• Keep project skill guidance in CLAUDE.md or .claude/CLAUDE.md
• Split specialized guidance by project rule files when needed

Category G: Trae / Trae-CN Practices

• Keep project skill guidance in project-managed .trae context
• Follow Trae project conventions for scoped instructions
• Carry forward existing Trae-specific constraints into new project outputs

Agents

• Project Scanner: use for large project boundary scan
• Tech Stack Analyzer: use for framework-aware skill design
• Quality Validator: use before final delivery
• Convention Detector: use when conventions are implicit

References

• Advanced Patterns
• Path Discovery
• Trae Target Mode
• Trae Skill Best Practices