Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Developer agent skill router - routes to 31 dev skills by category and signal keywords, manages 5 sub-agents.
.claude/skills/majiayu000-dev-router/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 0% | 0% |
> "Right skill, right sub-agent, right time."
Typical task types and the skill combinations that work well for them.
Skills:
Why: Research ensures patterns are followed.
Skills:
Why: Full server-authoritative stack from architecture to client. Validation ensures server builds correctly.
Skills:
Why: Prediction builds on server-authoritative foundation. Advanced TypeScript for generic prediction types.
Skills:
Why: Vite patterns + specific model loading. R3F fundamentals for scene integration.
Skills:
Why: Start with basics, add specific techniques based on problem. Object pooling for per-frame creation.
Skills:
Why: TypeScript for component props. UI animations for polish. E2E testing for UI validation.
Skills:
Why: Server-authoritative hit detection. Shooting prediction. Object pooling for projectiles.
Skills:
Why: Coverage tracking pattern. Advanced TypeScript for grid state types. R3F for visualization.
Skills:
Why: Mobile optimization patterns. Haptics for feedback. E2E testing with touch events.
Skills:
Why: Find existing patterns first. Load only the relevant domain skill. Validation to ensure fix works.
For complex tasks, load skills in this order:
Task({ subagent_type: "developer-code-research", ... })
↓
[Research completes, provides findings]
↓
Load domain skills based on findings
↓
Task({ subagent_type: "code-implementation", ... })
↓
[Implementation completes]
↓
Task({ subagent_type: "developer-validation", ... })
↓
[Validation passes]
↓
Task({ subagent_type: "commit-agent", ... })| Anti-Pattern | Why | Correct Approach | | ---------------------------------------------- | ------------- | --------------------------------- | | Loading all R3F skills at once | Context bloat | Load only relevant skills | | Loading validation before research | Wasted cycles | Research → Implement → Validate | | Skipping research for "simple" tasks | Miss patterns | Always research first | | Loading both basics and advanced TypeScript | Redundant | Advanced includes basics concepts | | Loading multiplayer for single-player features | Unused code | Only load when needed |
Other measured skills in the registry, with their headline benchmark lift.