nanmicoder/claude-code-skills
Overview
This skill is an Agent Teams orchestration decision engine that automatically decides whether to spawn Agent Teams or use Subagents for a given task. It evaluates task characteristics across five dimensions and returns a recommendation with confidence, a suggested team design, and a ready-to-use prompt template. The goal is to maximize parallel exploration value while minimizing coordination cost and file conflict risk.
How this skill works
On task receipt, the skill scores five dimensions: parallelism, inter-member communication needs, context isolation, file-conflict risk, and cost/benefit. It applies a decision matrix: 4–5 positive checks strongly favor Agent Teams, 0–1 favor Subagents, and 2–3 need additional context. When recommending Teams, it also suggests team size, roles, task granularity, and generates a spawn prompt and warnings.
When to use it
- Tasks that require multi-angle parallel analysis (code review, hypothesis debugging).
- Work that needs teammates to share findings, challenge conclusions, and iterate.
- Cross-layer development where frontend, backend, and test roles must coordinate.
- When the user explicitly requests creating a team or uses commands like /team.
- Task descriptions containing keywords: parallel, simultaneous, multi-person, collaborate.
Best practices
- Design clear, complementary roles and assign file ownership to avoid overlaps.
- Keep team size optimal (3–5 typical; avoid >7 due to coordination overhead).
- Aim task granularity for 15–30 minutes per unit with clear deliverables.
- Provide explicit context in spawn prompts: file paths, tech stack, expected output.
- Monitor progress and redirect unproductive explorations; close unused teammates promptly.
Example use cases
- Multi-angle PR review: security, performance, and test reviewers run in parallel and debate findings.
- Competitive hypothesis debugging: five members each propose and test a root-cause hypothesis.
- Feature implementation across layers: frontend, backend, and test engineer coordinate API contracts and integration tests.
- Research and design spike: UX researcher, architect, and devil-advocate explore tradeoffs and consolidate recommendations.
FAQ
It scores five dimensions (parallelism, communication need, context isolation, file conflict, cost/benefit) and applies a decision matrix: 4–5 positives → Agent Teams, 0–1 → Subagent, 2–3 → conditional.
What team size and task granularity are recommended?
Prefer 3–5 members for most tasks, 2–3 for simple reviews, 5–7 for high complexity. Design tasks that take 15–30 minutes each with clear, testable deliverables.
4 skills
This skill analyzes task features to decide between Agent Teams or Subagent, enabling parallel collaboration and cross-functional coordination.
This skill provides a concise LangChain 1.0 quick reference for building agents, tools, memory, and middleware workflows.
This skill generates Bilibili chapters from subtitles, ensuring correct timing, titles, and formatting for easy copying into the editor.
This skill converts SRT subtitles into structured JSON, extracting timing, duration, and text for analysis and downstream processing.
