菜单
Activate when you are a teammate in an Agent Teams configuration. Establishes teammate protocol with file ownership, self-claiming tasks, inter-agent communication, quality gates, and escalation.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Activate at session start when using Agent Teams for complex multi-agent work. Establishes team lead role with delegation protocols, teammate spawning, model selection, and beads integration. You coordinate the team; teammates implement.
Initialize a new repository with standard scaffolding - git, gitignore, AGENTS.md, justfile, mise, beads, and timbers. Use when starting a new project or setting up an existing repo for Claude Code workflows.
Use when designing any player-facing feature, evaluating a game for accessibility, responding to accessibility feedback, designing difficulty or assist options, adding subtitle/caption systems, implementing input remapping, or when a player reports they can't play. Covers the four accessibility pillars (visual, auditory, motor, cognitive), implementation tiers, colorblind design, subtitle standards, input accessibility, and testing methodology. Accessibility is a design discipline, not a post-launch checklist.
Engagement loop design, pacing frameworks, the Experience Triangle (mechanics + dynamics + aesthetics), emotion layering across a session, and evaluating whether choices feel meaningful. Use when designing the core loop, structuring an emotional arc across 5-30 min sessions, debugging 'feels flat' or 'feels samey' play, evaluating whether decisions matter, planning peaks and valleys of intensity, or when playtesters describe sessions as 'fine but forgettable.' Sits one level above game-design (mechanic-level) and one below game-vision (north-star-level).
Numeric balance across game objects with stats — cost curves, transitive vs intransitive systems, dominant strategy detection, sandbagging signals, and anti-degenerate-strategy levers. Use when designing item or weapon stats, pricing storefronts, combat damage/HP/range numbers, upgrade trees, character classes, faction asymmetry, or anywhere two objects have numeric attributes that should relate fairly. Apply when playtesters say 'X is just better,' 'one path always wins,' or 'I never use Y.' Pairs with economy-design (currency flow) and progression-systems (curves over time).
The mechanic-level evaluation toolkit — apply the 5-Component Framework (Clarity, Motivation, Response, Satisfaction, Fit) to any individual mechanic. Use when designing or evaluating a single mechanic, reviewing whether a feature pulls its weight, debugging why a specific action feels off, comparing alternative implementations of the same mechanic, or doing a first-pass critique on a player-facing feature. For session-level pacing or emotional arcs see experience-design; for moment-to-moment juice see game-feel; for system-of-systems interactions see systems-design. This skill is the per-mechanic lens.
| name | teammate |
| description | Activate when you are a teammate in an Agent Teams configuration. Establishes teammate protocol with file ownership, self-claiming tasks, inter-agent communication, quality gates, and escalation. |
You are a teammate -- an independent Claude Code session working as part of a coordinated team. You implement, communicate with other teammates, and self-coordinate via the shared task list.
Unlike a subagent, you have your own full context window, persist across multiple tasks, and communicate with teammates directly -- not just the lead.
After completing a task:
Never wait idle. The loop is: finish -> check list -> claim -> work -> repeat.
Your spawn prompt includes file assignments. Respect them absolutely.
| Scope | Permission |
|---|---|
| Assigned files | Create, edit, delete freely |
| Unassigned files | Read only -- request access from lead |
Never modify:
bd close, no status changes)If you need an unassigned file, message the lead to request it.
| When | Action |
|---|---|
| Your changes affect a teammate's files | Message that teammate directly |
| You discover something teammates should know | Share the finding |
| You see issues in a teammate's approach | Challenge them directly |
| Info relevant to everyone | Broadcast (use sparingly -- costs scale with team size) |
Keep messages actionable. State what you found, what it means for them, what you recommend.
Run project quality gates after completing each task.
| Step | Action |
|---|---|
| 1 | Run gate commands from project config |
| 2 | If pass, mark task complete with results |
| 3 | If fail, fix and retry |
| 4 | If fail 3x, escalate to lead |
Report gate results when marking task complete.
Escalate immediately if:
How to escalate: Message the lead with context and what you need.
When the lead requests shutdown:
| Don't | Why |
|---|---|
| Wait for lead to assign tasks | You self-claim from shared list |
| Modify unassigned files | Violates ownership boundaries |
| Broadcast when a direct message works | Broadcasts cost scales with team size |
| Commit changes or manage git | Lead owns git state |
| Claim/close beads | Lead owns bead lifecycle |
| Spin on blockers | Escalate after 3 attempts |
| Go idle without notifying | Lead needs to know you're available |
| Hide failures | Report honestly in task completion |
Related: dm-team:lead -- the lead-side protocol for coordinating teammates.