メニュー
Collaborative code review using Agent Teams where reviewers discuss findings, challenge each other, and converge on a unified assessment. Evolves dm-work:review from isolated parallel reviewers to a review team.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
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 | review |
| description | Collaborative code review using Agent Teams where reviewers discuss findings, challenge each other, and converge on a unified assessment. Evolves dm-work:review from isolated parallel reviewers to a review team. |
Team-based collaborative code review using Agent Teams.
Current /dm-work:review spawns 3 isolated subagents that report back independently. Findings may conflict, duplicate, or miss connections between concerns. With Agent Teams, reviewers can:
/dm-work:review (subagent-based) — reach for the team variant only when reviewers genuinely benefit from challenging each other| Role | Teammate | Model | Skills to inject |
|---|---|---|---|
| Architecture | Teammate 1 | opus | dm-arch:solid-architecture, dm-arch:data-oriented-architecture |
| Code Quality | Teammate 2 | opus | Language-specific dm-lang:*-pro skill |
| Security | Teammate 3 | opus | (none — general security knowledge) |
| Lead | You | opus | Synthesis and final assessment |
Code reviewer may split by domain (same as dm-work:review) for large reviews.
Still uses a haiku Explore subagent (not a teammate — too small for team overhead). Scout categorizes files, detects patterns, identifies hotspots, splits review domains.
Compatible with dm-work:review:
## Review: <scope> (<N> commits, <LOC> across <M> files)
### Verdict: <emoji> <summary>
**Critical** (<count>)
1. `file:line` - <description> [Consensus: arch+security agreed]
**High** (<count>)
...
**Architecture**: <verdict + summary>
**Code Quality**: <verdict + summary>
**Security**: <verdict + summary>
### Review Discussion Notes
[Key debates between reviewers and resolutions]
New addition: "Review Discussion Notes" section captures inter-reviewer debates and resolutions — this is the unique value of team review over subagent review.
--pr N for PR mode (GitHub review comments)--commits <range> for explicit scope--only <acs> to filter reviewer types--min-severity <level> to filter output--skip-beads for local mode without bead creation