agent-tiered-planning
适合混能团队，避免按工时误分配导致的质量问题与返工。
依赖组织流程与权限设定，不当分配可能造成任务误交付。
按能力层级为多种代理分配任务并制定升级与操作手册。
在多人/多模型（人工、强力模型、廉价模型）共同执行 backlog 时，把每个 issue 按“代理能力层级”打标签，并发布一份冷启动代理先读的 playbook。还规定当任务超出某层级要自动升级，防止小但需判断的活被廉价代理做出错的实现。适合需要按能力而非单纯按工时分配工作的团队。
▸ 展开 SKILL.md 英文原文
Plan and dispatch work across a mixed-capability pool of coding agents (humans, frontier models, cheaper models) by labeling every issue with an agent-capability tier orthogonal to its effort estimate, publishing a playbook cold agents read before picking up work, and enforcing an escalation rule when an issue exceeds its tier. Use when a repo's backlog is executed by more than one class of agent — at planning time (assign tiers as issues are written) and at pickup time (the protocol any cold agent follows). Pairs with design-direction-lock (how standard-tier agents safely take design-facing work), implementation-discipline (how any tier conducts itself inside an issue), and adversarial-review (the gate the upper tiers mandate).
帮我安装这个 skill：https://raw.githubusercontent.com/boshuthebuilder/ai-tradecraft-skills/main/plugins/coding/skills/agent-tiered-planning/SKILL.mdcurl -fsSL "https://raw.githubusercontent.com/boshuthebuilder/ai-tradecraft-skills/main/plugins/coding/skills/agent-tiered-planning/SKILL.md"# agent-tiered-planning A backlog executed by mixed-capability agents fails in a specific way: work is routed by *size*, but size is the wrong axis. A one-hour task can require deep judgment; a two-day task can be pure mechanical execution. When a cheap agent picks up a subtle-but-small issue, the failure is not a delay — it is a **wrong-but-merged implementation** that downstream issues inherit. Tiering makes capability an explicit, labeled dimension of every issue, so the routing decision is made once, at planning time, by whoever understands the work best — instead of implicitly, at pickup time, by the agent least equipped to judge it. The deep principle (shared with design-direction-lo