메뉴
统一路由仓库级请求到合适 lane 与交付深度。用于处理任何仓库内任务，尤其是在需要在 answer/git/github/bug/feature/refactor/review/architecture 之间选择路径、判断是否触发 brainstorming，或协调多 Agent 与多技能执行时。
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when handling local Git CLI daily operations such as checking status or diff, staging files, creating conventional commits, restoring local changes, reverting commits, or resolving merge/rebase/cherry-pick conflicts.
Use when explaining code, answering technical questions, comparing approaches, or providing repository-grounded analysis without entering an implementation or remediation workflow.
Use when evaluating system architecture, platform design, or high-level technical structure to identify production risks, evidence gaps, and prioritized remediation without directly redesigning or implementing the system.
Use when improving internal code structure without intentionally changing externally observable behavior, such as module splits, simplification, dependency cleanup, or maintainability and performance-oriented restructuring.
Use when feature ideas, issues, stakeholder notes, or partial specs need to be reviewed and turned into a concise, testable PRD with clear scope, user stories, acceptance criteria, and explicit open decisions.
Use when a PRD, feature spec, or acceptance criteria already exist and need to be turned into a concise test plan and traceable test cases with clear priorities, coverage, and execution focus.
| name | superagents |
| description | 统一路由仓库级请求到合适 lane 与交付深度。用于处理任何仓库内任务，尤其是在需要在 answer/git/github/bug/feature/refactor/review/architecture 之间选择路径、判断是否触发 brainstorming，或协调多 Agent 与多技能执行时。 |
superagents 是仓库内任务的统一编排入口。它负责三件事：先判定任务应该进入哪个 lane，再判定交付深度 Lite / Standard / Full，最后只装配完成当前任务所需的最小 Skill + Agent 组合。
它不替代 answer / git / github / fix-bug / develop-feature / refactor / review-code / architecture-review；这些都只是内部 lane。
以下情况应进入 superagents：
brainstormingrules/*.md、AGENTS.md、CLAUDE.md、skills/*/SKILL.md、Prompt 或 Agent 行为约束以下情况不要把它当成“额外一层”重复调用：
superagents 内部，只需继续当前 lanesuperpowers:using-superpowers，再做任何路由或实现动作。superagents；其他仓库技能只作为内部 lane，不作为外部直达入口。目标 / 范围 / 不做 / 关键假设 / 验收。XY、真实方案权衡、跨模块边界不清，必须切换到 superpowers:brainstorming；进入后按“一次一问”推进。lane + depth + evidence + next，避免黑盒式路由。Standard，并显式纳入 Prompt / Agent Behavior 视角。Route
lanedepthAlign
目标 / 范围 / 不做 / 关键假设 / 验收Compose
Execute
Lite: route -> execute -> reportStandard: research -> plan -> implement -> review -> verify -> reportFull: parallel-research -> plan -> parallel-implement -> dual-review -> full-verify -> reportVerify
file:line 或链接Report
直接执行 / 深度交互 对用户交付| Situation | Lane | Default Depth |
|---|---|---|
| 代码解释、概念问答、轻 Git/GitHub、handoff | answer / git / github / handoff | Lite |
| bug、回归、错误输出、失败路径 | fix-bug | Standard |
| feature、集成、行为变更、refactor | develop-feature / refactor | Standard |
| 代码审查、PR 审查、架构评审 | review-code / architecture-review | Standard |
| 难以稳定分类、存在多方案、边界不清 | research -> planning | Standard |
| 安全、高风险、跨模块复杂并发、规则或 Prompt 行为改动 | 对应 lane | Full 候选，至少 Standard |
Lite 只在命中 rules/fast-path.md 且风险足够低时采用。默认 Standard。
按需读取参考文件，避免把长说明一次性塞进上下文：
references/routing-matrix.md
references/orchestration-playbook.md
templates/universal-engineering-task-prompt.md
references/verification-evals.md
rules / skills / prompts / agent behavior 时读取出现以下信号时，说明路由正在退化：
Fullsuperagents 写成重复 AGENTS.md、rules/* 的总汇编lane / depth 判断Lite 处理对用户的最终输出至少应包含：
Done / Partial / Skipped若没有有效挑战点，深度交互 必须明确写出“当前路径合理，暂无更优替代建议”。
depth 一致rules/output-style.mdrules/deliverable-quality-gate.md详细矩阵、工作流、模板与 eval 规则放在配套参考文件中，按需加载。