菜单
Investigate a bug with competing hypotheses and parallel subagents. Use when the user asks to debug an error, find a root cause, or investigate why something is failing.
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
Review Harbangan plans and code with read-only advisory agents. Use `--plan <path>` for domain-driven plan review, and `--code [target]` for fixed reviewer-dimension code review with domain-advisor consultation.
Update project documentation by consulting all domain agents for accuracy. document-writer leads, other agents provide technical review. Use when user says 'update docs', 'write documentation', 'refresh docs', 'document this', or 'sync documentation'.
ACH-based debugging — spawns domain investigators to evaluate competing hypotheses across the Harbangan stack using formal evidence standards and arbitration. Use when user says 'debug this error', 'why is this failing', 'investigate bug', 'root cause analysis', or 'something is broken'.
Full lifecycle feature implementation — spawns teams, assigns tasks, monitors progress, verifies quality, and creates PRs. Agents remain idle after completion — use /team-shutdown to terminate. Use when user says 'implement this', 'build this feature', 'start working on X', or 'execute the plan'.
Analyze scope, explore codebase, and produce implementation plans. Spawns all 7 domain agents for parallel investigation from their domain perspective, and writes structured plans. Use when user says 'plan this feature', 'analyze scope', 'what would it take to build X', or 'explore before implementing'.
Launch a multi-reviewer parallel code review organized by quality dimensions (Security, Performance, Architecture, Testing, Accessibility). Use when user says 'review this code', 'security review', 'review my PR', 'code quality check', or 'architecture review'.
| name | team-debug |
| description | Investigate a bug with competing hypotheses and parallel subagents. Use when the user asks to debug an error, find a root cause, or investigate why something is failing. |
Run a structured, read-first debugging pass before making fixes.
.claude/agents/*.md ownership guidance, and references/hypothesis-testing.md..qwen/agents/, choosing the smallest relevant subset of:
rust-backend-engineerreact-frontend-engineerdatabase-engineerdevops-engineerbackend-qafrontend-qadocument-writer for clarity, docs, or expectation mismatches when relevantfile:line references and explicit verdicts such as confirmed, probable, inconclusive, or ruled out.