菜单
Conducts structured discovery interviews, produces a versioned REQUIREMENTS.md, identifies unknowns and risks, and generates a go-beast handoff plan for a software project.
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
| name | go-hawk |
| version | 1.1.0 |
| description | Conducts structured discovery interviews, produces a versioned REQUIREMENTS.md, identifies unknowns and risks, and generates a go-beast handoff plan for a software project. |
| when_to_use | Use when starting a new project, feature, or epic — or whenever the problem is underspecified and scope must be established before any implementation begins. Invoke before go-fox. |
go-hawk is the entry point of the go-beast pack. Its job is to turn a vague idea into a scoped, documented requirement that every other beast can act on without second-guessing.
User: "I want to build a task manager app."
→ invoke go-hawk
→ structured interview → requirements doc → handoff plan
Ask the following in order. Do not skip. Do not answer on the user's behalf.
Write a REQUIREMENTS.md in the project root:
# Requirements — <Project Name>
> version: 1 | date: YYYY-MM-DD | status: draft
## Problem statement
## Users and roles
## Functional requirements (numbered, testable, P0/P1/P2)
## Non-functional requirements (performance, security, availability)
## Out of scope
## Open questions
## Risks
Each functional requirement must be specific, testable, and prioritized.
List every open question that must be resolved before architecture begins. Flag them explicitly — do not let them slip into the backlog.
Recommend the go-beast sequence for this specific project. Typical:
go-hawk → go-fox → go-otter → go-beaver → go-wolf + go-lynx → go-eagle → go-bear → go-raven → go-owl
Adjust based on actual requirements. Always call go-bear early if auth, PII, or payments are in scope.
Pack meta-skills (load on demand, not in sequence):
go-mole — project briefing before any sessiongo-jay — AI context file editinggo-smith — create or audit skills for the go-* familygo-swift — Claude Code hook authoring [Claude Code only]REQUIREMENTS.md — versioned, dated, signed offConducts structured technical conversations — code walkthroughs, architectural debates, decision support, rubber-duck debugging, and Q&A — using mode-aware dialogue, one-question-per-turn discipline, explicit assumption labeling, and escalation signals that hand off to the right beast when a conversation has produced enough to warrant a formal artifact.
Conducts a multi-pass code review with structured scoring across seven dimensions (correctness, security, maintainability, testability, performance, design, observability), applies a calibrated 0–4 rubric per dimension, produces OIR-structured findings with BLOCKER/WARNING/SUGGESTION/NIT severity classification, and emits a SCORE_REPORT.md with a dimensional scorecard and a merge verdict.
Translates approved requirements into an architecture decision record (ADR), technology stack selection, Mermaid component diagram, and interface contracts.
Audits an existing system's architecture across five dimensions — structure, observability, reliability, scalability, and security posture — and produces a prioritized HTML report with findings, capability gaps, and concrete improvement proposals. Uses repomix to pack the codebase, reads ADRs and CONTEXT.md when present, and opens the report in the browser.
Designs entity-relationship models, defines schemas with conventions, writes safe migrations with rollback, plans indexing strategy, and reviews queries for N+1 and sequential scans.
Translates interface contracts and functional requirements into BDD scenario files (Given/When/Then), an acceptance test skeleton, and a SPEC.md that go-wolf and go-lynx must satisfy before implementation begins.