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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/andrem-sec/psc-comet --skill prd命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | prd |
| description | Product Requirements Document — specify and verify before implementing |
| version | 0.1.0 |
| level | 3 |
| triggers | ["prd","write requirements","spec this out","requirements first","before we build"] |
| pipeline | ["prd","plan-first"] |
| context_files | ["context/user.md"] |
| steps | [{"name":"Capture Intent","description":"What is this feature or change trying to accomplish? State the goal in one sentence."},{"name":"Scope Boundaries","description":"What is explicitly in scope? What is explicitly out of scope?"},{"name":"Draft Acceptance Criteria","description":"Write specific, verifiable criteria — not generic scaffolding"},{"name":"Refine Criteria","description":"Apply the quality gate — each criterion must be independently verifiable without subjective judgment"},{"name":"Identify Dependencies","description":"What must exist or be true before this can be built?"},{"name":"Confirm","description":"Present the PRD. Get explicit approval before handing off to plan-first."}] |
Define what to build and how to verify it before writing a line of code. The output is a confirmed specification that plan-first uses as its source of truth.
Without a PRD, implementation begins on an implicit spec. The spec lives only in the user's head and Claude's initial interpretation. When they diverge — and they always diverge on complex features — the mismatch is discovered at the end of implementation, not the beginning.
Auto-generated acceptance criteria are deliberately generic. They must be replaced with task-specific, independently verifiable criteria before the PRD is confirmed.
WRONG (scaffold — reject these):
- Implementation is complete
- Code compiles without errors
- Feature works as expected
RIGHT (specific, verifiable):
- POST /api/orders returns 201 with order ID when payload is valid
- POST /api/orders returns 422 with field-level errors when required fields are missing
- Order is written to orders table with status="pending" and timestamp within 1s of request
- Test file exists at tests/api/test_orders.py and all tests pass
The test: can you write a passing/failing test for this criterion without asking any clarifying questions? If yes, it is a good criterion. If no, refine it.
## PRD: [feature name]
Date: [date] | Status: DRAFT → CONFIRMED
### Goal
[One sentence — what problem does this solve?]
### Scope
In: [explicit list of what is included]
Out: [explicit list of what is excluded]
### Acceptance Criteria
1. [Specific, verifiable criterion]
2. [Specific, verifiable criterion]
3. [Specific, verifiable criterion]
### Dependencies
- [What must exist before this can be built]
### Open Questions
- [Anything that needs resolution before starting]
### Confirmation
Confirmed? (yes / modify / no)
After PRD is confirmed, pass it directly to plan-first. The plan should reference the acceptance criteria by number.
Do not start implementing while the PRD is still in DRAFT status.
Do not accept criteria you cannot write a test for. Push back and refine until each criterion is independently verifiable.
Do not include implementation details in the acceptance criteria — specify behavior, not mechanism.