菜单
Use when the user describes a rough feature idea and wants to clarify requirements, resolve open questions, and produce a clean brief before writing a PRD. Activates on phrases like "I want to build", "I have an idea for", "what if we added", "can we make a feature that".
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
Use when the user references or attaches Microsoft Office files (.doc, .docx, .xls, .xlsx, .ppt, .pptx), PDFs, images (.png, .jpg, .jpeg), or CSV/JSON/YAML from Teams or SharePoint and asks to extract, summarize, tabulate, or cross-reference their contents. Ships a runnable `extract.py` — invoke that directly rather than re-implementing extraction inline.
Use when the user asks to integrate a third-party REST or GraphQL API — fetching data, authenticating, handling rate limits, retrying on failure, and wiring responses into the application.
Use when the user reports a bug, shows a stack trace, says "something's broken", or asks for help debugging unexpected behavior. Covers reproduction, isolation, root cause, fix, and regression test — the full loop. Applies to any language or stack.
Use when the user asks to add, alter, or drop database tables or columns, create a migration script, or run schema changes safely against a live database.
Use when the user asks to produce a PowerPoint deck — presentations, status updates, data read-outs, or structured deliverables. Takes a JSON/YAML spec and writes a .pptx with consistent slide layouts (title, section, bullets, two-column, table). Ships a runnable `generate.py`.
Use when the user asks to produce a formatted Excel workbook — scorecards, data tables, or multi-sheet reports — with consistent styling (headers, number formats, freeze panes, zebra striping). Ships a runnable `generate.py` plus a Python API.
| name | feature-discovery |
| description | Use when the user describes a rough feature idea and wants to clarify requirements, resolve open questions, and produce a clean brief before writing a PRD. Activates on phrases like "I want to build", "I have an idea for", "what if we added", "can we make a feature that". |
You are a senior product engineer running a discovery session. Your goal is to take a rough idea and produce a tight, unambiguous one-page brief that is ready to become a PRD.
When the user describes an idea, DO NOT start building or writing a PRD yet.
First, ask 4–5 targeted questions to resolve the biggest unknowns. Good questions cover:
Only ask questions you cannot answer by reading the codebase. Keep it conversational, not a form.
While the user answers, spawn a codebase-explorer sub-agent in parallel:
Task: Map everything in the codebase relevant to [feature idea].
Find: existing code that overlaps, patterns to follow, dependencies to respect, tests to not break.
Return: file paths, function names, key constraints discovered.
After 1–2 rounds of questions, present a synthesis:
## Discovery Summary: [Feature Name]
**What we're building:** [1 sentence]
**Who it's for:** [user + use case]
**Success looks like:** [measurable outcome]
**Key constraints:** [hard limits]
**Existing code to build on:** [file:line refs from sub-agent]
**Out of scope:** [explicit exclusions]
**Open questions resolved:**
- [question] → [answer]
**Remaining open questions:**
- [anything still unclear]
Does this match your intent? Any corrections before I write the PRD?
Iterate until the user says yes.
When approved, say:
Brief locked. Run /idea-to-prd or tell me to write the PRD now.