菜单
Use when the user wants a service PRD or PRD-lite for an ML or LLM feature, agent, retriever, evaluation service, or internal ML capability. Focus on business problem, scope, users, KPI, and release acceptance.
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 | ml-llm-prd |
| description | Use when the user wants a service PRD or PRD-lite for an ML or LLM feature, agent, retriever, evaluation service, or internal ML capability. Focus on business problem, scope, users, KPI, and release acceptance. |
Read:
../../../ML LLM Delivery Pipeline/01 PRD Template.md../../../ML LLM Delivery Pipeline/ML LLM Delivery Pipeline and Required Documents.mdUse this skill for business-facing scoping of a single ML/LLM service or capability, not for full architecture.
Do not drift into:
Use when the user wants the rollout and ownership plan for an ML or LLM delivery: scope, workstreams, dependency map, rollout steps, blockers, readiness checklist, and transition notes.
Use when the user wants a quality and release-readiness plan for an ML or LLM service: dimensions, thresholds, datasets, review process, regression rules, and post-release quality review.
Use when the user wants the operational document for an ML or LLM service: ownership, alerts, common failures, incident response, degradation modes, recovery checks, and escalation contacts.
Use when the user wants the integration contract for an ML or LLM service: API/tool methods, input/output schema, data contract rules, prompt hierarchy, retries, confidence routing, and fallback behavior.
Use when the user wants to record one concrete architectural decision in an ML or LLM system, including alternatives, chosen option, rationale, risks, and impacted artifacts.
Use when the task is to create, update, sequence, or review the ML/LLM delivery document set as a pipeline rather than a single document. Covers PRD, system design, ADR, service spec, evaluation plan, delivery plan, runbook, security note, and prompt template.