菜单
Use when approved AI features need architecture decisions for direct calls, RAG, fine-tuning or agents, plus model gateway, knowledge stores, evaluation, observability and tenant security; use HLD for the whole system and agent architecture for tool-using runtimes.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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/peterbamuhigire/srs-skills --skill 11-ai-architecture-spec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。
基于 SOC 职业分类
正在显示 SKILL.md
| name | 11-ai-architecture-spec |
| description | Use when approved AI features need architecture decisions for direct calls, RAG, fine-tuning or agents, plus model gateway, knowledge stores, evaluation, observability and tenant security; use HLD for the whole system and agent architecture for tool-using runtimes. |
| metadata | {"portable":true,"compatible_with":["claude-code","codex"]} |
| Artefact | Source or provider | Required? | Missing behaviour |
|---|---|---|---|
| Approved AI feature requirements and HLD | PRD/SRS and Phase 03 architecture | Required | Stop if feature acceptance or data ownership is unresolved. |
| Data, evaluation, security, latency and cost constraints | AI, data, security and finance owners | Required | Return architecture options and verification gaps when current provider facts are unavailable. |
| Artefact | Consumer | Observable acceptance condition |
|---|---|---|
| AI Architecture Specification and ADR seeds | AI, data, service, security, test and operations teams | Each feature maps to a justified pattern; gateway, data boundary, evaluation, fallback, observability, cost and tenancy controls have testable contracts. |
| Evidence | Consumer | Acceptance condition |
|---|---|---|
| Source and decision trace | Reviewer and downstream owner | Each material statement cites an approved input, named decision or qualified open issue. |
| Completed verification record | Release or phase gate owner | Every applicable check records pass/fail; unavailable checks remain not assessed. |
Read-only is the default for analysis, review, evaluation and planning. Read and search access to authorised project artefacts are required. Editing is limited to an explicitly requested project deliverable. Execution may run document, syntax or validation checks. Network access is used only for facts that require current verification. Do not publish, spend, change production, approve policy, or claim certification without explicit authority.
If any required capability is unavailable, return the narrowest useful qualified AI Architecture Specification and ADR seeds draft plus a gap register showing the missing item, affected sections, risk and owner. Never convert an unassessed check into a pass.
| Choice | Action | Failure or risk avoided |
|---|---|---|
| Need is grounded retrieval over governed sources | Choose RAG with citation/evaluation path | Fine-tuning does not mask knowledge freshness |
| Task needs bounded tool planning | Route to agent architecture | Agent controls are not omitted |
The AI-distinctive architecture artefact. Sits alongside the multi-tenancy spec and the generic HLD. Captures the model gateway, vector store, eval harness, prompt registry, observability bus, and the multi-tenant AI security boundaries.
Read HLD, multi-tenancy spec, AI feature PRD spec, AI data spec. Identify in-scope AI features, models, patterns, and the tenant boundaries.
Include the problem/non-AI alternative, human and affected-non-user map, model/system/input/output boundaries, human control, consent/notice, drift signals, staged promotion, rollback and re-promotion criteria. Use the oversight reference before selecting a runtime pattern.
The AI plane is a sub-set of the application plane plus a small set of dedicated control-plane services:
Diagram with Mermaid; place every AI service.
For each AI feature select the pattern:
| Pattern | When | Components |
|---|---|---|
| Direct LLM call | input is self-contained, no external data | gateway + prompt + model |
| RAG | grounding in customer data | gateway + retrieval + reranker + prompt + model + citation post-processor |
| Agent | multi-step, tool-using, planned | gateway + planner + tool catalogue + executor + audit log + per-step approval UI |
| Fine-tune | repetitive narrow task, cost reduction | training pipeline + model artefact + eval suite + rollback artefact |
| Classical ML | structured prediction | feature store + model artefact + monitoring |
State the verdict per feature with rejected alternatives.
The gateway is the sole egress to model providers. Capture:
For each retrieval index: store technology, partitioning model (per-tenant index / namespace / metadata-filter), embedding model + version, dimensions, ANN parameters, freshness, encryption posture, key management.
Versioned, tagged, changes proposed via PR with regression eval attached. State the registry source-of-truth, deploy pipeline, rollback procedure.
The eval harness is a first-class production system, not a notebook. State: dataset store, judge-LLM, CI gate hook, scheduled regression, alerting on score drop.
AI-specific signals: tokens in/out per request, model latency per provider, fallback rate, abstention rate, citation rate, judge-LLM score, cost per tenant per feature, content-filter trips, red-team alerts.
ADR seeds: model choice per feature, RAG-vs-fine-tune, vector store choice, eval threshold, abstain policy, content filter, fallback policy.
AI_Architecture_Spec.md sections: 1) AI Plane Diagram, 2) Feature-to-Pattern Map, 3) Model Gateway, 4) Vector Store, 5) Prompt Registry, 6) Eval Harness, 7) Observability, 8) Security Boundaries, 9) ADR Seed Index, 10) Traceability.
