Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/seed-forge/harness-ai-kit/devlab-ai-agent-engineeringnpx skills add seed-forge/harness-ai-kit --skill devlab-ai-agent-engineeringgit clone --depth 1 https://github.com/seed-forge/harness-ai-kitWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/seed-forge/harness-ai-kit/devlab-ai-agent-engineering)<a href="https://agentmods.dev/skills/seed-forge/harness-ai-kit/devlab-ai-agent-engineering"><img src="https://agentmods.dev/badge/skills/seed-forge/harness-ai-kit/devlab-ai-agent-engineering.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00135 | $0.02423 |
| Opus 5 | $0.00068 | $0.01211 |
| Sonnet 5 | $0.00027 | $0.00485 |
| Haiku 4.5 | $0.00014 | $0.00242 |
Grade A, and why
devlab-ai-agent-engineering scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 148 lines — stays where its author put it; the contents beside it link to each section on GitHub.
devlab-ai-agent-engineering
用途
把"要做一个 AI 智能体应用"从零散试错升级为可复用的工程化方法论。回答两类问题:
- 拆分:一个 AI Agent 应用应该拆成哪几层/哪几块?各层职责与边界是什么?
- 避坑:LLM/Agent 工程里反复踩的坑,如何在设计期就规避(不是给某个具体 bug 的补丁,而是给方法论原则)。
本技能是方法论层(讲原则与结构),具体某个技术栈的落地细节交由 AI Agent 结合项目给出选项/建议;各业务领域的落地范例见
references/。
适用场景
- 新建 NL2SQL / Text2SQL、RPA、语音数字人、RAG 问答等任意 LLM 驱动应用,需要先定架构。
- 现有 Agent 应用"越改越乱"、规则与 LLM 边界混乱、Prompt 散落、模型成本失控,需要重新梳理分层。
- 评审一个 AI 应用设计方案是否踩了已知的结构性坑。
不适用场景
- 纯粹的模型微调/训练(本技能聚焦应用工程,非模型训练)。
- 具体某框架的 API 用法(交由
-usage类技能或 AI 现场给建议)。 - 非 AI 的普通后端/前端架构(用对应 devlab-srv-* / devlab-web-* 技能)。
输入
- 业务目标与输入/输出形态(自然语言查询?桌面操作?语音对话?)。
- 现有代码/架构现状(可选,用于重构场景)。
- 约束:延迟预算、成本预算、可解释性/审计要求、私有化程度。
输出
- Agent 分层架构方案(各层职责 + 边界 + 数据流)。
- 规则 vs LLM 的分工决策 + 兜底/超时/缓存策略。
- 多模型触点路由表(每个触点的 primary/fallback 模型与选型理由)。
- Prompt 治理方案(统一注册/版本/复用)。
- 结构性风险清单(对照"已知坑方法论")。
核心方法论
1. 分层管道(Layered Pipeline)
把 Agent 拆成单向、可独立测试的阶段,每阶段有明确输入/输出契约:
输入理解 → 信息抽取 → 策略决策 → 执行/编译 → 结果验证 → 输出
│ │ │ │ │
意图识别 槽位/条件 按类型选策略 调用/生成 正确性校验
知识增强检索 metric抽取 strategy- (SQL/动作/ (schema/
实体消歧 per-type TTS) 断言)
- 中间表示解耦:在"理解"与"执行"之间引入中间表示(如 DSL),让 NLP 层与工程实现层解耦,便于多目标复用(如一份 DSL 编译到多种 SQL 方言)。
- 按类型分策略:先做请求分类(如明细/指标/排名/对比),再路由到 strategy-per-type,避免一个巨型分支处理所有情况。
- 每层可独立评测:每个阶段都能单独喂输入、断言输出,是后续评测闭环的基础。
2. 规则优先 + LLM 兜底(Hybrid)
- 默认 rule_first:规则/检索能确定的走确定性路径(快、稳、可解释、零成本)。
- 规则置信度低于阈值或产出不足时,才触发 LLM 兜底(慢路径)。
- 策略可配:
rule_first/llm_first/parallel/adaptive。 - ⚠️ 兜底触发阈值必须语义正确:典型坑是"最小条件数=0"导致"0 条件也算充足"→ 兜底永不触发。阈值要表达"何时判定规则不足",而非形式化默认值。
3. 多模型触点路由(Touchpoint Routing)
- 每个 LLM 触点独立配置
primary_model+fallback_models。 - 按难度分配模型:简单触点用便宜/快模型,复杂触点(如复杂条件抽取)用强模型。
- 路由集中在配置,不在代码里散落硬编码模型名。
4. Prompt 统一注册治理
- 所有 prompt 集中注册管理(统一命名/版本/复用),禁止散落在各处字符串。
- 新增触点的 prompt 必须遵循已有注册规范与配置结构,保持一致性。
5. 韧性三件套:超时 / 缓存 / 降级
- 超时:LLM 调用必设超时;超时值要匹配真实耗时(典型坑:默认 5s 太短,返回 200 但未完成即 timeout)。
- 缓存:可缓存的触点开启缓存,降本增稳。
- 降级:LLM 不可用/超时时有确定性兜底或明确的失败语义,不能静默丢结果。
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 148 lines · 135 tokens per session scan A 324e1cf2dc5f
devlab-ai-agent-engineering is a skill published in the GitHub repository seed-forge/harness-ai-kit (22 stars, last pushed 4d ago), licensed Apache-2.0. It adds 135 tokens to every session and 2,423 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
audit-langfuse-llm
Run a PDCA quality audit on LLM/AI features: traces, prompts, costs, evals, grounding, hallucination. Use for "audit LLM quality", "check Langfuse", "audit prompts", "check AI quality", "audit AI costs", "check traces". Jailbreak/OWASP LLM → audit-llm-security. Token caps → plan-llm-cost-guardrails.
Prompt Refiner
Improves AI prompts to be clearer, more specific, and produce more consistent outputs.
Prompt Injection Defense Auditor
Reviews LLM application prompts and input handling for direct and indirect prompt injection vulnerabilities, then writes defensive scaffolding.
audit-llm-security
Read-only OWASP LLM Top 10 audit of app-facing AI: prompt injection, data leakage, unsafe output/agency, RAG risks, misinformation, and unbounded spend. Use when "audit LLM security", "prompt injection", "jailbreak my chatbot", or "is my AI safe?". General app security → audit-security.
prompt-engineer
Writes, refactors, and evaluates prompts for LLMs — generating optimized prompt templates, structured output schemas, evaluation rubrics, and test suites. Use when designing prompts for new LLM applications, refactoring existing prompts for better accuracy or token efficiency, implementing chain-of-thought or few-shot…
prompt-master
Generates optimized prompts for AI tools. Activates only when the user explicitly asks to write, fix, improve, or adapt a prompt for a specific AI tool (LLM, Cursor, Midjourney, image AI, video AI, coding agents, etc.). Does not activate for general conversation, coding tasks, document writing, or other…