Install any skill in seconds. Free to start, no credit card required.
Get Started Free →AI agent and LLM system engineering reference covering single-agent dev (ReAct, tool calling, plan-execute), multi-agent coordination (swarm, role decomposition, file locking), LLM security (prompt injection, jailbreak defense, output filtering), RAG architecture (chunking, hybrid retrieval, rerank), and prompt engineering / evaluation (RAGAS, LLM-as-Judge). Use when building AI agents, designing RAG pipelines, orchestrating multi-agent workflows, hardening LLM apps, or writing prompts.
.claude/skills/telagod-building-agent-systems/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 12% | 0% |
> 判断先于执行:决定「是否做 / 选什么 / 如何取舍」(栈、方案、架构、权衡)前,先读领域判断内核 skills/_kernel/ml/SKILL.md——它管 judgment,本秘典管 execution;冲突时以内核判断为准。
> 单 Agent 是器,多 Agent 是阵。先选规模,再选模式。
| 意图 | 加载 | 核心 | |------|------|------| | 单 Agent 开发(工具调用、ReAct) | agent-dev | ReAct / Plan-Execute / Reflection | | 多 Agent 协同(>=3 文件 or >=2 并行) | multi-agent-coordination | 蚁群仿生、文件锁、依赖图 | | 多 Agent 协议细节(消息素、收阵报告) | multi-agent-protocol | Codex 原生协议、角色定义 | | LLM 安全(注入、越狱、输出过滤) | llm-security | OWASP LLM Top 10 视角 | | RAG 系统(向量、检索、重排) | rag-system | Chunking / 混合检索 / Cohere rerank | | Prompt + 评估 | prompt-and-eval | Few-shot / CoT / RAGAS / LLM-as-Judge |
单步任务(一文件、一查询) → 直接执行(不需要 Agent 框架)
多步任务(计划 + 工具) → 单 Agent (ReAct)
复杂任务(>5 步、需反思) → 单 Agent (Plan-Execute / Reflection)
独立并行任务(>=3 文件、>=2 流) → 多 Agent (TeamCreate)
跨域协作(角色明确) → 多 Agent (角色分工)犹豫时优先 TeamCreate — 串行降级容易,并行升级难。
Prompt 即代码须版控 | 输入输出皆验证 | 成本效果平衡 | 持续评估迭代 | 安全边界明确[unverified]| 信号 | 启用 TeamCreate | |------|-----------------| | 涉及 ≥3 独立文件 | ✅ | | 需 ≥2 并行流 | ✅ | | 总步骤 >10 | ✅ | | 用户明确要求 | ✅ | | 单一探索任务 | ❌(用 explorer 或单 Agent) | | 单文件改动 | ❌(用 worker 或直接执行) | | 单步任务 | ❌(直接执行) |
详细生命周期、文件锁规则、依赖感知、过载保护、降级链:multi-agent-coordination.md
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 23,624 | 24,858 | +5% | 1 | 1 | 0% | 3,686 | 4,328 | +17% | 0 | 0 | — |
case-02 | fail→pass | 33,391 | 23,339 | -30% | 1 | 1 | 0% | 5,402 | 4,724 | -13% | 0 | 0 | — |
case-03 | fail→pass | 24,956 | 21,176 | -15% | 1 | 1 | 0% | 3,778 | 4,360 | +15% | 0 | 0 | — |
case-04 | pass→pass | 16,779 | 15,657 | -7% | 1 | 1 | 0% | 2,492 | 3,249 | +30% | 0 | 0 | — |
case-17 | fail→pass | 14,190 | 14,013 | -1% | 1 | 1 | 0% | 1,891 | 3,077 | +63% | 0 | 0 | — |
case-05 | pass→pass | 12,992 | 8,543 | -34% | 1 | 1 | 0% | 1,998 | 2,315 | +16% | 0 | 0 | — |
case-06 | fail→fail | 4,914 | 7,175 | +46% | 1 | 1 | 0% | 739 | 2,087 | +182% | 0 | 0 | — |
case-07 | fail→fail | 14,013 | 7,597 | -46% | 1 | 1 | 0% | 2,195 | 2,235 | +2% | 0 | 0 | — |
case-08 | fail→pass | 11,401 | 5,843 | -49% | 1 | 1 | 0% | 1,651 | 1,905 | +15% | 0 | 0 | — |
case-09 | fail→pass | 13,667 | 8,882 | -35% | 1 | 1 | 0% | 1,962 | 2,206 | +12% | 0 | 0 | — |
case-10 | fail→fail | 16,141 | 16,466 | +2% | 1 | 1 | 0% | 2,471 | 3,463 | +40% | 0 | 0 | — |
case-11 | fail→pass | 14,561 | 15,345 | +5% | 1 | 1 | 0% | 2,048 | 3,116 | +52% | 0 | 0 | — |
case-12 | fail→fail | 14,842 | 9,195 | -38% | 1 | 1 | 0% | 2,101 | 2,378 | +13% | 0 | 0 | — |
case-13 | fail→fail | 14,136 | 10,368 | -27% | 1 | 1 | 0% | 2,093 | 2,598 | +24% | 0 | 0 | — |
case-14 | fail→pass | 13,700 | 12,708 | -7% | 1 | 1 | 0% | 2,088 | 2,977 | +43% | 0 | 0 | — |
case-15 | fail→pass | 11,002 | 5,152 | -53% | 1 | 1 | 0% | 1,568 | 1,758 | +12% | 0 | 0 | — |
case-16 | pass→fail | 14,370 | 11,546 | -20% | 1 | 1 | 0% | 2,000 | 2,695 | +35% | 0 | 0 | — |
case-18 | pass→pass | 13,585 | 6,052 | -55% | 1 | 1 | 0% | 1,927 | 1,831 | -5% | 0 | 0 | — |
case-19 | pass→pass | 23,082 | 19,064 | -17% | 1 | 1 | 0% | 3,217 | 4,057 | +26% | 0 | 0 | — |
case-20 | pass→pass | 19,887 | 20,608 | +4% | 1 | 1 | 0% | 3,136 | 4,175 | +33% | 0 | 0 | — |
case-21 | pass→pass | 30,321 | 19,102 | -37% | 1 | 1 | 0% | 6,175 | 5,100 | -17% | 0 | 0 | — |
case-22 | pass→pass | 25,756 | 24,545 | -5% | 1 | 1 | 0% | 5,262 | 5,985 | +14% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.