agent-runtime skillA
agent-runtime is agent-read markdown (skill) from yuecao365/offercome: Agent 运行时深挖：循环与事件、工具协议与沙箱、子 agent、预算终止、恢复、输出契约。.
Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.
What the file says
## 面试官在意什么 这本是 ai-agent 的运行时那一层：模型外面那圈代码怎么写。面试官要的不是框架名，而是候选人自己写过或改过循环之后留下的判断：状态存哪、预算超了怎么办、工具报错怎么回给模型、跑一半停了怎么续、换个服务商结构化输出为什么就坏。追问的方式是给一个故障现象让候选人定位到层：死循环是终止条件还是工具结果的问题，上下文爆是历史没裁还是工具结果太长，重放不一致是状态存了两份还是事件缺字段。能把 Claude Code、Codex、Manus 这类公开资料里的做法和自己的取舍对上的，算实。 ## 项目 / 实习怎么深挖 简历上出现下面这类经历时从哪里切、追什么。追到候选人能说出机制、数字的来源与一次真实的故障或取舍才算实；只有框架名与结论、说不出自己那一段的，记为危险信号。通用的追问方法见 project-deep-dive。 - 简历出现自研 agent 循环 / 运行时 → 追一步的定义、状态存在哪、预算超了发生什么、能不能从某一步续跑、最长一次跑了多少步、每步的 token 与缓存命中 - 简历出现工具权限 / 沙箱 → 追分了几档、谁批准、误调过什么、执行类工具隔离了什么边界、超时怎么处理 - 简历出现多 agent / 子 agent → 追为什么拆、子 agent 的上下文与预算、产出怎么回主流程、协作失败怎么定位 - 简历出现 checkpoint / 断点续跑 → 追存了什么、哪些不能重跑、写操作的幂等键、一次真实的恢复 - 简历出现 trace / 回放 / 调试工具 → 追记了哪些事件、按步回放怎么做、靠它定位过什么故障 - 简历出现多服务商 / 结构化输出 → 追哪个服务商坏过、修在哪一层、收敛与抢救的顺序 ## 常见失守与危险信号 - 运行时循环与状态：循环全交给框架的 maxIterations，说不出超限后发生了什么；状态既存消息列表又存事件，两边对不上 - 工具协议、权限分级与沙箱：工具越多越强；敏感工具没有确认或 dry-run；工具报错把栈信息原样喂回模型；沙箱只说"用 Docker" - 子 agent 与编排：为了"多"而多 agent；子 agent 与主 agent 共享同一份无限增长的上下文；说不出协作失败时看哪个 agent 的轨迹 - 中断恢复与持久化：说"重跑一遍就行"；把重试和续跑混为一谈；不知道写操作要幂等键 - 事件日志与轨迹可观测：只记最终输出不记步；trace 里没有每步的 token、缓存、耗时；定位故障靠重跑 - 预算与终止条件：只有步数上限；预算超了直接抛错给用户；没想过给模型一步"不许调工具直接结论" - 结构化输出契约与多服务商：只会"加一句请输出 JSON"；不知道 schema 是否随请求下发；把服务商差异修在每个 agent 里 …
Read the whole file at its exact version.
How to install
mdr add yuecao365/offercome/agent-runtime@git:20260920.a4843f5mdr add yuecao365/offercome/agent-runtime@sha256:7bf220de77edfccdPin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.
[](https://markdownregistry.com/a/art_enmo3u6kwlvvintn)
1 badge views in 30 days
Versions
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (8824 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
yuecao365/offercome · 23 stars · license MIT · pushed 2026-09-23 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_enmo3u6kwlvvintn GET https://markdownregistry.com/api/v1/resolve?ref=yuecao365/offercome/agent-runtime GET https://markdownregistry.com/api/v1/blob/7bf220de77edfccd7c8e47e71ce1df78e81bc539c953f7edc13047d6ff44b9d9
Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.