代理实施流程
- 作者仓库星标 0
- 作者仓库 skills-registry
- 领域
- 人工智能
- 作者声明的智能体
- 未找到明确声明;不据此推断已兼容或已测试
- 静态检查
- 88 / 100 · 启发式扫描,不代表运行安全
- 运行环境证据
- 未执行;当前没有隔离运行测试证据
- 作者 / 版本 / 许可证
- @tomevault-io · 未声明许可证
- 上下文用量评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部接口密钥
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 运行环境要求
- 未检出明确要求
- 检测到的文件与系统行为
-
- 只读
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 无(仅作为资料)
需要注意: 未声明工具权限范围;实际权限由运行环境决定。
# AI Agent Implementation Workflow
## What This Skill Produces
Use this skill to implement or update agent behavior with stable contracts. The expected result is:
- explicit agent goal and decision boundaries
- predictable tool invocation flow
- schema-safe outputs for downstream consumers
- graceful fallback when tools/providers fail
- targeted tests or smoke checks proving behavior
## When to Use
Use this skill when you need to:
- add a new agent workflow
- refine tool usage logic for an existing agent
- fix agent output/schema instability
- harden retry/fallback behavior in agent loops
- align agent behavior across CLI/API/UI surfaces
Common trigger phrases:
- "implement this agent behavior"
- "add tool-calling to the agent"
- "fix unstable agent responses"
- "make the agent robust"
- "agent output schema keeps breaking"
## Procedure
1. Define contract first
- Lock input/output schema and required fields.
- Clarify what is best-effort vs required behavior.
2. Constrain tool boundaries
- List which tools can be called and for what reasons.
- Keep side-effecting actions explicit and auditable.
3. Implement minimal orchestration
- Prefer small deterministic control flow over deep branching.
- Make retries bounded and reason-aware.
4. Handle degraded mode intentionally
- Return actionable errors when hard requirements are missing.
- Use safe fallback only when it preserves contract meaning.
5. Verify behavior incrementally
- Add focused tests/smokes for primary path + fallback path.
- Confirm output schema is stable across paths.
6. Validate integration surface
- Ensure consuming endpoints/UI can parse new outputs.
- Avoid silent breaking changes in event/JSON structure.
## Quality Checks
Before finishing, confirm that:
- output schema is deterministic and documented
- tool usage boundaries are explicit
- retries/fallbacks are bounded and observable
- failures are actionable, not silent
- integration consumers remain compatible
---
> Source: [Bryan-Roe/Aria](https://github.com/Bryan-Roe/Aria) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->
章节与文件索引
按说明文件中的顺序排列。
这项技能会产生什么
何时使用
程序
质量检查
文件/命令 -> tools/providers · output/schema · retry/fallback · CLI/API/UI · input/output · tests/smokes · endpoints/UI · event/JSON
内容校验值 -> 955dfe675356
当前原文的自动摘译,尚未人工审校。
这项技能会产生什么
使用此技能通过稳定的合约来实施或更新代理行为。预期结果是: 明确的代理目标和决策边界 可预测的工具调用流程 为下游消费者提供架构安全的输出 当工具/提供者失败时进行优雅的回退
何时使用
当您需要执行以下操作时,请使用此技能: 添加新的代理工作流程 细化现有代理的工具使用逻辑 修复代理输出/架构不稳定性 强化代理循环中的重试/回退行为
程序
首先定义合约锁定输入/输出模式和必填字段。澄清什么是尽力而为与要求的行为。约束工具边界 列出可以调用哪些工具以及出于什么原因。
质量检查
在完成之前,请确认: 输出模式是确定性的,并且记录的工具使用边界是明确的重试/回退是有限的,并且可观察到的故障是可操作的,而不是静默的