intent-and-automation

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Skill: TraceMem Intent and Automation Modes

Purpose

用途

This skill explains how to choose the correct

intent

and

automation_mode

for a Decision Envelope. Correct classification is critical for potential policy checks and audit clarity.

本Skill说明如何为Decision Envelope选择正确的

intent

和

automation_mode

。正确分类对于潜在的策略检查和审计清晰度至关重要。

When to Use

使用场景

When calling
```
decision_create
```
and you need to populate the arguments.
When determining if a task requires human-in-the-loop (
```
propose
```
) or can be done automatically (
```
autonomous
```
).

调用
```
decision_create
```
并需要填充参数时。
确定任务是否需要人工介入（
```
propose
```
模式）或可自动执行（
```
autonomous
```
模式）时。

When NOT to Use

不适用场景

Do not invent new automation modes. Stick to the strict list.

请勿自创新的自动化模式，严格遵循现有列表。

Core Rules

核心规则

Intent is Hierarchical: Intents must be dot-separated strings ordered from general to specific (Category -> Entity -> Action). Example:
```
customer.order.refund
```
.
Automation Mode is Binding: The mode you select declares your authority level.
- ```
propose
```
  : You will only read data and suggest actions. You will NOT write/execute.
- ```
approve
```
  : You will execute, but only after explicit human approval (often enforced by policy).
- ```
override
```
  : You are explicitly breaking a rule (requires high permission).
- ```
autonomous
```
  : You will execute immediately without human intervention.

Intent具有层级结构：Intent必须是点分隔的字符串，按从通用到具体的顺序排列（类别 -> 实体 -> 操作）。示例：
```
customer.order.refund
```
。
自动化模式具有约束性：你选择的模式声明了你的权限级别。
- ```
propose
```
  ：仅读取数据并建议操作，不进行写入/执行。
- ```
approve
```
  ：将执行操作，但必须经过明确的人工批准（通常由策略强制执行）。
- ```
override
```
  ：明确打破规则（需要高权限）。
- ```
autonomous
```
  ：无需人工干预，立即执行操作。

Correct Usage Pattern

正确使用模式

Choosing an Intent

选择Intent

Structure:

<Domain>.<Entity>.<Action>

Good:

security.access_log.scan

billing.invoice.void

support.ticket.reply

Bad:
```
scan_logs
```
,
```
fix_thing
```
,
```
decision_1
```
.

结构：

<领域>.<实体>.<操作>

正确示例：

security.access_log.scan

、

billing.invoice.void

、

support.ticket.reply

。

错误示例：
```
scan_logs
```
、
```
fix_thing
```
、
```
decision_1
```
。

Choosing Automation Mode

选择自动化模式

Are you just looking?
- Mode:
```
autonomous
```
  (if read-only duties are pre-approved) or
```
propose
```
  (if you are just gathering info for a human).
Are you planning to change state (write/delete)?
- If you need permission first: Mode
```
propose
```
  (stop after planning).
- If you have permission but need confirmation: Mode
```
approve
```
  (TraceMem might force this anyway).
- If you are fully trusted logic: Mode
```
autonomous
```
  .

仅查看数据？
- 模式：
```
autonomous
```
  （如果只读职责已预先批准）或
```
propose
```
  （如果仅为人工收集信息）。
计划更改状态（写入/删除）？
- 若需先获得权限：模式为
```
propose
```
  （完成规划后停止）。
- 若已有权限但需确认：模式为
```
approve
```
  （TraceMem可能会强制要求此模式）。
- 若为完全可信的逻辑：模式为
```
autonomous
```
  。

Example

示例

json

{
  "intent": "network.firewall.block_ip",
  "automation_mode": "autonomous",
  "actor": "security-agent-v2"
}

json

{
  "intent": "network.firewall.block_ip",
  "automation_mode": "autonomous",
  "actor": "security-agent-v2"
}

Common Mistakes

常见错误

Mismatched Mode: interacting as
```
propose
```
but then trying to
```
decision_write
```
(TraceMem may block this or flag it as a violation).
Inconsistent Intents: Using
```
user.create
```
in one decision and
```
create.user
```
in another. Be consistent.

模式不匹配：以
```
propose
```
模式交互，但尝试执行
```
decision_write
```
（TraceMem可能会阻止此操作或将其标记为违规）。
Intent不一致：在一个决策中使用
```
user.create
```
，在另一个决策中使用
```
create.user
```
。请保持一致性。

Safety Notes

安全注意事项

Policy Triggers: Policies are often attached to specific intents. Using the wrong intent might bypass safety checks or trigger unnecessary alarms.
Escalation: If you start as
```
autonomous
```
but policy returns
```
requires_exception
```
, you effectively switch to an approval flow. The initial mode was your desired mode.

策略触发：策略通常与特定Intent关联。使用错误的Intent可能会绕过安全检查或触发不必要的警报。
升级流程：如果初始选择
```
autonomous
```
模式，但策略返回
```
requires_exception
```
，则会自动切换到审批流程。初始模式是你的期望模式。