aiproof-cli 0.1.2

CLI for aiproof, the static analyzer for AI prompts (ESLint for prompts).
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

# AIP020: system-message-overloaded

**Severity:** Info | **Category:** Efficiency

System prompts that try to do too much increase cognitive load on the model. This rule flags system messages that exceed size or imperative density thresholds, suggesting a refactor into role-specific instructions or multi-turn setup.

## Detection

The rule applies two independent gates to system-role prompts:

- **Gate A (Size):** Flags if word count exceeds ~1500 tokens
- **Gate B (Imperatives):** Flags if total imperative clauses (you must, always, never, etc.) exceed 8

If either gate fires, an Info diagnostic is emitted.

## Example

**Bad:**
```
You are a careful assistant. You must verify facts. You should check sources. 
Always cite your work. Never make assumptions. Please be clear. Ensure accuracy.
Do not speculate. Don't guess. Make sure to...
```

**Good (split into role + runtime guidance):**
```
You are a careful assistant.
```
Then in the first user message:
```
Please verify all facts and cite sources. Always check before answering.
```

## References

- Token optimization practices for prompt engineering