Instruction Clarity

---
id: instruction_clarity
name: Instruction Clarity
---

# Instruction Clarity

Evaluates whether the plugin's prompts (SKILL.md, agent .md files, command .md files)
are clear, specific, and unambiguous enough that an LLM will follow them consistently.

## What to look for

- **Structured phases or steps** — Does the plugin break work into defined stages?
- **Explicit scope** — Is it clear what the plugin does and does NOT do?
- **Output format specification** — Does it tell the LLM exactly how to format responses?
- **Unambiguous language** — Are instructions precise, or could an LLM interpret them multiple ways?
- **Role definition** — For agents, is the persona/mindset clearly established?

## Scoring rubric

| Score | Criteria |
|-------|----------|
| **5** | Numbered workflow phases or clearly structured steps (sub-steps not required for simple plugins). Explicit scope boundaries. Defined output format with example. MUST/MUST NOT rules. |
| **4** | Clear role definition. Specific do/don't rules. Structured sections. Output format mentioned but not demonstrated. |
| **3** | Instructions tell the agent its goal, but lack phases, scope boundaries, or output format. Agent can complete the task but results vary between runs. |
| **2** | Instructions are <10 lines or use vague language ("do a good job", "be helpful"). Agent will guess at behavior. |
| **1** | No SKILL.md / agent .md, or only a title with no instructions. |

## What a 5 looks like

- SKILL.md with numbered workflow phases
- Explicit "you MUST" / "you MUST NOT" rules
- Defined output format with example
- Scope boundaries ("only review files in the PR, not the entire repo")

## Scoring examples

### Score 5 — deep-review