| name | prompt-quality |
| description | Checks whether a task prompt follows the anatomy-of-a-prompt guidelines (9 layers: Task, Context Files, Reference, Success Brief, Rules, Conversation, Plan, Alignment, Push Intent). Auto-invoked before responding to any new task. Outputs a quality report — confirmed or missing layers — then proceeds without blocking. Never invoke for follow-ups, continuations, or responses to clarifying questions.
|
Prompt Quality
Analyze the incoming prompt against the anatomy-of-a-prompt guidelines and output a quality report as the very first content of your response.
When to run
Run on every message that introduces a new task. Skip for:
- Short follow-ups ("yes", "do it", "go ahead", "ok", "sure", "looks good")
- Continuations that reference prior context without introducing a new goal
- Responses to clarifying questions you asked
Step 1 — Classify the task
Simple: single-step action, quick question, rename, small bug fix, research lookup
Complex: multi-step implementation, feature creation, architecture decision, skill creation, multi-file refactor, full design task
Complexity signals: "implement", "build", "create", "design", "refactor across", "skill", "feature", "architecture", multi-file scope, multi-step scope.
Step 2 — Check required layers
Simple tasks require layers 1, 3, 4, 5. Complex tasks require all 8. Layer 9 is conditional — required whenever the task will modify files in a git repository.
| Layer | Name | Present when |
|---|
| 1 | Task | States what to do AND why / what success looks like — "so that…", "in order to", "the goal is", or an explicit success criterion alongside the task |
| 2 | Context Files | Lists specific files to read first, each with a description — file paths, [[filename]] links, or backtick references with explanations |
| 3 | Reference | Provides an example of good output AND extracted rules — pasted content or "here is an example" followed by Always/Never patterns |
| 4 | Success Brief | Covers all four: output type/length, recipient reaction, what to avoid, and success definition ("success means…") |
| 5 | Rules | States explicit constraints OR asks Claude to flag violations before proceeding — "never", "always", "stop and tell me if", "my standards" |
| 6 | Conversation | Explicitly defers execution — "DO NOT start", "don't execute yet", "ask me questions first", "clarify before starting" |
| 7 | Plan | Asks Claude to prove understanding before producing output — "list the rules that matter", "which constraints apply", "before you write, list…" |
| 8 | Alignment | Requests an execution plan and alignment — "execution plan", "5 steps maximum", "only begin once we've aligned" |
| 9 | Push Intent | When the task will modify files in a git repository — specifies what to do with the changes: commit, push, open PR, or nothing |
Step 3 — Output
Place this block as the very first output, before any other content:
All required layers present:
---
Prompt quality: All layers present
---
One or more required layers missing:
---
Prompt quality: Missing Layer X (Name), Layer Y (Name)
---
Then respond normally. Never block or delay execution based on this check.
