Evaluate prompt
Skill Snowflake-Labs/devops-snowflake-coco-agents/skills/idd/evaluate-prompt
Audit an existing prompt for IDD alignment. Scores it across four dimensions (Goal, Requirements, Constraints, Output), identifies gaps, and offers to hand off to the rewrite skill. Use when: "evaluate this prompt", "audit my prompt", "is this IDD", "score my prompt".From its SKILL.md
npx -y skills add Snowflake-Labs/devops-snowflake-coco-agents --skill evaluate-promptAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.8 KB, as published. Nobody here has run it
Stopping Points
Ask the user to provide the prompt to evaluate:
ask_user_question:
header: "Prompt"
question: "Paste the prompt you want to evaluate, or provide a file path."
type: text
defaultValue: ""
If a file path is given, read the file. If pasted directly, use as-is.
Optionally ask for context:
ask_user_question:
header: "Context"
question: "What does this prompt do? (helps calibrate the evaluation)"
type: text
defaultValue: "A cortex exec / CI/CD automation prompt"
Evaluation
Score each of the four IDD dimensions from 0–3:
| Score | Meaning |
|---|---|
| 0 | Absent — not present at all |
| 1 | Implicit — hinted at but not stated |
| 2 | Partial — present but incomplete or vague |
| 3 | Full — explicit, specific, and sufficient |
Dimension 1 — Goal (Desired State)
Does the prompt declare what should exist or be true after execution?
- 0: No goal — only "do X" instructions
- 1: Role framing ("You are a reviewer") without desired state
- 2: Vague goal ("find bugs and report them")
- 3: Specific desired state with quality criterion ("surface issues precise enough that a fix agent can act without follow-up")
Dimension 2 — Requirements (Intent Statements)
Are the requirements outcome-focused (what should happen) vs step-by-step (how to do it)?
- 0: No requirements
- 1: Pure procedure ("scan files, create issue, exit")
- 2: Mix of intent and procedure
- 3: Pure intent statements ("one issue per root cause", "body must include file/line/why/expected")
Dimension 3 — Constraints
Does the prompt specify scope limits, safety rules, and what not to do?
- 0: No constraints
- 1: Single vague constraint ("don't add comments")
- 2: Partial constraints (scope defined but safety missing)
- 3: Full constraints (scope + safety + error handling + negative examples)
Dimension 4 — Output (Glass Box)
Does the prompt specify what success looks like including observable evidence?
- 0: No output specification
- 1: Implicit ("create issues") — no format, no summary
- 2: Partial format ("use this title template") but no summary/Glass Box reporting
- 3: Full Glass Box — format + stdout summary + success/failure signals
Presenting Results
Show the scores as a table:
Dimension Score Notes
──────────────────────────────────────────────────────
Goal X/3 <what's there / what's missing>
Requirements X/3 <what's there / what's missing>
Constraints X/3 <what's there / what's missing>
Output X/3 <what's there / what's missing>
──────────────────────────────────────────────────────
IDD Score XX/12 <overall assessment>
ICR estimate ~X <rough ops-per-intent if inferable>
Classification:
- 0–4: Syntaxist prompt — command relay, low agentic readiness
- 5–8: Transitional — some intent, significant gaps
- 9–12: IDD-compliant — high ICR potential, strong agentic readiness
Then ask:
ask_user_question:
header: "Next step"
question: "What would you like to do?"
options:
- label: "Rewrite it using IDD"
description: "Hand off to the rewrite-prompt skill"
- label: "Just show me what to fix"
description: "List the specific gaps with examples"
- label: "Done"