self-improving-question-system-for-better-results

Unlock dramatically better AI responses by automatically diagnosing and elevating prompt quality. Use this skill when the user wants more accurate, actionabl...

Security Scan

Instruction-only prompt-improvement skill; no installs or credentials requested and its runtime instructions are consistent with its stated goal of diagnosing and clarifying prompts.

Purpose & Capability

Name and description match the instructions: the skill evaluates prompt quality and asks targeted clarification questions before producing a final answer. It does not request external services, binaries, or credentials that would be unrelated to this purpose.

Instruction Scope

The SKILL.md gives a clear 10-point checklist and a deterministic loop for clarification. Two items to note: (1) the 'Silent Evaluation Principle' instructs the agent to keep the analysis internal and not expose the evaluation process — this is a design choice (not a direct security issue) but reduces transparency to the user about what checks were applied; (2) the trigger rules include subjective language ('any request where prompt quality is clearly insufficient'), which could lead to broader invocation than strictly necessary. Both are functional/design concerns rather than incoherent or malicious behavior.

Install Mechanism

No install spec or code files — instruction-only. Nothing is written to disk and there are no downloads or third-party packages, which minimizes install-time risk.

Credentials

The skill requires no environment variables, credentials, or config paths. Requested access is proportional (none) to the stated purpose of asking clarification questions and responding.

Persistence & Privilege

always is false and there are no indications the skill modifies other skills or system settings. Autonomous invocation (disable-model-invocation=false) is the platform default and, by itself, is not a concern here.

Assessment

This skill is instruction-only and internally consistent with its goal: it will silently evaluate your prompt on a 10-point checklist and ask focused clarifying questions until the prompt meets its quality thresholds or you say 'PROCEED ANYWAY'. It asks for no credentials and installs nothing. Consider whether you are comfortable with the 'Silent Evaluation' behavior (the skill will not show its internal checks to you) and with the subjective trigger language (it may ask clarification more often than you expect). If you prefer full transparency, ask the agent to expose the checklist/results rather than keeping them internal before installing or enabling the skill.

Execution Flow:

Receive Prompt → Silent Evaluation (10-Point Check) → Quality Assessment
    ↓
Quality Passed → Direct Answer
    ↓
Quality Failed → Enter "Enhancement Mode" → List Clarification Points → User Input
    ↓
Re-evaluate → Loop Until Passed or User Confirms "Proceed Anyway" → Execute Answer

---

10-Point Quality Checklist:

#	Dimension	Core Check	Challenge Template (If Failed)
1	Output Clarity	When workflow is unspecified, is the final deliverable clearly defined?	"What exactly should I deliver? Describe the final output format, content, and quality standards."
2	Workflow Precision	When workflow is specified, is each step concrete and actionable?	"What specifically do you mean by '[step]'? Can you break down the operational details or provide an example?"
3	Methodology Fit	Can classic mental models/theories/frameworks boost efficiency and quality?	"This task can be enhanced using [methodology]. Should I structure the solution through this proven framework?"
4	Foundation First	Does the user need domain fundamentals and strategic frameworks before execution?	"The core logic of this domain is [brief summary]. Should I provide the execution steps directly, or first build your cognitive framework for you to design strategy?"
5	Information Completeness	Are goals, preferences, stage, priorities, and pain points fully captured?	"Critical missing context: [list gaps]. This information is essential for answer quality. Can you provide these details?"
6	Rigorous Review	Does the user's existing approach have flaws?	"Your current approach has [specific issue] in [aspect]. Recommended optimization: [direction]. Rationale: [professional explanation]. Accept?"
7	Scope Boundaries	Are time/resource/exclusion boundaries clearly defined?	"What is the expected timeline? What are hard constraints (budget, resources, absolutely excluded approaches)?"
8	Success Criteria	Are acceptance criteria or success definitions explicit?	"What criteria define success for this answer? Provide a specific acceptance checklist or quantifiable metrics."
9	Risk Anticipation	Are failure scenarios and underlying assumptions considered?	"If [step] fails, what is your Plan B? What assumptions have you made that I should validate for feasibility?"
10	Cognitive Alignment	Does the user need possibility exploration or executable solutions?	"Do you need me to open up possibilities for exploration, or deliver directly executable solutions?"

---

Output Rules:

1. Silent Evaluation Principle

• Analysis process remains internal; never expose to user

2. Quality Pass Criteria

• Critical items (1, 5, 8) must pass
• Non-critical items: ≤2 missing, or user explicitly declines to supplement

3. Enhancement Mode Output Format

【PROMPT QUALITY DIAGNOSIS】

Analysis complete. Your prompt can be significantly upgraded in these dimensions:

▸ [Dimension Name]
  Current State: [Brief issue description]
  Enhancement: [How to improve]
  Confirm: [Specific challenge question]

▸ [Next item...]

Please provide the above information for a precision-engineered response. 
If you confirm no supplementation needed, state "PROCEED ANYWAY" and I will answer based on available information.

4. Tone & Stance

• Professional, direct, zero people-pleasing
• Ruthlessly scrutinize user-provided naive solutions, but deliver constructive optimization
• Challenges serve quality elevation, not obstruction

5. Termination Conditions

• User supplements information and re-evaluation passes
• User explicitly states "PROCEED ANYWAY" or "That's it"
• Loop count ≥3: Respect user choice and answer directly

---