ia-meta-prompting

Stress-test agent outputs with structured reasoning patterns and verification checkpoints.

30|3|Updated Feb 22, 2026

Search Tags:#edge cases #architecture review #structured output #confidence scoring #meta-prompting #adversarial reasoning #decision validation

Ready to run

npx skills add https://github.com/iliaal/whetstone --skill ia-meta-prompting

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ia-meta-prompting
Source: https://github.com/iliaal/whetstone/tree/main/plugins/whetstone/skills/ia-meta-prompting
Command: npx skills add https://github.com/iliaal/whetstone --skill ia-meta-prompting

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill reduces costly mistakes by forcing careful reasoning, explicit assumptions, and verification before an agent presents a final decision or answer.

Core Features & Use Cases

Reasoning patterns for validation: Apply modifiers like think, verify-think, adversarial, edge cases, and confidence tiers to expose weak spots.
Assumption and failure-mode surfacing: Enumerate implicit assumptions and run a premortem to identify systemic risks.
Structured output options: Constrain responses and format outputs (including JSON) for downstream reliability.
Real-world use case: Validate an architecture proposal by challenging it from opposing perspectives, checking for edge-case failures, and producing a verified recommendation instead of an untested plan.

Quick Start

Ask your agent to run /verify-think with /edge on the design and return a VERIFIED ANSWER with the main reasoning left prominent.

Frequently Asked Questions about ia-meta-prompting

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I stress-test an architecture review before committing to a design?▼

You can validate an ambiguous architecture plan by challenging it from opposing perspectives and checking for edge-case failures. Applying structured reasoning modifiers exposes weak spots and implicit assumptions, ultimately producing a verified recommendation instead of an untested plan.

What is adversarial reasoning for decision validation?▼

Adversarial reasoning challenges a proposed plan from opposing perspectives to expose weak spots and systemic risks. It forces explicit assumption enumeration and runs a premortem to identify failure modes, ensuring claims are grounded before finalizing high-stakes outputs.

How do I format agent outputs as JSON with confidence scoring?▼

You format agent outputs as JSON with confidence scoring by applying structured output modifiers during the reasoning process. This constrains responses into pattern-aware structured results, utilizing VERIFIED or REVISED markers and confidence tiers to ensure downstream parsing reliability.

Do I need any external dependencies to run a premortem on security-sensitive plans?▼

No external dependencies are required to run a premortem on security-sensitive plans. The process relies entirely on applying internal reasoning patterns and verification checkpoints to enumerate assumptions, identify failure modes, and validate claims within your existing environment.

When should I use meta-prompting for edge case analysis?▼

Use meta-prompting for edge case analysis when dealing with high-stakes scenarios where mistakes are expensive. It forces careful reasoning and verification checkpoints before presenting a final decision, making it ideal for security-sensitive validation and ambiguous architecture proposals.

Can I constrain text outputs to specific confidence tiers during architecture reviews?▼

Yes, you can constrain text outputs to specific confidence tiers during architecture reviews by applying structured output modifiers. This produces pattern-aware structured results with VERIFIED or REVISED markers, ensuring claims are grounded and formatted for downstream parsing reliability.