evidence-discipline

Enforce evidence-based claim validation in AI agent workflows.

51|17|Updated Feb 10, 2026
One-click install
npx skills add https://github.com/shynlee04/hivemind-plugin --skill evidence-discipline
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: evidence-discipline
Source: https://github.com/shynlee04/hivemind-plugin/tree/main/skills/evidence-discipline
Command: npx skills add https://github.com/shynlee04/hivemind-plugin --skill evidence-discipline

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

This Skill ensures that AI agents only make claims or declare tasks complete after providing verifiable evidence, preventing errors and maintaining integrity in AI-driven workflows.

Core Features & Use Cases

  • Mandatory Verification: Enforces that all assertions (e.g., code compilation, test success, subagent results) are backed by command output or explicit validation steps.
  • Evidence Chain Tracking: Promotes a clear audit trail of how conclusions were reached, linking actions to outcomes.
  • Use Case: When an AI agent claims a feature is "done," this Skill will prompt it to first run npm test and npx tsc --noEmit, and only then allow the "done" claim if all checks pass.

Quick Start

Use the evidence-discipline skill to verify that all tests pass before claiming the current task is complete.

Frequently Asked Questions about evidence-discipline

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

FAQPage Schema
How do I enforce evidence-based verification before an AI agent claims a task is complete?

To enforce mandatory verification, configure your AI agent workflow to run predefined verification commands like npm test and npx tsc --noEmit before allowing any task completion claims. The evidence-discipline mechanism intercepts agent assertions and demands verifiable command output as proof, preventing unsubstantiated claims and ensuring rigorous validation within automated workflows.

What is evidence-based claim validation in AI agent workflows?

Evidence-based claim validation is a mechanism that prevents AI agents from declaring tasks complete without verifiable proof. It requires executing predefined verification commands and inspecting their outputs to back assertions like code compilation or test success, maintaining integrity by linking actions to verifiable outcomes.

How do I track the evidence chain for AI agent decisions and conclusions?

You track the evidence chain for AI agent decisions by enforcing mandatory verification steps that link actions to outcomes. This process creates a clear audit trail of how conclusions were reached, requiring execution of validation commands and inspecting their outputs before accepting subagent results or task completion claims.

Can I use verification commands to audit subagent results in automated workflows?

Yes, you can use verification commands to audit subagent results in automated workflows. This approach applies to scenarios requiring rigorous proof before accepting subagent results, enforcing the execution of predefined validation commands and explicit inspection of their outputs to ensure integrity and prevent errors.

Does evidence-discipline work with standard code quality and testing tools?

Evidence-discipline works with standard code quality and testing tools by enforcing the execution of predefined verification commands and inspecting their outputs. It prompts the AI agent to run validation commands like npm test and npx tsc --noEmit, only allowing claims to pass if all required checks succeed.

What are the limitations of enforcing mandatory verification in AI agent scripts?

The limitation of enforcing mandatory verification in AI agent scripts is that it strictly requires predefined verification commands to exist and pass before any task completion claim is accepted. It cannot validate claims without explicit command execution and output inspection, making it unsuitable for tasks lacking verifiable testing steps.