What types of failures can be simulated?

The skill currently supports three scenarios: ambiguous-spec (requirements), failing-tests (test recovery), and security-finding (vulnerability response).

Where can I find the results of a simulation?

A comprehensive markdown report is automatically generated in 'docs/training/simulation-{date}.md' containing scores, justifications, and recommendations.

Does this skill modify my actual source code?

No, all outputs are written to a dedicated 'docs/training/' directory and are clearly labeled as 'SIMULATION MODE' to prevent confusion with real workflow outputs.

How does the scoring system work?

The skill evaluates agent behavior against specific items in their 'Non-Normal' checklists, assigning a PASS or FAIL based on whether the agent stopped, documented the issue, and followed escalation paths.

Which agents can I test with this skill?

It is designed to test the Agent Triforce ecosystem, specifically Prometeo (PM), Forja (Developer), and Centinela (QA).

Agent Failure Simulation & Testing

Name: Agent Failure Simulation & Testing
Author: ArtemioPadilla

byArtemioPadilla

•

セキュリティとテスト

Verifies AI agent reliability by running controlled failure simulations and scoring adherence to emergency procedures.

The Simulate Failure skill is a diagnostic tool for the Agent Triforce framework, designed to stress-test AI agents under non-ideal conditions. By generating synthetic failure scenarios—such as ambiguous requirements, broken tests, or security vulnerabilities—it evaluates whether agents (Prometeo, Forja, and Centinela) correctly pivot to their 'Non-Normal' emergency checklists rather than proceeding blindly. All activities are strictly isolated in a simulation mode, providing detailed audit reports and PASS/FAIL metrics to ensure your AI workforce remains compliant with safety, security, and quality protocols.

主な機能

01Ensures safety with an isolated simulation environment that never modifies production files

02Generates deliberately ambiguous specifications to test PM requirement logic

03Simulates security vulnerabilities like SQL injection to audit QA responses

043 GitHub stars

05Provides detailed PASS/FAIL scoring against agent-specific Non-Normal checklists

06Creates synthetic test failure artifacts to verify developer recovery protocols

ユースケース

01Benchmarking the reliability of AI agent configurations before production deployment

02Auditing the decision-making process of AI agents when faced with contradictory project requirements

03Training AI agents to recognize and report security risks instead of attempting unsafe fixes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add artemiopadilla/agent-triforce simulate-failure

For use in Claude.ai and ChatGPT

Download Skill