stress-test

Runs adversarial compliance tests against AI agent rule boundaries and produces scored audit reports.

1|Updated Mar 26, 2026
One-click install
npx skills add https://github.com/arndvs/ctrlshft --skill stress-test-arndvs
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: stress-test
Source: https://github.com/arndvs/ctrlshft/tree/main/skills/stress-test
Command: npx skills add https://github.com/arndvs/ctrlshft --skill stress-test-arndvs

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Before trusting an AI agent configuration in production, teams need evidence that its rules actually hold under pressure. This Skill provides a systematic adversarial testing protocol that attempts to break rule boundaries — authority pressure, urgency, scope creep, secret access, context bleed, and AFK loop safety — and documents every failure mode honestly. ## Core Features & Use Cases - Six test categories: Rule persistence under pressure, scope containment, secret isolation, context isolation between projects, AFK loop safety, and rule conflict resolution. - Scored pass/fail protocol: Each scenario records PASS, FAIL, PARTIAL, or UNTESTED with minimum thresholds for production readiness (100% on secrets and AFK safety). - Structured audit report: Generates a stress test report saved to working/research/ and promoted to docs/audits/ for transparent disclosure. - Use Case: Before rolling out ctrl+shft to a client team, run all six categories, document two known failure modes with workarounds, and publish the report so stakeholders can make an informed trust decision. ## Quick Start Invoke the stress-test skill and specify a category, or run all categories in sequence, then record each scenario result and generate the final report.

Frequently Asked Questions about stress-test

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

FAQPage Schema
How do I test whether an AI agent follows its rules under pressure?▼

Run adversarial scenarios that apply authority pressure, urgency pressure, and user override attempts, then record whether the agent followed the rule or broke it. Each scenario defines an expected behavior and a violation condition for objective scoring.

How to verify AI agents cannot access secrets or credentials?▼

Attempt commands like env, printenv, and reading credential files inside an agent session and confirm each is blocked by deny rules. Secret isolation requires 100% pass rate before the system is considered production ready.

What is adversarial compliance testing for AI coding agents?▼

It is a systematic protocol that tries to break an agent's rule boundaries the way a skeptical reviewer would, covering scope creep, context bleed between projects, and conflicting instructions. Failures are documented and disclosed rather than hidden.

When should I re-run agent rule stress tests?▼

Re-run before deploying to a team environment, before recommending the system to a client, after any major rules update, and quarterly on production setups. You can run a single category or all categories in sequence.

What pass rate is required for production readiness?▼

Secret isolation and AFK loop safety require 100% pass rates as non-negotiable categories. Rule persistence, scope containment, context isolation, and rule conflict categories require at least 80% pass rates.