Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add UnboundCompute/security-agent-skills --skill evaluating-model-guardrailsgit clone --depth 1 https://github.com/UnboundCompute/security-agent-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/unboundcompute/security-agent-skills/evaluating-model-guardrails)<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/evaluating-model-guardrails"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/evaluating-model-guardrails/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/evaluating-model-guardrails"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/evaluating-model-guardrails.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00119 | $0.01558 |
| Opus 5 | $0.00060 | $0.00779 |
| Sonnet 5 | $0.00024 | $0.00312 |
| Haiku 4.5 | $0.00012 | $0.00156 |
Grade A, and why
evaluating-model-guardrails scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Evaluating model guardrails: measure the boundary, don't collect one jailbreak
A guardrail is a claim: "the model will refuse this class of request." A single successful jailbreak is an anecdote; what you need is a measurement, the rate at which the boundary holds across a structured battery of attacks. Evaluating guardrails means grounding probes in the actual policy, attacking each along known bypass axes, and scoring outcomes reproducibly, so you can state how strong the boundary is, not just that someone once beat it.
When to use
- You are reviewing a model deployment, a system prompt, or a content-safety layer.
- Before shipping an assistant that must refuse or constrain a defined set of requests.
- You need a defensible bypass rate, not a single proof-of-concept transcript.
Scope check
Evaluate models and deployments you own or are authorized to test. Use benign, clearly-scoped probes against a defined policy; do not generate real harmful output against systems you do not control. If you can't name the authorization, stop.
The loop
-
Ground the probes in the stated policy. Get the actual list of what this deployment must refuse or constrain (its safety policy, its system prompt's rules, its allowed scope). Every probe targets a specific rule, so a result maps to a policy line, not a vibe. An undefined policy is the first finding: you cannot evaluate a boundary no one has drawn.
-
Build a baseline probe set. For each rule, write direct requests that should be refused and benign near-misses that should be allowed. The near-misses matter: a guardrail that refuses everything is broken differently from one that refuses nothing. Record baseline refuse/allow behavior before attacking.
-
Attack along the bypass axes. Take each refused probe and apply the known transformations: multi-turn escalation (warm up, then pivot), encoding and obfuscation (alternate scripts, spacing, invisible characters, indirection), role and context reframing (fiction, hypothetical, translation, "for research"), and instruction-hierarchy conflict (content claiming higher authority than the system rule). Each axis is a separate test of the same rule.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 129 lines · 119 tokens per session scan A 3900aa4187fb
evaluating-model-guardrails is a skill published in the GitHub repository UnboundCompute/security-agent-skills (5 stars, last pushed yesterday), licensed MIT. It adds 119 tokens to every session and 1,558 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-05.
Other skills, from other repositories
model-supply-chain
Reviews AI/ML model supply chains for security risks including model provenance verification, training data lineage, fine-tuning pipeline integrity, inference dependency review, and backdoor detection. Auto-invoked when reviewing systems that download pre-trained models, fine-tune foundation models, or deploy models…
prodcheck-review
Review this codebase against the prodcheck pre-production checklists — security, performance, scale, integrations and post-launch readiness. Use when asked to check whether a project is ready to ship, to audit an area before launch, or to work through a specific checklist. Produces evidence with file:line citations…
rag-poisoning-and-data-exfiltration
Test Retrieval-Augmented Generation (RAG) systems for data poisoning, prompt injection via retrieved documents, and data exfiltration through manipulated context windows. Use this skill when assessing RAG-based chatbots, knowledge bases, enterprise AI assistants, or any system that augments LLM responses with external…
integrate-arcjet-guard-genkit
Integrate Arcjet security into a Genkit JS agent using @arcjet/guard — wrap ai.defineTool, put guardMiddleware on generate({ use }) for unwrapped / MCP / filesystem tools, and read a caller-owned id from generate({ context }). Use when asked to add Arcjet to genkit, rate limit its tools, screen inbound messages, or…
integrate-arcjet-guard-tanstack-ai
Integrate Arcjet security into a TanStack AI chat() app using @arcjet/guard — put guardMiddleware first on chat({ middleware }) so onBeforeToolCall gates tools, and read a caller-owned id from chat({ context }). Use when asked to add Arcjet to TanStack AI, rate limit its tools, screen inbound messages, or block prompt…
integrate-arcjet-guard-claude-managed-agents
Integrate Arcjet security into Claude Managed Agents (hosted REST+SSE, beta managed-agents-2026-04-01) using @arcjet/guard — screen user.message / initialevents before sessions.events.send, and gate custom tools on agent.customtooluse. Use when asked to add Arcjet to Claude Managed Agents, rate limit custom tools, or…