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 auditing-ml-inference-endpoint-abusegit 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/auditing-ml-inference-endpoint-abuse)<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/auditing-ml-inference-endpoint-abuse"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/auditing-ml-inference-endpoint-abuse/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/auditing-ml-inference-endpoint-abuse"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/auditing-ml-inference-endpoint-abuse.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 | $0.00201 | $0.02038 |
| Opus 5 | $0.00101 | $0.01019 |
| Sonnet 5 | $0.00040 | $0.00408 |
| Haiku 4.5 | $0.00020 | $0.00204 |
Grade A, and why
auditing-ml-inference-endpoint-abuse 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 6d 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 — 139 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Auditing ML inference endpoint abuse: a served model is an asset and a meter, both attackable
A model exposed as an inference endpoint is two things worth attacking at once: a metered resource that costs money per call, and a confidential asset that queries can reconstruct. The abuses follow from that. If the endpoint is unauthenticated or weakly keyed, anyone can call it. If there is no per-caller rate or spend limit, a caller runs up unbounded inference cost, a denial-of-wallet against expensive model serving. Beyond cost, the model itself leaks to a determined querent: systematic queries reconstruct the model or its decision boundary (model extraction), and carefully chosen queries recover whether a specific record was in the training set or reconstruct sensitive training data (membership and inversion inference). Over-informative responses, full class probabilities, raw embeddings, confidence vectors, make both extraction and inversion far easier. The audit treats the endpoint as a cost meter and a confidential asset and checks the controls on both. You audit this by calling the endpoint as an attacker would: unauthenticated, at volume, and systematically.
When to use
- A model is deployed as a callable prediction or embedding endpoint (classifier, recommender, scorer).
- The endpoint may be unauthenticated, weakly keyed, or lack per-caller rate and spend limits.
- Responses may return full probabilities or embeddings, and the model or its training data is confidential.
Scope check
Test inference endpoints only on models and services you own or are authorized to assess, on non-production deployments. Volume and systematic querying incur real cost and exercise a real confidentiality boundary, so use a test deployment and never run up spend or extract from a model that is not yours. If you can't name the authorization, stop.
The loop
- Establish the intended access, cost, and confidentiality bounds first. Name who may call the endpoint, the per-caller rate and spend limit, and what the response should reveal. This is the false-positive killer: an endpoint that authenticates callers, enforces per-caller rate and spend limits, and returns only the minimal answer (a label, a top-k, a coarse score) is behaving correctly. Name the intended bounds, then test each.
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.
- 6d ago First seen · 139 lines · 201 tokens per session scan A 1ceed4a31f97
auditing-ml-inference-endpoint-abuse is a skill published in the GitHub repository UnboundCompute/security-agent-skills (5 stars, last pushed 2d ago), licensed MIT. It adds 201 tokens to every session and 2,038 once invoked, about $0.0010 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-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…
linear-claude-skill
Manage Linear issues, projects, and teams.