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 hunting-code-interpreter-and-tool-sandbox-escapegit 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/hunting-code-interpreter-and-tool-sandbox-escape)<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/hunting-code-interpreter-and-tool-sandbox-escape"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/hunting-code-interpreter-and-tool-sandbox-escape/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/hunting-code-interpreter-and-tool-sandbox-escape"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/hunting-code-interpreter-and-tool-sandbox-escape.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 | $0.00207 | $0.02092 |
| Opus 5 | $0.00103 | $0.01046 |
| Sonnet 5 | $0.00041 | $0.00418 |
| Haiku 4.5 | $0.00021 | $0.00209 |
Grade A, and why
hunting-code-interpreter-and-tool-sandbox-escape 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 — 137 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hunting code-interpreter and tool sandbox escape: the model writes code and something runs it
An AI application that runs model-generated code, a code interpreter, an agent tool that shells out, a notebook backend, is executing attacker-influenceable instructions, because the model's output is shaped by its input and its input can be attacker-controlled. The sandbox that runs that code is therefore a security boundary around hostile code, and it is often built as if the code were trusted. The escapes are concrete. The runtime may have network access, so generated code reaches internal services or exfiltrates. It may see a writable host filesystem or a shared mount, so code reads or writes outside the jail. It may carry credentials, an API key, a cloud role, a token, in its environment, so code steals them. The isolation may be thin, a shared kernel, a container with a host mount, a subprocess with no namespace, so a known primitive escapes it. And with no CPU, memory, time, or output bound, one run starves the host. The hunt is to run code in the sandbox and see what of the host it can touch. You hunt this by making the model emit probing code and observing what succeeds.
When to use
- An AI feature executes model-generated code or shell, or runs agent tools, inside a sandbox.
- The sandbox may have network access, a writable or shared filesystem, or credentials in its environment.
- Isolation may be thin (shared kernel or mount) or resource limits may be missing.
Scope check
Test sandbox escape only on AI applications and runtimes you own or are authorized to assess, in a non-production sandbox. Running probing code exercises real execution and can reach real resources, so use an isolated test deployment and never touch data, credentials, or hosts that are not yours. If you can't name the authorization, stop.
The loop
- Establish the intended sandbox boundary first. Name what the runtime is allowed to touch: no network or a narrow allowlist, an ephemeral filesystem with nothing host-shared, no ambient credentials, hard CPU/memory/ time/output limits, and isolation strong enough for hostile code. This is the false-positive killer: a runtime with no egress, no host mount, no credentials in scope, enforced resource bounds, and real isolation is behaving correctly. Name the intended boundary, then test each edge.
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 · 137 lines · 207 tokens per session scan A 1db21009e010
hunting-code-interpreter-and-tool-sandbox-escape is a skill published in the GitHub repository UnboundCompute/security-agent-skills (5 stars, last pushed 3d ago), licensed MIT. It adds 207 tokens to every session and 2,092 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.