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 ariana-dot-dev/cloud-sandbox-vm-skills --skill run-untrusted-codegit clone --depth 1 https://github.com/ariana-dot-dev/cloud-sandbox-vm-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/ariana-dot-dev/cloud-sandbox-vm-skills/run-untrusted-code)<a href="https://agentmods.dev/skills/ariana-dot-dev/cloud-sandbox-vm-skills/run-untrusted-code"><img src="https://agentmods.dev/badge/skills/ariana-dot-dev/cloud-sandbox-vm-skills/run-untrusted-code/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/ariana-dot-dev/cloud-sandbox-vm-skills/run-untrusted-code"><img src="https://agentmods.dev/badge/skills/ariana-dot-dev/cloud-sandbox-vm-skills/run-untrusted-code.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 | $0.00061 | $0.00375 |
| Opus 5 | $0.00030 | $0.00187 |
| Sonnet 5 | $0.00012 | $0.00075 |
| Haiku 4.5 | $0.00006 | $0.00038 |
Grade C, and why
run-untrusted-code scanned grade C with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
`curl -fsSL https://box.ascii.dev/install | sh`, then `box login "$BOX_API_KEY" --json`. Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
`curl -fsSL https://box.ascii.dev/install | sh`, then `box login "$BOX_API_KEY" --json`. What it actually says
Run untrusted code (Box)
Use this right before executing code you generated or fetched when running it locally would be risky: shell scripts, scraped snippets, dependency installs, anything with side effects. Pattern: create → run → delete, so nothing touches the host.
Why Box
- Isolation — a separate VM; the host filesystem and network are never exposed.
- Disposable & cheap —
box deletewhen done; per-second billing, a run costs cents. - Real machine — generated code that needs
pip install, subprocesses, or Docker actually works.
Prefer another tool for: thousands of tiny <500 ms runs (E2B/Modal microVMs).
Prereqs
curl -fsSL https://box.ascii.dev/install | sh, then box login "$BOX_API_KEY" --json.
Recipe
box_id="$(box new --json | jq -r 'select(.event == "ready") | .id')"
trap 'box delete "$box_id" --json' EXIT # never leak a VM
box ssh "$box_id" -- "mkdir -p /run && cat > /run/main.sh" < ./main.sh
box ssh "$box_id" -- "cd /run && timeout 120 bash main.sh"
Always wrap in timeout and delete in a trap/finally.
Limits
EU-only; disk-level snapshots; 100 active VMs soft cap.
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.
- 9d ago First seen · 36 lines · 61 tokens per session scan C d06e6ce5e5d2
run-untrusted-code is a skill published in the GitHub repository ariana-dot-dev/cloud-sandbox-vm-skills (2 stars, last pushed 2mo ago), licensed MIT. It adds 61 tokens to every session and 375 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
superdesign
Design or redesign frontend UI, presentations, and graphics on the Superdesign canvas with a choice of leading AI models. Use whenever the user wants to design a page, feature, flow, slide deck, or brand-new product; improve or reproduce existing UI; compare design results across top models; explore visual variants…
declauding
Removes LLM prose tics from drafts — staged reveals, "it's not X, it's Y", significance tags, abstraction agency, coy headers, fragment cadence, the flatter slop patterns (copula avoidance, participle tails, forced triads, chatbot residue), the flat-certainty patterns the pass itself produces (bare adverbs, juridical…
image-to-svg
Convert raster images (photos, paintings, illustrations, line art) into SVG vector reproductions. Use when the user uploads an image and asks to reproduce, vectorize, trace, or convert it to SVG. Also use when asked to decompose an image into shapes, create an SVG version of a picture, or faithfully reproduce artwork…
declaring-invariants
Find tests that enumerate a domain by copying it, and declare the invariants a codebase depends on. Reports where a parametrize list, for-loop, or it.each iterates a hand-written subset of a dict/set/tuple/Enum that exists in the source, and names the members nothing covers. Use when reviewing tests, when a module…
developing-preact
Specialized Preact development skill for standards-based web applications with native-first architecture and minimal dependency footprint. Use when building Preact projects, particularly those involving data visualization, interactive applications, single-page apps with HTM syntax, Web Components integration, CSV/JSON…
generative-thinking
Break out of a locked problem frame by picking one disciplined move — reframe, provocation (Po), random stimulus, SCAMPER, inversion, perspective shift, constraint play, or family traversal — and committing to it before evaluating. Use when stuck, when options feel narrow or obvious, when iterations produce variations…