AGENT SKILL

sandbox-zip

Read and write zip archives in a Code node or CodeAct action, with fflate running on the host under a 50 MB inflation cap

License 许可
AGPL-3.0
最近更新
2026-08-18
Stars
★ 1
Forks
0

01 前言 Front Matter

name
sandbox-zip
description
Read and write zip archives in a Code node or CodeAct action, with fflate running on the host under a 50 MB inflation cap

02 资源文件 1

03 SKILL.md 正文

Zip archives in the sandbox

Specifier: @nodetool-ai/sandbox-zip. Declare it in the node's packages property and import it at the top of the body.

Why this one runs on the host

fflate is pure JavaScript and the sandbox compiler admits it, so this pack could have shipped as a guest module — and an earlier version did. It does not, because of the cap: a zip bomb is a policy question, and a policy enforced inside the guest is enforced by code the guest can simply not call. The guest's 64 MB heap does not replicate it either — fflate inflates into host memory first. So the library stays on the host, with unzip refusing anything that inflates past 50 MB, and there is no second route around that number.

unzip — extract an archive

import { unzip } from "@nodetool-ai/sandbox-zip";

const files = await unzip(inputs.archive);          // { "a/b.txt": Uint8Array }
const readme = new TextDecoder().decode(files["README.md"]);
return { names: Object.keys(files), readme };

Entry values arrive as real Uint8Arrays. Decode text with TextDecoder.

zip — build one

import { zip } from "@nodetool-ai/sandbox-zip";

const archive = await zip({
  "report.md": inputs.markdown,                     // strings are UTF-8 encoded
  "data.json": JSON.stringify(inputs.rows)
});
await workspace.writeBytes("out.zip", archive);
return { bytes: archive.length };

Entry values are Uint8Array or string.

Gotchas

  • Both exports are async.
  • 10 MB in, 50 MB out. The archive itself is capped at 10 MB; the sum of the entries unzip produces — and the sum of what zip is handed — is capped at 50 MB, checked as the entries are walked so a bomb stops part way.
  • No streaming. The whole archive is materialized. For a payload near the cap, write entries to the workspace as you go instead.

04 同仓库更多 Skills