skill package

Sandbox Script Runner

Runs uploaded package scripts in an isolated shell sandbox with approval gates, live output, and artifact collection.

SandboxReviewedvv1.0.0publicAdmin

92% eval4.8 rating1,044 installs6 targets

Edit in Build

Permissions, source, and what this skill claims to do.

Contract

What this skill may access, and where it can install.

read_fileslow risk

Mount uploaded package files and workspace inputs.

write_filesmedium risk

Collect artifacts, reports, dist files, and changed workspace files.

shellhigh risk

Execute the approved command inside an isolated sandbox.

networkmedium risk

Allow outbound network only for approved hostnames.

Compatibility

CodexClaudeAntigravityOpenCodeGrokVS Code

Source

README and SKILL.md for the current version.

README

Sandbox Script Runner turns uploaded scripts into controlled, trace-backed runs. It is designed for real shell execution inside a sandbox, not the application server.

SKILL.md

---
name: sandbox-script-runner
description: Execute uploaded agent-skill scripts inside an isolated shell sandbox. Use this for package tests, build scripts, artifact generation, and command-output explanations.
---

# Sandbox Script Runner

Execute uploaded agent-skill scripts inside an isolated shell sandbox. Use this for package tests, build scripts, artifact generation, and command-output explanations.

Use this skill when the user needs a controlled agent workflow with clear inputs, bounded actions, traceable decisions, and a final artifact that another agent or developer can use immediately.

## Workflow
1. Inspect the package manifest, scripts, and workspace files before suggesting commands.
2. Present detected commands with risk labels and require explicit shell approval before execution.
3. Normalize paths and network allowlists before mounting files or starting a command.
4. Run exactly the approved command in the isolated sandbox with a timeout and output limit.
5. Stream stdout, stderr, permission events, and command exit metadata into the trace.
6. Collect artifacts from approved output paths and explain what changed.

## Permissions
- read_files: Read package files and workspace inputs.
- write_files: Persist sandbox artifacts and changed files.
- shell: Run the approved command in the isolated sandbox.
- network: Use only approved public hostnames when network access is necessary.

## Compatibility
- Codex
- Claude
- Antigravity
- OpenCode
- Grok
- VS Code

## Output Format
- Command summary
- Permission decisions
- stdout and stderr highlights
- Exit status
- Artifacts generated
- Recommended next command

## Examples
- "Run npm test for this uploaded skill package and summarize failures."
- "Execute scripts/build-report.mjs and collect artifacts/report.md."
- "Run this package without network access and show the trace."

Reception

Community feedback for this skill.

No reviews yet.

Loading studio…

--- name: sandbox-script-runner description: Execute uploaded agent-skill scripts inside an isolated shell sandbox. Use this for package tests, build scripts, artifact generation, and command-output explanations. --- # Sandbox Script Runner Execute uploaded agent-skill scripts inside an isolated shell sandbox. Use this for package tests, build scripts, artifact generation, and command-output explanations. Use this skill when the user needs a controlled agent workflow with clear inputs, bounded actions, traceable decisions, and a final artifact that another agent or developer can use immediately. ## Workflow 1. Inspect the package manifest, scripts, and workspace files before suggesting commands. 2. Present detected commands with risk labels and require explicit shell approval before execution. 3. Normalize paths and network allowlists before mounting files or starting a command. 4. Run exactly the approved command in the isolated sandbox with a timeout and output limit. 5. Stream stdout, stderr, permission events, and command exit metadata into the trace. 6. Collect artifacts from approved output paths and explain what changed. ## Permissions - read_files: Read package files and workspace inputs. - write_files: Persist sandbox artifacts and changed files. - shell: Run the approved command in the isolated sandbox. - network: Use only approved public hostnames when network access is necessary. ## Compatibility - Codex - Claude - Antigravity - OpenCode - Grok - VS Code ## Output Format - Command summary - Permission decisions - stdout and stderr highlights - Exit status - Artifacts generated - Recommended next command ## Examples - "Run npm test for this uploaded skill package and summarize failures." - "Execute scripts/build-report.mjs and collect artifacts/report.md." - "Run this package without network access and show the trace."