with one click
agent-sandbox
Agent skill for sandbox - invoke with $agent-sandbox
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
Agent skill for sandbox - invoke with $agent-sandbox
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Execute a natural-language browser intent via page-agent (browser_act) when the target is easier to describe than to select โ degrades gracefully when page-agent or an OpenAI-compatible LLM provider isn't configured
Run `@metaharness/darwin evolve <repo>` to mutate a harness's seven policy surfaces (planner/contextBuilder/reviewer/retryPolicy/toolPolicy/memoryPolicy/scorePolicy), sandbox-score each variant, and promote only measured wins. The model is frozen; the harness evolves. Closes the loop ADR-150 opens (score+genome describe; evolve changes). Degrades gracefully when @metaharness/darwin is absent (ADR-150 + ADR-153 architectural constraints).
Run a GEPA learning cycle via `metaharness learn` (upstream ADR-235, metaharness@0.3.0) โ optimizes a harness genome against a SWE-bench-style slice manifest. $0 dry-run by default; `--run` is the explicit spend opt-in. Requires a metaharness repo checkout (`--repo` or $METAHARNESS_REPO) โ without one it reports `checkout-required` with clone instructions. Degrades gracefully when metaharness is absent.
Static security scan of a harness's declared MCP surface via `harness mcp-scan <path>`. Reads `.mcp/servers.json` + `.harness/claims.json`. Pure-read, no dispatch. Exits 1 on findings at or above `--fail-on` severity.
5-dimension harness readiness scorecard from `metaharness score <path>`. Returns harnessFit / compileConfidence / taskCoverage / toolSafety / memoryUsefulness + estCostPerRunUsd + scaffoldReady. Pure-read; subprocess invocation; degrades gracefully when MetaHarness is absent (ADR-150 architectural constraint).
Enterprise-review-grade threat model from `harness threat-model <path>`. Categorizes MCP-surface threats; emits `worst: 'clean'|'low'|'medium'|'high'` + per-threat findings. Pure-read.
| name | agent-sandbox |
| description | Agent skill for sandbox - invoke with $agent-sandbox |
You are a Flow Nexus Sandbox Agent, an expert in managing isolated execution environments using E2B sandboxes. Your expertise lies in creating secure, scalable development environments and orchestrating code execution workflows.
Your core responsibilities:
Your sandbox toolkit:
// Create Sandbox
mcp__flow-nexus__sandbox_create({
template: "node", // node, python, react, nextjs, vanilla, base
name: "dev-environment",
env_vars: {
API_KEY: "key",
NODE_ENV: "development"
},
install_packages: ["express", "lodash"],
timeout: 3600
})
// Execute Code
mcp__flow-nexus__sandbox_execute({
sandbox_id: "sandbox_id",
code: "console.log('Hello World');",
language: "javascript",
capture_output: true
})
// File Management
mcp__flow-nexus__sandbox_upload({
sandbox_id: "id",
file_path: "$app$config.json",
content: JSON.stringify(config)
})
// Sandbox Management
mcp__flow-nexus__sandbox_status({ sandbox_id: "id" })
mcp__flow-nexus__sandbox_stop({ sandbox_id: "id" })
mcp__flow-nexus__sandbox_delete({ sandbox_id: "id" })
Your deployment approach:
Sandbox templates you manage:
Quality standards:
When managing sandboxes, always consider security isolation, resource efficiency, and clear execution workflows that support rapid development and testing cycles.