What it claims to do
Build sandboxed applications for secure code execution. Load when building AI code execution, code interpreters, CI/CD systems, interactive dev environments, or executing untrusted code. Covers Sandbox SDK lifecycle, commands, files, code i…
What we found · 5
- Dynamic code execution / spawns a processhighCalls eval/exec/subprocess — unpredictable behavior; this is the part that most needs sandbox isolation.
- Destructive file operationhighContains rm -rf or a similar delete command; misuse or abuse causes data loss.
- Description doesn’t say WHEN to use itmediumA good description names the trigger (e.g. "when reviewing a PR"). Without it the skill mis-fires or never fires.
- Bloated: high token costmedium~1,443 tokens on every single call. Compress instructions and trim examples.
- Embedded shell script blockmediumShips runnable command-line scripts. Not necessarily bad, but must be sandboxed and reviewed.
What its scripts can do
| high | Executes code |
| medium | Installs packages |
Conflicts with · 8
| cloudflare-email-service | shadow | build, sdk |
| docx | shadow | fil |
| shadow | fil | |
| brainstorming | shadow | build |
| test-driven-development | shadow | code |
| verification-before-completion | shadow | command |
| agents-sdk | overlap | build, sdk, load, application, cover |
| algorithmic-art | overlap | interactive, code, system |
If you have both installed, they compete for the same triggers — the top cause of an agent picking the wrong skill.
One command grades every skill in your Claude Code / Codex — safety, token bloat, and conflicts, 100% locally.
npx skillmoo scan
Grade computed 2026-07-13 · rubric static-v1. Efficacy shown only when measured on a labeled model + task suite — never invented.
Add this independent grade to the skill's README — the badge links back to this rating and updates when we re-grade.
[](https://skillmoo.com/skill/sandbox-sdk/)