Loading the registry…
LLM Sandbox is a lightweight and portable sandbox environment designed to run large language model (LLM) generated code in a safe and isolated mode.
The publisher performed an authenticated act of publication — proving control of a namespace in the official MCP registry, or appearing in a first-party catalogue an operator curates. That establishes WHO published it. It is not a security finding, and attested artifacts carry Flagged verdicts too.
160 critical findings reported by cisco-skill-scanner, agent-audit-kit, 0 of them in the installed artifact. The composite score alone does not reflect this — read the per-scanner reports below.
This artifact ships from 2 sources and their verdicts differ. The worst is github (Flagged). Installing from a cleaner source does not mean the others are safe — check the source you actually install from.
github: https://github.com/vndee/llm-sandbox, 248 findings
Choose your runtime: the CLI writes the mcpServers config entry for that target.
nerlo install --target claude-code -- llm-sandbox
Install respects the composite badge: Clean proceeds, Caution prompts for confirmation, and Flagged is refused. Write operations require an API token.
5 of 11 scanners examined this package. The rest do not apply to it — language, ecosystem and packaging decide which analyzers can say anything, and a scanner that cannot examine an artifact reports nothing rather than passing it.
v2.0.11
92 findings · 17.2s
v0.3.26
148 findings · 11.8s
v0.1.0
0 findings · 29.3s
v0.1.0
2 findings · 29.3s
v0.1.0
0 findings · 20.8s
v1.4.0
This scanner found nothing in this artifact that it assesses. What each scanner covers
0 findings · 0.2s
No report in this scan: capslock, govulncheck, osv-scanner, trivy, trivy_image.
1 scan on record. Every scan's full results are retained immutably for 24 months.
| Completed | Composite | Change | Scanners | Status |
|---|---|---|---|---|
| Aug 3, 2026 | Scan Halted | — | agentshield ·n/a: not applicablecisco-skill-scanner: completeagent-audit-kit: completenerlo-behavioral: completenerlo-install-instruction: completenerlo-multi-source: complete | rejected |
Findings in files this artifact installs stand as reported; a model re-read the rest and said which ones it believes are false positives. This is a second opinion published beside the evidence, not a correction to it: the per-scanner reports above are unchanged, every dismissed finding is still listed there at its original severity, and the score and the Flagged badge are computed from those raw severities alone. Nothing below moved them.
Out of 242 findings across every scanner that examined this package. 160 of 160 findings rated critical by their scanner were among the dismissals.
Not reviewed is not the same as clean. Those 5 findings stand exactly as their scanners reported them, at full severity, as they would have without this review.
Two keys had to turn. The model had to name a reason from a fixed list of six, and a deterministic check of ours — over which files this artifact actually installs — had to independently agree. Anything it could not corroborate is in “not reviewed” above, at full severity.
142 Agent instruction file — The match is in a file that instructs an AI coding agent — a CLAUDE.md or similar. It is documentation of intent, and our own check confirmed it is not part of what gets installed.
Rules: AAK-AGENT-001, AAK-AGENT-002, TOOL_ABUSE_UNDECLARED_NETWORK · Files: CLAUDE.md, .github/copilot-instructions.md, /repo/repo/CLAUDE.md · reported by agent-audit-kit, cisco-skill-scanner
48 Documentation or example — The match is in documentation or sample code, confirmed by our own check to be outside the installed artifact.
Rules: COMMAND_INJECTION_SHELL_TRUE, RESOURCE_ABUSE_INFINITE_LOOP, opt.nerlo-rules.nerlo-install-run-bundled-artifact and 9 more · Files: /repo/examples/security_integration_tests.py, /repo/examples/security_policy_examples.py, /repo/examples/security_policy_presets.py and 15 more · reported by cisco-skill-scanner, nerlo-install-instruction, agent-audit-kit
27 Test or fixture file — The match is in a test or fixture. Our own check confirmed the file is not part of what gets installed.
Rules: COMMAND_INJECTION_EVAL, COMMAND_INJECTION_OS_SYSTEM, YARA_command_injection_generic and 2 more · Files: /repo/examples/security_integration_tests.py, /repo/tests/conftest.py, /repo/tests/test_advanced_security_scenarios.py and 3 more · reported by cisco-skill-scanner
6 CI or editor configuration — The match is in build, CI or editor configuration, confirmed by our own check to be outside the installed artifact.
Rules: AAK-GHA-IMMUTABLE-001, MANIFEST_INVALID_NAME · Files: .github/workflows/main.yml, .github/workflows/docker-publish.yml, /repo/.github/ISSUE_TEMPLATE/SKILL.md and 1 more · reported by agent-audit-kit, cisco-skill-scanner
These findings are in files this artifact installs and are not placeholder values. No false-positive basis can be corroborated for such a finding, so this review never dismisses one; they were not sent to the model and stand at their scanner's severity.
4 HIDDEN_DATA_FILE
Files: /repo/.pre-commit-config.yaml, /repo/.sonarcloud.properties, /repo/.bandit and 1 more · reported by cisco-skill-scanner
3 RESOURCE_ABUSE_INFINITE_LOOP
Files: /repo/llm_sandbox/interactive.py, /repo/llm_sandbox/pool/base.py · reported by cisco-skill-scanner
2 MANIFEST_MISSING_LICENSE
Files: /repo/SKILL.md, /repo/paper/SKILL.md · reported by cisco-skill-scanner
2 SOCIAL_ENG_VAGUE_DESCRIPTION
Files: /repo/SKILL.md, /repo/paper/SKILL.md · reported by cisco-skill-scanner
1 AAK-SSRF-TOCTOU-001
Files: pyproject.toml · reported by agent-audit-kit
1 EXCESSIVE_FILE_COUNT
Files:
3 The model called these false positives and our own check did not agree, so the dismissal was refused.
Rules: SECRET_PASSWORD_VAR · Files: /repo/examples/test_security.py, /repo/tests/test_advanced_security_scenarios.py · reported by cisco-skill-scanner
2 The model gave no reason we were willing to publish, and a verdict without one is not usable evidence.
Rules: DATA_EXFIL_HTTP_POST · Files: /repo/tests/test_advanced_security_scenarios.py · reported by cisco-skill-scanner
Potential resource abuse via infinite loops
The artifact contains code that may lead to infinite loops, specifically identified in `/repo/llm_sandbox/interactive.py` and `/repo/llm_sandbox/pool/base.py`. These loops could consume excessive system resources, impacting performance or stability. Further review of these code sections is warranted before deployment.
Concern level assessed by the review: High. This is the review's scale, not the registry badge.
Reviewed by gemini-2.5-flash. The model sees the findings, not your code, and cannot change a score.
1 FILE_MAGIC_MISMATCH
Files: /repo/llm_sandbox/language_handlers/artifact_detection/r_plot_detection.py · reported by cisco-skill-scanner