sandbox-python-executor
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill defines a policy for executing Python code via host-native tools for verification and deterministic computation. As specified in
SKILL.md, it includes safety mitigations such as treating scripts as untrusted, performing read-only operations unless otherwise requested, and requiring manual script inspection prior to execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data sources including archives, JSON/YAML manifests, and user-provided files, establishing an attack surface for indirect prompt injection.
- Ingestion points: Parsing of manifests, inspection of archives, and transformation of user files described in
SKILL.md. - Boundary markers: Absent; the instructions do not specify the use of delimiters or clear boundaries for untrusted data passed to the execution tool.
- Capability inventory: Access to host-native Python execution (Code Interpreter) with file system and diagnostic capabilities as documented in
SKILL.mdandagents/openai.yaml. - Sanitization: The skill mandates that the agent treat repository-native scripts as untrusted input and inspect them before execution to identify potential risks.
Audit Metadata