modal-container-images
Pass
Audited by Gen Agent Trust Hub on Jul 31, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill provides instructions for mounting sensitive local directories like '~/.aws' into remote containers (e.g., in 'references/local-files-mounts.md'), which can lead to the exposure of credentials to the cloud environment.
- [COMMAND_EXECUTION]: The skill documents the 'run_commands' and 'dockerfile_commands' methods, enabling the execution of arbitrary shell instructions within the build environment as seen in 'SKILL.md'.
- [EXTERNAL_DOWNLOADS]: The documentation includes patterns for fetching content from external sources, including git repositories and public package registries like PyPI and apt.
- [REMOTE_CODE_EXECUTION]: The skill describes workflows where remote code is fetched and executed, such as 'git clone' followed by installation or execution within the container, presenting a risk if source URLs are untrusted.
- [PROMPT_INJECTION]: The skill enables an indirect prompt injection surface by processing external definitions to perform privileged actions.
- Ingestion points: Image configuration parameters (packages, commands, files) in 'SKILL.md' and associated reference files.
- Boundary markers: None present.
- Capability inventory: Shell execution (run_commands), file writes (add_local_file), and network access across all scripts.
- Sanitization: None present.
Audit Metadata