datalad
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: Executes shell commands through
datalad runanddatalad containers-runto record computational provenance and automate data processing steps. - [EXTERNAL_DOWNLOADS]: Retrieves dataset structure and content from well-known scientific hosting services, including OpenNeuro, DANDI, and the DataLad superdataset registry. Installs the DataLad software suite and git-annex binaries from the official PyPI registry.
- [DYNAMIC_EXECUTION]: Utilizes runtime template expansion (e.g.,
{inputs},{outputs}) to construct and execute shell commands based on specific dataset states and user-provided arguments. - [INDIRECT_PROMPT_INJECTION]: Ingests and processes data from external repositories which may contain unverified metadata or instructions within the Git history or run records.
- Ingestion points: Dataset cloning from remote URLs (GitHub, OpenNeuro, DANDI) documented in
references/data-access.md. - Boundary markers: No explicit delimiters are used to separate external dataset content from agent instructions.
- Capability inventory: Full shell execution capability via
datalad runanddatalad containers-runas specified inSKILL.md. - Sanitization: The skill does not implement specific sanitization or validation of commit messages or metadata retrieved from remote sources.
Audit Metadata