benchling-integration
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from the Benchling platform, which introduces a potential attack surface for indirect prompt injection if the source data contains malicious instructions.\n
- Ingestion points: Data is retrieved from Benchling using methods like
benchling.dna_sequences.list(),benchling.containers.list(), and other resource list/get operations described inSKILL.mdandreferences/sdk_reference.md.\n - Boundary markers: The instructions and examples do not include specific delimiters or explicit 'ignore instructions' prompts to wrap the external biological data or experimental metadata.\n
- Capability inventory: The skill demonstrates capabilities to write data to the local file system (e.g., the CSV export example in
SKILL.md) and modify data on the Benchling platform through various update and workflow methods.\n - Sanitization: The provided code snippets demonstrate direct processing and exporting of fetched data without explicit sanitization or validation of the content against instruction injection patterns.
Audit Metadata