The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions utilize the official benchling-sdk Python package, which is the standard library for interacting with the Benchling platform. It also references common ecosystem packages such as biopython, python-dotenv, and httpx for standard development tasks.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and process data from the Benchling API (such as entity names, notebook entry content, and custom metadata fields), it possesses a surface for indirect prompt injection. If the source data in the Benchling tenant were to contain malicious instructions, they could potentially influence the agent's behavior when processing that data.
Ingestion points: Data retrieved via various SDK methods like benchling.dna_sequences.list(), benchling.entries.get(), and benchling.custom_entities.list() as described in SKILL.md and references/sdk_reference.md.
Boundary markers: The provided examples do not explicitly define delimiters for external data.
Capability inventory: The skill includes capabilities for writing to the local file system (CSV export in SKILL.md), reading local files (Bio.SeqIO.parse), and performing network operations via the Benchling API.
Sanitization: The provided code snippets do not include explicit sanitization of field values before processing, relying on the agent's default handling of API responses.

benchling-integration