The Agent Skills Directory

[DYNAMIC_EXECUTION]: The included script scripts/inspect_latch_sdk.py uses importlib.import_module to dynamically inspect the locally installed Latch SDK. This is a diagnostic feature used to confirm symbol availability and signatures for compatibility checking, and it is restricted to a hardcoded set of SDK modules.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing external bioinformatics files (e.g., FASTQ, CSV) and records from the Latch Registry. This risk is mitigated by the skill's instructions to use strict type definitions (LatchFile, LatchDir), perform input validation, and require explicit user confirmation before executing paid compute or destructive operations.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install the latch SDK from the official Python Package Index (PyPI) using uv. These are standard dependencies for the intended platform and are considered safe.
[COMMAND_EXECUTION]: The skill involves the execution of scientific tools via shell commands. It emphasizes secure practices, such as using argument lists instead of raw shell strings and employing subprocess.run(..., check=True) to prevent command injection and handle errors reliably.

latchbio-integration