The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/inspect_latch_sdk.py uses importlib.import_module to verify the availability of SDK symbols. A manual review of the script confirms that the modules targeted for inspection are hardcoded within the script and do not accept untrusted external input, ensuring the environment check is safe.
[COMMAND_EXECUTION]: The skill provides instructions for using the latch CLI and Python subprocess.run to manage bioinformatics pipelines. It proactively includes security guidance recommending the use of argument lists and check=True to prevent command injection risks during execution.
[CREDENTIALS_UNSAFE]: The documentation explicitly warns against manual handling of sensitive authentication data, such as the ~/.latch/token file. It directs users to use the official OAuth-based login flow and highlights the importance of not logging secrets or signed URLs.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the official Latch SDK via PyPI and references documentation hosted on official domains and GitHub repositories. These are verified as trusted sources related to the skill's intended purpose.
[INDIRECT_PROMPT_INJECTION]: While the skill interacts with external Latch APIs and Registry data, it incorporates boundary markers and validation steps. It instructs agents to fetch schemas, validate parameters against those schemas, and obtain user confirmation for cost-intensive operations, significantly reducing the surface for indirect injection attacks.

latchbio-integration