The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents the 'state_load' function, which is used to restore DataFrame transformations and models. The documentation acknowledges that these state files can contain serialized Python objects that execute upon loading. While the instructions explicitly warn the user to only load state files from trusted sources, the mechanism itself involves the deserialization of local data that can trigger code execution.
Evidence: 'references/io_operations.md' and 'references/machine_learning.md' note that state files can execute deserialized Python objects and emphasize loading only trusted state.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large datasets from various file formats and remote cloud storage services. This creates an attack surface where malicious instructions embedded in processed data could potentially influence the agent's behavior if not properly isolated.
Ingestion points: 'vaex.open()', 'vaex.from_csv()', and 'vaex.open_many()' functions described in 'SKILL.md' and 'references/io_operations.md'.
Boundary markers: The documentation does not specify explicit prompt boundary markers for data ingestion, though it recommends schema validation.
Capability inventory: The skill utilizes 'Bash', 'Write', 'Read', and 'Edit' tools, allowing for file system operations and network access to S3 or GCS.
Sanitization: The documentation focuses on numerical validation and schema consistency rather than content-based sanitization for language model safety.
[EXTERNAL_DOWNLOADS]: The skill contains instructions for installing necessary software and fetching metadata from external services.
Evidence: 'SKILL.md' provides commands to install 'vaex-core', 'vaex-hdf5', and other integration packages using 'uv pip install'.
Evidence: 'SKILL.md' instructs the agent to fetch citation metadata from 'arxiv.org', a well-known service, which is documented neutrally.
[COMMAND_EXECUTION]: The skill uses shell commands for environment setup, package management, and running local services.
Evidence: 'SKILL.md' and 'references/io_operations.md' include 'uv' command usage and 'vaex server' startup instructions.

vaex