The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a Python utility, examples/wrds_conn.py, that programmatically reads the ~/.pgpass file. This file is a standard location for storing sensitive database credentials (host, port, database, username, and password). While intended to facilitate database connections for the user, reading sensitive configuration files from the home directory constitutes high-severity data exposure.
[EXTERNAL_DOWNLOADS]: The SKILL.md file instructs the user to install the pixi package manager using an external shell script hosted at https://pixi.sh/install.sh. This introduces a dependency on a remote resource that is not managed within the skill's own environment.
[REMOTE_CODE_EXECUTION]: The skill explicitly recommends a 'curl | bash' execution pattern (curl -fsSL https://pixi.sh/install.sh | bash) in SKILL.md. Piping remote content directly into a shell is a high-risk practice as it allows for the execution of unverified code from an external server.
[DYNAMIC_EXECUTION]: The skill is designed to generate shell scripts (Slurm .sh files) that are then submitted for execution on an HPC cluster using the sbatch command. This pattern of generating and then executing scripts at runtime is a form of dynamic code execution.
[INDIRECT_PROMPT_INJECTION]: The skill creates a broad attack surface for indirect prompt injection by generating executable Slurm scripts based on user-provided instructions or external data queries without explicit sanitization or strict boundary markers.
Ingestion points: User-supplied logic for Slurm job parameters and SQL queries defined in SKILL.md and references/wrds-polars-pipeline.md.
Boundary markers: None; the skill does not use specific delimiters or instructions to prevent the agent from obeying embedded commands in processed data.
Capability inventory: Includes powerful system commands such as sbatch, srun, ssh, and direct database access via connectorx.
Sanitization: No evidence of input validation or escaping for the generated shell scripts.

hpc